Productivity, Inequality, and the Digital Economy: A Transatlantic Perspective

Productivity, Inequality, and the Digital Economy This page intentionally left blank Productivity, Inequality, and ...

Author: Nathalie Greenan | Yannick L'Horty | Jacques Mairesse

38 downloads 766 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Productivity, Inequality, and the Digital Economy

This page intentionally left blank

Productivity, Inequality, and the Digital Economy A Transatlantic Perspective

edited by Nathalie Greenan, Yannick L’Horty, and Jacques Mairesse

The MIT Press Cambridge, Massachusetts London, England

( 2002 Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. This book was set in Palatino on 3B2 by Asco Typesetters, Hong Kong, and was printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Productivity, inequality, and the digital economy : a transatlantic perspective / edited by Nathalie Greenan, Yannick L’Horty, and Jacques Mairesse. p. cm. Revised contributions and discussions from a two-day international conference on ‘‘Information and Communications Technologies, Employment, and Earnings,’’ organized by the editors, held in Nice in June 1998, and sponsored by the Conseil de l’Emploi, des Revenus et de la Cohe´sion Sociale (CERC), and by the 10th CRESTNBER Franco-American economic seminar. Includes bibliographical references and index. 1. Industrial productivity—Europe—Congresses. 2. Industrial productivity—United States—Congresses. 3. Labor productivity—Europe—Congresses. 4. Labor productivity—United States—Congresses. 5. Information technology—Economic aspects—Europe—Congresses. 6. Information technology—Economic aspects— United States—Congresses. 7. Employees—Effect of technological innovations on— Europe—Congresses. 8. Employees—Effect of technological innovations on—United States—Congresses. 9. Income distribution—Europe—Congresses. 10. Income distribution—United States—Congresses. 11. Unemployment—Europe—Congresses. 12. Unemployment—United States—Congresses. 13. Solow, Robert M.—Congresses. I. Greenan, Nathalie. II. L’Horty, Yannick. III. Mairesse, Jacques. HC240.9.I52 P764 2002 3380 .064—dc21 2001059646

Contents

Contributors vii Foreword ix Introduction: The Puzzling Relations between the Computer and the Economy 1 Nathalie Greenan, Yannick L’Horty, and Jacques Mairesse I The Productivity Puzzle

17

1 The Mismeasurement Hypothesis and the Productivity Slowdown: The Evidence 19 Jack E. Triplett Comments

47

Dietmar Harhoff 2 Information Technology, Organizational Transformation, and Business Performance 55 Lorin M. Hitt and Erik Brynjolfsson Comments

93

Jacques Mairesse 3 Innovation and Employment: A Critical Survey Vincenzo Spiezia and Marco Vivarelli Comments

133

Marie-Claire Villeval

101

vi

Contents

II The Inequality Puzzle

139

4 Technological Bias and Employment Inequality: A Macroeconomic Perspective 141 Henri R. Sneessens Comments

171

Jean-Pierre Laffargue 5 Technical Change and the Structure of Employment and Wages: A Survey of the Microeconometric Evidence 175 Lucy Chennells and John Van Reenen Comments

225

Eric Maurin 6 By What Means Does Information Technology Affect Employment and Wages? 229 Kathryn Shaw Comments

269

Nathalie Greenan Conclusion: Computers, Productivity, and Wages; Reflections on the Economics of the Information Age 279 Timothy F. Bresnahan Index

299

Contributors

Tim F. Bresnahan Stanford University and NBER USA Erik Brynjolfsson, MIT Sloan School of Management and NBER USA Lucy Chennells Bank of England UK Nathalie Greenan Centre d’Etudes de l’Emploi and CEPREMAP France Dietmar Harhoff INNO-tec, Ludwig-Maximilians University, Munich Germany Lorin M. Hitt University of PennsylvaniaWharton School USA Jean-Pierre Laffargue CEPREMAP and Paris I University France

Yannick L’Horty EPEE, Evry Val d’Essone University France Jacques Mairesse CREST-INSEE, EHESS, and NBER France Eric Maurin CREST-INSEE France Kathryn Shaw Carnegie Mellon University and NBER USA Henri R. Sneessens Catholic University of Louvain and Free Faculty of Lille Belgium Vincenzo Spiezia International Labor Office Switzerland Jack E. Triplett Brookings Institution USA

viii

John Van Reenen University College London and CEPR UK Marie-Claire Villeval CNRS-GATE France Marco Vivarelli Catholic University of Piacenza Italy

Contributors

Foreword

For many casual observers, the spectacular development of information and communications technology (ICT) could account for the remarkable economic growth record in the United States and the relatively poor performance and mass unemployment in many European countries during the 1990s. It could similarly explain the large increase in wage inequalities and/or unemployment of unskilled workers on both sides of Atlantic since the 1980s. These current views, however, have raised deep-seated puzzles for professional economists. If computers can be seen everywhere, they do not show up evidently in growth and productivity trends, nor in data on the skill structure of employment and wages. This is, on the one hand, the well-known Solow paradox or productivity paradox, and on the other hand, what can be called in parallel fashion the inequality paradox. In fact, in searching for the effects of ICT, economists hardly find them in macro statistics. They have to analyze carefully the micro data evidence to disentangle such effects from the various other phenomena at work. Our principal aim in organizing this volume has been to provide a comprehensive examination of these two general paradoxes or sets of puzzling issues. We have asked among the best experts from both sides of the Atlantic to cover and discuss the evidence, both empirical and theoretical, and at the micro and the macro level. Their challenge was to be synthetic and to write in language accessible to a general audience interested in economic questions, and not only to the economists specializing in the field. Besides an introductory overview and a concluding essay, the volume is composed of six chapters, each focusing on different aspects of the issues at stake, and each of them being followed by a set of comments.

x

Foreword

This volume originated in a two-day international conference on Information and Communications Technologies, Employment, and Earnings, which we organized in Nice and held in June 1998. The conference was sponsored by the Conseil de l’Emploi, des Revenus et de la Cohe´sion Sociale (CERC), and by the 10th CREST-NBER Franco-American economic seminar. Eric Brousseau, University of Nancy II; Michel Dolle´, CERC; Jean-Luc Gaffard, University of Nice– Sophia Antipolis; Michel Gollac, Centre d’Etudes de l’Emploi; JeanPierre Laffargue, University Paris I; and Alain Rallet, University of Paris IX Dauphine, were the members of the scientific committee. We are thankful to the authors and the discussants for having gone through a number of extensive revisions of their initial contributions to the conference. We are indebted with Jean-Pierre Laffargue and Michael Piore for providing helpful comments and insightful suggestions on a first draft of the book manuscript. We are also grateful to The MIT Press for having supported our project with goodwill and patience, and to Françoise Leveleux and MarieChristine Thomas from CERC and Sophie Bontemps from Centre d’Etudes de l’Emploi for their efficient help at its different stages. Nathalie Greenan, Yannick L’Horty, and Jacques Mairesse

Introduction: The Puzzling Relations between the Computer and the Economy Nathalie Greenan, Yannick L’Horty, and Jacques Mairesse

The economic consequences and effects of the development and diffusion of information and communication technologies (ICTs) are being widely discussed on both sides of the Atlantic, in Europe and the United States. The issues concern economic growth and employment outcomes, on the one hand, and employment and wage inequalities, on the other. That is to say, ICTs are viewed as driving simultaneously economic growth and inequalities. Advocates for the ‘‘new economy’’ say that the effects of ICTs on American growth and employment are important, a few of them going even so far as to attribute to ICT some twenty million jobs created in the United States since the 1991 recession. By way of contrast with the U.S. situation, the voices raised in Europe argue that the (alleged) slow pace of ICT diffusion is part of the explanation of European unemployment and more generally a symptom of ‘‘Eurosclerosis.’’ Indeed, in the 1990s, macroeconomic performances took different directions in Europe and in the United States. To illustrate this point, we examine the French and American growth and employment records since the 1991 recession in the United States and the 1993 recession in France.1 During that period (until 1999), the gross national product (GNP) grew at a slower rate in France than in the United States, by about a 1.3 percentage point each year (figure I.1). The gap in employment rates has also kept widening over the period, the average differential being about 1.2 percentage point lower in France than in the United States (figure I.2). Since the early 1990s, advocates for the ‘‘new economy view’’ have sought in ICT the major impetus behind growth. This view, although not yet a dominant one, has been gaining the support of economists and of the general public as well. Many economists, however, have

2

Nathalie Greenan, Yannick L’Horty, Jacques Mairesse

Figure I.1 U.S. and French growth since the last recession. Source of data: BEA and INSEE, national account.

Figure I.2 U.S. and French employment since the last recession. Source of data: BLS and INSEE, national account.

Introduction

3

been reluctant to wholeheartedly discard what they call the Solow ‘‘productivity paradox.’’ Economics Nobel laureate Robert Solow is widely quoted as writing in 1987: ‘‘You can see the computer age everywhere but in the productivity statistics.’’ Solow wrote this at a time when the U.S. economy was confronted with a prolonged period (since the mid-1970s) of poor productivity performance. It actually was not until the late 1996 that the U.S. productivity growth significantly quickened. While it may not be theoretically or factually inconsistent to see large positive impacts of ICT on output and employment growth and none, or even negative ones, on productivity growth, such a conjunction does not seem too likely, and particularly so at the macro level and over the long run. Indeed, both common sense and economic history tend to agree that in the long run, technical change is a basic source of productivity and productivity is a basic source of economic growth. If ‘‘paradox’’ is too strong a term to describe the relations between ICTs and economic growth and productivity, one can say at least that it has the character of a ‘‘puzzle’’ and has thus created a ‘‘puzzling’’ problem for economists. Another set of issues concerns the effects of ICTs on the structures of wages and employment. Since the early 1980s, and on both sides of the Atlantic, inequalities have been growing. Depending on the relative ‘‘flexibility’’ or ‘‘rigidity’’ of the labor markets, such increasing inequality expresses itself in a declining share of unskilled workers in total employment (and an augmenting share in total unemployment), as in France, or mainly affects the magnitude of wage differentials, as in the United States (figure I.3). Both phenomena were observed to be present in a number of other European countries. The general opinion of economists is that these persistent and pervasive trends in the relative wages and employability of skilled versus unskilled workers is inherent to the nature of ICT. The increasing use of computers and reliance on computer applications in all spheres of activities, from management to production and from manufacturing industries to services, put strong demands on technical skills, while also requiring more general capabilities and good background education. In other words, in professional economic parlance, technical progress as embodied in the development of ICT use is not ‘‘neutral’’ but ‘‘skill biased.’’ There are of course other possible explanations for the rising inequalities in the labor force, and particularly the development of international trade. Western firms achieve competitive advantage by trading with

4


Figure I.3 Comparison of wage inequalities in the United States and France. The ratio D9/D1 is an indicator of wage disparities. D9 represents the lowest wage in the upper 10 percent group of workers with the highest wages and D1 the highest wage in the lowest 10 percent group of workers with the lowest wages. Source of data: OECD, unemployment outlook data base.

‘‘cheap’’ labor developing countries and/or by reallocating their labor-intensive activities to these regions. When the effects of ICT on growth and productivity are viewed together with the effects on wages and employment, the economic picture becomes even more puzzling. Even if in pure economic logic the various effects can be reconciled, they do not really fit well with one another. In particular, it seems hard to believe that, at both the micro level of the firm and the macro level of the economy, computers have changed the skills composition of the work force and increased wage differentials, whereas they have had no significant impact on productivity. Finally, there is one more consideration that makes these issues even more puzzling and the debates more controversial. We have taken for granted the general opinion, also shared by most political leaders and their economic advisers, that the pace of ICT diffusion in Europe has been much slower than in the United States. But the facts are not so clear-cut. The importance of ICT in an economy depends on both the growth of the ICT sector and the evolution of ICT use in other sectors. The United States (and Japan) are far ahead of Europe in the production of computer equipment and generic software. However, they do not

Introduction

5

Figure I.4 Comparison of ICT use at work in the United States and France. A polynomial was used to derive the time series trend. Sources of data: U.S. Bureau of the Census for noninstitutional computer use at work by population aged 18 and over in 1984, 1989, 1993, and 1997; INSEE and DARES, Enqueˆte techniques et organisation du travail, for computer use in France in 1987 and 1991 and DARES, Enqueˆte condition de travail, for comparable figures for 1993 and 1998.

have the leading edge in telecommunications equipment or in the service activities that are located in proximity of their consumers (maintenance, consulting, etc.). If ICT use is roughly measured by the share of computer users in the workplace, Europe does not seem to be much behind the United States. This indicator shows, for example, that France was lagging the United States by about three years in the 1980s and early 1990s, but by 1998, the gap had become negligible, the share of ICT users reaching 50 percent of the work force in the two countries (figure I.4).2 The diffusion of ICT has been pervasive in Western economies, with various organizational consequences and impacts on economic performances, both at the macro and micro levels. But differences between countries in the diffusion of ICT use do not seem large enough to account for growth, employment or inequality differentials patterns. Our aim in producing this volume has therefore been to explore the puzzling relations between the computer and the economy through a careful examination of the empirical evidence and theoretical analysis of this evidence. We tried to adopt a ‘‘transatlantic point

6


of view’’ by combining evidence from both the United States and Europe, at both the micro and the macro levels. In focusing on the economic impact of ICT use, and not on the economic effects of the production of hardware or software, a broad definition of what constitutes ICT is generally privileged.3 The volume is a collection of nontechnical surveys. They set out the key questions, assumptions, and tools that economic research has produced to investigate the impacts of ICTs. They illustrate various methodological approaches, combining lessons from fieldwork or from small statistical samples of firms, national surveys of management practice, results of microeconometric studies, and theoretical analyses at the macro level. More precisely, the volume is composed of two parts: the first deals with the productivity puzzle and the second one with the inequality puzzle. There are three chapters to each part: one mostly focused on measurement or econometric issues, one focused on micro mechanisms backed up by fieldwork results, and a third one taking a macroeconomic and theoretical point of view. Each chapter is followed by the comments of a discussant. The volume ends with a concluding chapter by Timothy Bresnahan. The Productivity Puzzle The conclusions from applied research on the relation between productivity and ICT diffusion largely depend on whether they are based on a macroeconomic or a microeconomic analysis. The productivity paradox may be in part a reflection of the difficulties of passing from one level of analysis to the other. The first part of the volume thus addresses the issue from three different angles: Does the productivity paradox amount mainly to problems of statistical measurement (of product, price, quality of equipment, etc.)? Can it be explained by the effects of ICTs on the organizations of firms and market structures? What do we know about the effects of ICTs on employment levels? Is the Productivity Paradox a Statistical Fiction? For one group of observers, the use of ICTs has not increased the productivity of firms. Poor implementation of computer technology on an unfavorable micro- or macroeconomic environment may have

Introduction

7

even resulted in lowered productivity. More generally, while ICTs may improve areas of performance (product differentiation, manufacturing flexibility, lead times, etc.), there has been no measurable gain in productivity. These observers believe, however, in a possible increase of productivity in the long run, with the unfavorable trend mainly reflecting a short- to medium-run adaptation process. For another group of observers, the favorable effects of ICTs are measurable using existing tools when the decomposition of measured output changes between volume and price changes is revised to better reflect improvements in the quality of goods and services produced. The first chapter, by Jack Triplett, discusses this ‘‘measurement hypothesis,’’ reviewing what is known about measurement bias in U.S. statistics, pre- and post-1973. The author explains that the mismeasurement hypothesis is about differential mismeasurement. The hypothesis requires that mismeasurement be worse after the productivity slowdown began than it was previously, the price index being more strongly upward biased, and/or the output measures being more strongly downward biased. Triplett’s analysis uses the available, late 1999, statistics, which have been revised after the Boskin Commission’s recommendations on the measurement of the consumer price index (CPI). Subsequent to the Commission’s report, the Bureau of Labor Statistics made a number of changes to its methodology which have been estimated to reduce the growth in the CPI index by about a 0.7 percentage point each year. This revision plus other changes introduced by the Bureau of Economic Analysis resulted in an increase of the growth rate of U.S. GDP since 1959 of about 0.2 percentage points per year and in the growth rate of labor productivity since 1973 of 0.4 percentage points per year. Based on these numbers, Triplett’s conclusion is that the measured productivity slowdown since 1973 cannot be explained by a mismeasurement problem. The evidence for mismeasurement is persuasive, but the evidence for differential mismeasurement, which is what the mismeasurement hypothesis requires, is not. Jack Triplett also examines the ‘‘new economy view’’ of innovation and technical change and its implications. In this view we are overwhelmed by an unprecedented flow of innovations and new products not reflected in the productivity numbers. For these innovations to improve productivity growth, they must arrive at an increasing rate and not only in an increasing number: an ever larger number of new products is required just to keep the productivity rate constant.

8


The empirical work thus far is inconclusive as regards this ‘‘increasing rate hypothesis.’’ The Effects of ICT on Productivity and Organization Lorin Hitt and Erik Brynjolfsson in their survey chapter consider a variety of case studies on ICT implementation, explore the available econometric evidence of its impact on productivity, and examine, in particular, studies that condition these impacts on organizational innovations. They show that while there are strengths and weaknesses in all the individual studies, collectively they paint a very compelling picture in which firm organization plays a critical role in enhancing the efficiency of ICT use. ICTs improve productivity by enabling complementary organizational innovations. The cost of investments in ICTs is relatively low, but the efficiency of these investments is conditional on organizational changes, which can be very costly. In order to obtain the full benefit of ICT use, firms must allow more autonomy to the lower levels of their hierarchies, reduce the number of hierarchical layers, hire a workforce with higher skills and education, spend more on training, and so on. Whenever the costs of organizational change are not taken into account in firms that get reorganized, the measured productivity impact of ICT, or their impact on the firm financial value, may be overstated. Because these costs are very difficult to measure, they tend to be ignored in most studies. ICTs are also more frequently associated with increases in the intangible component of output, for example, when they allow new ways of delivering value to consumers as well as new kinds of interactions with suppliers. In fact, both the output and the relevant inputs raise measurement problems. However, overall, there is no evidence of a productivity paradox, at the firm level. The Employment Effect of ICT The effect of technical progress on employment differs whether we consider a shorter or a longer time horizon. In the short run, when the demand level is fixed, labor-saving technical progress translates into a reduction in the level of employment. In the long run, when prices are fully flexible, demand adjustments are conducive to increased economic activity and employment. In a general inter-

Introduction

9

temporal equilibrium framework, this positive long-term effect counterbalances the negative short-term one. Labor-saving technical change finally raises growth, productivity, and wages but has no impact on the employment level. The question is, does this textbook presentation really apply to ICTs? Vincenzo Spiezia and Marco Vivarelli, in the third chapter, start by examining the relationship between ICT diffusion and employment levels, from a macroeconomic and mainly theoretical point of view. Firms that produce ICTs form a new sector in the economy, and ICTs are used by firms to generate new processes, new organizations, new products, and new services. From a static and anecdotal point of view, it is easy to point out the labor-saving impact of ICT use. But from a dynamic perspective, all the indirect effects of a new technology have to be taken into account when making a macro assessment. Because it is a generic technology, ICT diffuses throughout the whole economy and induces a large range of compensation effects. As far as the available empirical evidence is concerned, contrasting results emerge according to the different levels of analysis. Microeconometric studies tend to find a positive correlation between ICT and employment at the firm level. At the sector level, it is relevant to introduce the distinction between product innovation and process innovation. Finally, at the economy level, contrasting empirical results may arise from variations in the relative importance of product and process innovations and from different degrees of effectiveness in compensation mechanisms. The authors thus conclude that the pattern of ICT diffusion in Europe cannot be directly related to the persistence of high levels of unemployment. The Inequality Puzzle There is a growing concern in advanced countries that the situation of less-skilled workers has deteriorated, both in their ability to secure jobs and in their ability to earn a living wage. The conventional facts in this domain differ on both sides of the Atlantic: the deterioration of unskilled employment is more marked in Europe and the increase in wage inequality is more important in the United States. Many observers have claimed that the computerization of the economy has something to do with these trends. The second part of the book aims at investigating such claim.

10


Economists put forward the hypothesis of skill-biased technological change to explain the growing over-representation of the least skilled workers in unemployment figures in many countries since the 1980s. The concept of technological bias was first introduced in economic analysis to account for the strong tendency of production processes to become more capital intensive. What is new in the recent debate around the technological bias is the focus on the skill composition of the workforce. Today’s skill-biased technological change induces an upward drift in the relative efficiency of skilled workers and a downward drift in the cost share of unskilled workers. As a macroeconomic concept, skill-biased technological change has no direct implication on how wages and employment adjust, but when it is included in a general equilibrium framework, it implies an increase in wage inequality between skilled and unskilled workers. Moreover wage rigidity leads to an increase in employment inequality across skills. Of course, technological bias is only one explanation for the unemployment of unskilled workers or for an increase in the relative wage of skilled workers. Other explanations focus on competition from low-wage economies, certain aspects of the expansion of the service sector, the existence of a ladder effect in labor markets, the changes in structure of relative labor costs, the role of education and training, and so on. Each of these explanations has different consequences in terms of economic policy. Is skill-biased technological change the main explanation of unemployment among unskilled workers in Europe and of increased wage inequality in the United States? Can anything be said about the respective roles of the other explanations? What are the consequences for economic policy? The Impact of ICTs on the Structure of Employment and Wages at the Macroeconomic Level It is very easy to misinterpret aggregated data on the average rate of unemployment and its dispersion across skills, age groups, and the like. In order to avoid such misinterpretation, we need a consistent analytical framework. Henri Sneessens proposes such a framework. Although it does not distinguish information technology improvements from other technological changes, it provides an important starting point to better understand ICT macro impacts. It shows that

Introduction

11

an increase in unemployment rates or in wage dispersion across skill levels is not necessarily implied by skill-biased ICT diffusion. Sneessens presents a general and simple macroeconomic model in which observed changes in the difference of unemployment rates between low- and high-skilled workers result from both macroeconomic and structural shocks. Macroeconomic shocks are supposed to be initially symmetric: that is to say, they have the same direct initial impact on skilled and unskilled workers, as during a global productivity slowdown, for instance. Structural shocks do not. They are supposed to be initially asymmetric. A change in labor force composition, a higher minimum wage, or skill-biased technological change are examples of such shocks. The chapter shows that the final impacts may differ from the initial and direct impacts. Thus sometimes macroeconomic shocks generate asymmetric effects, changing the dispersion of both unemployment and wage rates. Conversely, structural shocks may have some symmetric effects via, for example, the flexibility of wages and endogenous changes in the labor force composition. As a result mismatch indicators like relative employment rates or the standard deviation of unemployment rates per skill are uninformative as to the nature of the shocks. As in the case of the productivity paradox, measurement issues limit our empirical ability to assess correctly the nature of the phenomenon and its underlying causes. The Impact of ICTs on the Structure of Employment and Wages at the Firm Level For many observers the automation of industrial production processes during the 1960s and 1970s allowed the productivity of relatively unskilled workers to be increased. This movement is alleged to have been the cause of a ‘‘de-skilling’’ of the workforce. In the 1980s and 1990s, numerous studies by American economists on the skill impact of technological change describes precisely the opposite effect: ICT adoption benefits to skilled white-collar employees to the detriment of the unskilled blue collars. Lucy Chennels and John Van Reenen survey the effects of technical change on skills, wages, and employment by examining the recent econometric evidence at the industry, firm, plant, and worker levels. They focus on studies that use direct measures of technology rather

12


than indirect measures like estimated time trends. But they adopt a large definition of technological improvements that is not limited to ICT. Although their main concern is labor demand, they give some consideration to the analysis of labor supply and wage-setting mechanisms. Chennells and Van Reenen show that a major problem in this research area is the absence of a conventional definition of technological change, which complicates and weakens comparisons. It is also difficult to extrapolate at the macroeconomic level the results of econometric studies mainly based on micro data. But in their opinion, the main difficulty in assessing the impact of ICT on the skill mix and relative wages lies in the need to control for unobserved heterogeneity and endogeneity. Partly because of weaknesses in the available data, only a small number of studies deal with these methodological issues even though failure to do so may induce strong biases in estimated coefficients. Econometric studies show considerable evidence of a correlation of technology with skill structure at the firm level, supporting the skill-biased technical change hypothesis. Technological improvement by the firm generally induces an increase in the relative demand for skilled labor. In some studies this result appears robust to fixed firm and/or worker effects, but such is not always the case. Studies conducted at the worker level produce another interesting, although less clear-cut result: average wages seem to increase with ICT use. This result is found in studies based on cross-sectional data, but not in studies that try to take advantage of longitudinal data to control for workers’ heterogeneity. Thus it seems difficult to interpret the computer–wage correlation as a causal effect of technical change on wages. If there is a wage premium for computers users, this premium may not be directly due to the actual use of a computer and is likely to reflect the fact that computer users tend to be selected among the most able workers who were already earning higher wages. By What Means Do ICTs Affect Employment Patterns and Wages? What are the channels, direct or indirect, through which the diffusion of ICTs may reinforce the share of skilled workers in employment or raise their relative wage? What role does organizational change or selection behavior play in the whole process?

Introduction

13

ICTs can have a positive or a negative effect on the demand for skills, but the net outcome is an increase in the relative demand for skilled workers. Kathryn Shaw in her chapter uses the steel industry as an example of an industry that is undergoing a drastic evolution with important changes in its labor force skill composition, new organizational practices, and increasing computerization. Her work is based on detailed studies of working practices in the steel mills, mainly in the United States but also in France and Japan. Shaw’s principal conclusion is that ICT use is complementary with changes in human resource management (HRM) practices. Overall, it is the combined effects of ICT use and innovative organizational practices that shape the firm’s demand for production workers. Innovative firms in the steel industry need workers with special skills that were not previously required: more advanced cognitive skills, communications skills, negotiations skills, and interpersonal skills. These new skill requirements are only partially fulfilled by formal education. When they are hiring a worker, innovative firms are not only concerned by his diplomas: they are looking for the right person. This is why hiring decisions are often based on a screening process that did not exist twenty years ago. Another measurement problem lies in this selection process: although operatives in the steel industries have different abilities than in the past, their skill level has remained unchanged. On the whole, ICT use and HRM practices do not lead necessarily to a decrease in the employment of the unskilled workers. If ICT contribute to improve firm performances, it can raise output and hence the level of employment of the less skilled as well as the skilled workers. There are many changes in labor demand and wage policies that are not influenced by ICTs. The trends that have been observed in the steel industry result from various causes: the lower demand for steel, international competition, development of new technologies leading to the existence of minimills, among other things. This chapter clearly suggests that the channels that would lead to a skill-biased technological change at the macro level are probably quite indirect and that trying to isolate the ICT effect is particularly difficult. The economic and management literature often describes a process where firms that invest in new generations of computers change the skill structure of their labor force, with a preference for more skilled workers: they increase their management staff and en-

14


hance the skill level of their production workers. Kathryn Shaw tells us that in the steel industry, things seldom work that way. First, the change in the skill structure of the workforce is more often the outcome of the process of plant or firm births and deaths. Second, companies that already have the most skilled workforce are usually the most innovative ones in terms of ICT use and HRM practices. To Conclude In the final chapter Timothy Bresnahan has transformed what was originally an after-dinner speech at the end of intense conference proceedings into thought-provoking reflections on the economics of the information age, and this puts much of what we learn in the preceding chapters into a coherent perspective. He accounts for the productivity puzzle by stressing the potential benefits of what he calls organizational computing (OC). But it is only through a combination of linked improvements in the production and organization of the firm, as well as in the difficult to measure quality attributes of output (availability, reliability, variety, etc.) that these benefits will materialize and be appropriated. This process of co-invention, as much as that of organizational computing per se, also explains the increasing demand for workers with higher and different skills, and hence the inequality puzzle. Bresnahan ends in predicting that by the same causes having similar effects, much of what we have seen in the past will continue to prevail in the era of the Internet, with current interorganizational computing. Without going as far as Timothy Bresnahan in his explanation of the puzzling relations between computers and the economy and his forecast about the near future, let us summarize some of the important messages of the surveys gathered in this volume. ICTs have important economic consequences, but they are not as massive as is usually thought. For a proper assessment, one has to distinguish these consequences at the firm level, the market level, and the macroeconomic level. At the firm level, ICT use is correlated with an increase in the share of skilled workers, when a large and informal definition of skill is adopted, and with an increase in performance, depending on the complex relationship between the ICT and the firm’s organization. However, there are no clear effects on firms’ growth and wage policies. The main reason is that the uses of ICTs remain largely

Introduction

15

open ended. ICTs are compatible with changes in work organization such as, at one extreme, increased workers’ decision rights, multitasking, and horizontal information structure, and at the other extreme, increased hierarchical control and supervision, higher formalization of work content, and vertical information structure. Many firms are on a continuous learning curve, using processes that have not yet brought about easily replicable and stabilized patterns of use. Thus, unlike changes in the technical characteristics of ICTs as described, for example, by Moore’s law, changes in ICTs use are slow, unspectacular, and complex. This is not to deny the impressive progress embedded in every new generation of computers or automated machines, but it is important to underline the difference between the innovations in equipment and the innovations in the use of that equipment. There is no revolution on the user’s side comparable to the one observed on the producer’s side. At the market level, the outcome of ICT changes are also very difficult to determine. Use of computer technology not only changes the way companies work internally but also affects intercompany relations. The changes may allow large companies to work like small ones. They may also allow small companies to work like big ones, when they are organized in networks and when they pool certain resources. More generally, ICTs change the state of relations between consumers and suppliers, modify distribution conflicts between principals and agents, and alter market structures. By changing what determines the limits between organization and market, computer technology transforms the nature of the firm as well as that of the market. Computer technology may increase firm and market efficiency greatly, but we do not know exactly who is going to benefit from this greater efficiency in the end. At the macroeconomic level, ICT diffusion cannot really account for the employment, productivity, or growth trends, on either sides of the Atlantic. The increasing wage and employment inequalities between unskilled and skilled workers are also far from being fully understood and ascertained, even if technological change remains the dominant explanation in the United States. Research in this area that distinguishes between different types of technologies has found that the effect of their adoption on the return to skills is ambiguous. Some technologies, like advanced manufacturing tools, seem to favor relatively unskilled labor, while others, like personal computers, work against it. Overall, these technologies do not seem to

16


be intrinsically destructive of unskilled jobs. Their industry-level or macro-level impact depends on their relative weight in overall investment. Complementary investment that accompanies adoption of new technologies, whether tangible or intangible as in organizational change, play an essential role in their effects on employment opportunities and wage differences. This volume is thus another illustration of the missing bridge between results obtained at the micro and macro levels. Clearly, there are no direct and mechanical linkages but a long and twisting path from ICTs to any macroeconomic variable. This view falls somewhere in between those of the optimistic supporters of the new economy and the pessimistic detractors of the new technologies. Notes 1. Although this volume has its origin in a Franco-American conference and the three editors are French, we compare the U.S. figures with the French ones mainly because the macroeconomic trends in France are close to the European Union average during the 1990s. 2. This is not true for households. Government statistics in both countries show that by the end of the 1990s nearly 40 percent of American households owned a personal computer compared to nearly 19 percent of French households. 3. There is no firmly established definition and measure of the ICT diffusion because of insufficient data, the rapidly changing supply of technologies, and the differing measurement approaches. A broad definition could include all the products and processes that embed a microprocessor, and a narrow one could be confined to certain office technologies such as the use of personal or mainframe computers. Definitions may thus differ from one chapter of this volume to another, depending on the questions and methodologies.

I

The Productivity Puzzle


1

The Mismeasurement Hypothesis and the Productivity Slowdown: The Evidence Jack E. Triplett

1.1

Introduction

The Nice conference took place at nearly the end of a period in which productivity research was dominated by the great productivity slowdown that began in 1973. From 1949 to 1973 nonfarm multifactor productivity increased yearly 1.9 percent in the United States, and labor productivity grew at the rate of 2.9 percent. From 1973 to 1996, these growth rates were 0.4 percent and 1.4 percent, respectively (table 1.1).1 With the publication in 1999 of revised U.S. national accounts data, productivity researchers became aware that the second half of the 1990s was marked by a resurgence of productivity growth in the United States. Attention quite naturally turned from analysis of the post-1973 slowdown to the post-1995 acceleration of productivity. Although this chapter reviews economic measurement issues in the post-1973 period, not the post-1995 period, economic measurement issues are always timely. Economists have a natural reaction to almost any unexpected economic development: Is it a problem with the data? Indeed, that is the right reaction, for problems of measurement are always with us. Reviewing whether the post-1973 productivity slowdown is due to measurement error is of value to future economic analysis as much as it is to economic history. The mismeasurement hypothesis suggests that some or all of the measured productivity slowdown is accounted for by increased mismeasurement of output since 1973. For several reasons many economists find the mismeasurement hypothesis appealing. First, most other hypotheses for explaining the productivity slowdown in the 1970s (e.g., oil price shocks) have been eliminated by productivity research. Because economic measurements are never

20

Jack E. Triplett

Table 1.1 U.S. labor and multifactor productivity, average annual rates of change, 1949–96, and selected subperiods Nonfarm business

Manufacturing

Estimated nonmanufacturing

Output per hour 1949–73 1973–96 1973–79 1979–96

2.9 1.4 1.2 1.5

2.6 2.7 2.1 2.9

3.0 1.1 0.9 1.2

Multifactor productivity 1949–73 1973–96 1973–79 1979–96

1.9 0.4 0.4 0.4

1.5 0.7 0.6 1.1

2.1 0.4 0.8 0.3

Sources: Output per hour: U.S. Department of Labor, Bureau of Labor Statistics, http:// www.bls.gov/ lpc/ home.htm. Multifactor productivity: U.S. Department of Labor, Bureau of Labor Statistics, http://www.bls.gov/mfp/home.htm.

precise, the mismeasurement hypothesis cannot be subjected to a direct statistical test, so it more or less wins (or perhaps loses) by default. The mismeasurement hypothesis is the last remaining singlehypothesis explanation for the slowdown. Second, the U.S. productivity slowdown was concentrated in nonmanufacturing. Since 1973 economywide multifactor productivity declined more than did manufacturing multifactor productivity. In the private nonfarm business sector, the post-1973 slowdown was 1.5 index points (1:9 0:4; compare the 1949–73 and 1973–96 lines in the bottom panel of table 1.1), and the manufacturing slowdown was 0.8 points (1:5 0:7). Thus productivity in the nongoods sectors fell nearly two percentage points, or more than the decline in the aggregate figures.2 Nongoods, or services in general, are ‘‘hard-tomeasure’’ sectors, so if their measured productivity lags behind the productivity of ‘‘easy-to-measure’’ sectors, it is natural to suspect that measurement is the villain in the productivity slowdown. Third, the share of poorly measured sectors of the economy has been growing, as Griliches (1994) has pointed out. The goodsproducing sectors—agriculture, mining, manufacturing, and construction—accounted for just under 50 percent of U.S. employment in 1940; now, the share is around 30 percent, and shrinking. If we do not know how to measure a ‘‘services economy,’’ and the share of services is rising, then the U.S. economy’s output may be more

The Mismeasurement Hypothesis and the Productivity Slowdown

21

poorly measured now than it was in the past, even if our ability to measure the output of every sector is as good as it ever was. Fourth, as Griliches (1994) has also noted, much U.S. high-tech equipment investment has gone to sectors of the economy where output is poorly measured. In the 1992 capital flow table (Bonds and Aylor, 1998), about 43 percent of U.S. computer equipment (computers plus peripherals) went to four industrial sectors: finance, wholesale trade, miscellaneous renting and leasing, and business services. Adding insurance and communications pushes the share above half. Many economists suspect that the potential productivity growth generated by high-tech capital equipment in banking, insurance, business services, wholesale trade, and similar ‘‘hard-tomeasure’’ industries has been lost in the productivity numbers. Finally, the ‘‘new economy’’ view contends that so many innovations, new products, and so forth, are clearly visible that the stagnant productivity figures just can’t be right. If the mismeasurement hypothesis applies to the United States, it probably applies to other countries as well. The post-1973 productivity slowdown was widespread across OECD countries. As in the United States the share of services has been growing in all advanced countries. Although price statistics have been subject to criticism in the United States (CPI Commission, 1996), in my experience no country’s government statistics reflect satisfactorily the measured output of the services industries, nor solve the difficult problem of adjusting price indexes for quality change (which is a major part of the problem in measuring price and output in services industries). This chapter reviews the mismeasurement hypothesis of productivity growth. It is a companion to another study (Triplett 1999a) that reviews the ‘‘Solow productivity paradox,’’ the puzzling fact that productivity growth has fallen precisely during the computer– ‘‘information age’’ technological revolution.3 One of the explanations for the computer paradox is the mismeasurement hypothesis, which is the topic of this chapter. 1.2

The ‘‘New Economy’’ View of the Productivity Slowdown4

Many economists believe there must be a mismeasurement story in the productivity slowdown, because they see more technical change, more new products, more changes in consumer service, in methods of delivery, and other innovations than is consistent with the modest

22

Jack E. Triplett

rise in government productivity numbers. We have a ‘‘new economy’’ in this view, inundated with an unprecedented flow of innovations and new products, and none of this flow of the new is discernible in the productivity numbers. This new economy view is prevalent in the newspapers, in business publications and even in Federal Reserve Bank reviews; we hear it at conferences. It once was true, the story goes, that products were standardized and therefore easy to measure. Today, we are told, there is an unprecedented stream of new products and quality improvements and customized products to meet market niches, product cycles are shortening to an unprecedented degree, new services from industries such as banking and finance are being introduced with a rapidity that is unprecedented historically, and the chairman of the Federal Reserve Board has been quoted to the effect that the unprecedented current level of technological innovations is a once in a century phenomenon that will yield an enormous upward surge in productivity. In the new economy view, the productivity slowdown is not directly related to computers. Rather, people are stacking up and cumulating anecdotes, whether from within their own companies or from what they read in the newspapers or hear other people saying. Those cumulated anecdotes do not seem consistent with the aggregate productivity numbers. From this point of view, it is not so much a belief that the computer has increased productivity, but rather a belief that productivity has improved, based on other evidence. Indeed, in a widely cited book review, Robert Solow (1987) makes the same point: ‘‘[The authors], like everyone else, are somewhat embarrassed by the fact that what everyone feels to have been a technological revolution, a drastic change in our productive lives, has been accompanied everywhere . . . by a slowing-down of productivity growth, not by a step up’’ (emphasis supplied). No doubt the anecdotes about new products, new services, new methods of distribution, and new technologies are valid observations. Although no one knows how to count the number of ‘‘new’’ things, I would not seriously dispute the proposition that there is more that is new today than there was at some time in the past. Yet the anecdotes wholly lack historical perspective, and for that reason they are misleading as evidence on the productivity slowdown and the mismeasurement hypothesis.


23

Table 1.2 Productivity from new products Periods

0

1

2

3

10

Number of products

100

110

121

133

259

20 673

Number of new products

10

11

12

13

26

67

Productivity change

10%

10%

10%

10%

10%

10%

To have an impact on productivity, the rate of new product and new technology introductions must be greater than in the past. A simple numerical example makes the point (table 1.2). Suppose that all productivity improvements come from the development of new products. Suppose further that in some initial period 100 products existed and that 10 percent of the products were new. In the following period, there must be 11 new products just to keep the rate of productivity growth constant, and in the period after that 12 new products are required. At the end of ten years, a constant productivity rate requires 26 new products each year, and after twenty years, 62 new products, and so on it goes, as the arithmetic of compounding increases shows. As the economy grows, an ever larger number of new products is required just to keep the productivity growth rate constant. Most of the anecdotes advanced as evidence for the ‘‘new economy’’ amount to assertions that there are a greater number of ‘‘new’’ things. A greater number of new things is not necessarily a greater rate of new things. As an example, many economists have cited the number of products carried in a modern grocery store as evidence of increased consumer choice, of marketing innovations, and so forth.5 One study reported that in 1994 there were more than twice as many products in the average grocery store than in 1972 (19,000, compared with 9,000). But in 1948 the number was 2,200, so the 1948 to 1972 rate of increase (from 2,200 to 9,000, or 6.0 percent a year) was nearly twice as great as the 1972 to 1994 increase (3.5 percent a year). Thus it is true that in 1994 there were many more products in grocery stores than there were two decades before; but the rate of increase has fallen. Some other illustrations enhance the point. The Boskin Commission cited welfare gains from the increased availability of imported fine wines. Because of the great reduction in transportation costs, we now get Australian wine in the United States at low prices (as low, in my experience, as in Australia). That is certainly an increase in the

24

Jack E. Triplett

number of commodities available, and an increase in welfare. But is the increase in tradable commodities a larger proportionate increment to choice and to consumption opportunities than the increments that occurred in the past? Diewert (1993) cites an example, taken from Alfred Marshall, of a new product in the nineteenth century: decreased transportation costs, owing to railroads, made fresh fish from the sea available in the interior of England for the first time in the second half of that century. Considering the very small number of consumption goods then available to the average worker, and even allowing for the fact that the fresh fish were undoubtedly initially consumed mostly by the middle class, was the introduction of fresh fish a smaller proportionate increase in the number of new commodities than is the availability of Australian wine and similar goods a century later? Perhaps the best answer to this question is: we do not know. But we also have looked at the decade of the 1990s with far too short a historical perspective. In developing a related point, Mokyr (1997) refers to ‘‘the huge improvements in communications in the 19th century due to the telegraph, which for the first time allowed information to travel at a rate faster than people. . . . The penny post, invented . . . in the 1840s, did an enormous amount for communications—compared to what was before. Its marginal contribution was certainly not less than Netscape’s.’’ One could go on. My numerical example above implies that every new product has the same significance as an earlier one. If we simply count, then the average new product of the 1990s must equal the significance of automobiles and appliances in the 1920s and 1930s (e.g., home air conditioners first became available in the early 1930s), and of television and other communications improvements in the 1940s and 1950s (e.g., mobile telephones were introduced in the 1940s). If the average significance of each new product in the 1990s is not as great as for new products from the past, then the number of them must be greater still to justify the new economy view of the paradox. The same proposition holds for quality change. It is amazing to see quality improvements to automobiles in the 1990s, great as they have been, held up as part of the unprecedented improvement story, or—as in a press account I read recently—quality change in automobiles given as an example of the new economy, contrasted with a


25

ton of steel in the old. Actually the first thing wrong with that contrast is that quality change in a ton of steel has been formidable. Second, quality change in automobiles is a very old problem in economic statistics; it did not emerge in the 1990s as a characteristic of the new economy. Hedonic price index methodology was developed in the 1930s to deal with quality change in automobiles (Court 1939). The study by Raff and Trajtenberg (1997) suggests that the rate of quality improvement in automobiles was greater in the first decade of the twentieth century than in its last decade. Much of what has been said about the new economy is true. What has been lacking is a proper historical appreciation for the magnitudes and significance of new product introductions and quality change in the past, which is necessary for an appreciation of the relative contribution of the new economy to productivity and to economic growth. While the number of new products and ‘‘new things’’ is greater than before, the real issue is whether the rate of introduction of new products is unprecedented historically. I do not believe that we know the answer to that question. If the number of new products is a measure of productivity improvement, then we must have an increase in the rate of introduction of new products and not just an increase in the number, in order to have increased productivity growth. The mismeasurement hypothesis has gained acceptability partly because some economists have mistakenly been counting new innovations on an arithmetic scale, and—finding more of them—have thought they have evidence confirming it. They ought to be looking at a logarithmic scale, a scale that says you must turn out ever greater numbers of ‘‘new products’’ (or new products of greater significance) to keep the current rate of new products up to the rates of the past. We look at the new products and new technical changes at the end of the twentieth century, and we are tremendously impressed by them. We should be. It is clear those new products are increasing welfare, and the technical innovations are contributing to output. But are they increasing at an increasing rate? Is the number of new products increasing more rapidly on a logarithmic scale? That is not clear at all. For the ‘‘new things’’ to improve productivity, they must be increasing at an increasing rate. I think it safe to assert that the empirical work in economic history that would confirm the in-

26

Jack E. Triplett

creasing rate hypothesis has not been carried out.6 Until it is, the mismeasurement hypothesis must be evaluated by examining information about possible biases in economic statistics, which is the topic of the next section. 1.3

The Mismeasurement Hypothesis

Current Understatement of Productivity Growth First, we must find evidence that the current rate of productivity growth is understated (not just mismeasured). Many economists believe that price measures used for deflation are overstated, which would lead to an understatement of deflated (real) output and therefore of productivity. In 1996 the Advisory Commission to Study the consumer price index, known as the Boskin Commission, estimated that the U.S. consumer price index contains an upward bias of 1.1 percentage points per year. Eldridge (1999, tab. 3) reported that 57 percent of the U.S. nonfarm business sector output was deflated with components of the CPI. Data in her article imply that applying the Boskin Commission’s bias estimate to real output growth would raise current estimates of productivity growth by about 0.6 percentage points a year.7 By late 1999 the available U.S. statistics reflected the extraordinary speed with which the U.S. statistical system had addressed the Boskin Commission’s recommendations. Subsequent to the Commission’s report, the Bureau of Labor Statistics made a number of changes to CPI methodology which have been estimated to reduce CPI growth by about 0.7 points per year. Then, in June 1999, BLS released a ‘‘research series’’ that carried methodological improvements in the CPI—the recent ones and also changes made in the 1980s—back to 1978 (Stewart and Reed 1999). This lowered the average rate of increase in the CPI. When the Bureau of Economic Analysis (BEA) revised the U.S. national accounts in October 1999, it incorporated the parts of the CPI research series that apply to the accounts, plus other changes, including treatment of software as a part of investment, and therefore of final output, and an improved measure of banking output. In total, revisions raised the growth rate of U.S. real GDP for the period 1959 to 1996 by about 0.2 percentage points a year, from about 3.2 percent per year to about 3.4 percent


27

(U.S. Department of Commerce 1999b), and by about the same amount for the 1973 to 1996 period. In November 1999 the Bureau of Labor Statistics published new labor productivity numbers that used the revised nonfarm business sector output from BEA. These revised labor productivity figures are reported in table 1.1. As a result labor productivity estimates for 1973 to 1996 were revised upward by 0.4 percentage points, somewhat more than the revision to GDP (because the revisions to nonfarm business sector output were greater than the revisions to GDP). Because so much of the recent discussion of measurement issues has involved the CPI, it is important to note that productivity growth could also be mismeasured because of biases in non-CPI deflators used in personal consumption expenditures (PCEs) and biases in investment, government and foreign trade deflators. Additionally there are nondeflator biases in GDP—for example, measurement errors in current price measures of consumption and investment.8 An assessment of output mismeasurement requires answers to those nonCPI measurement questions. For productivity, one should also ask whether there are measurement biases in inputs. Exploring any of these matters is beyond the scope of the present chapter. Differential Mismeasurement It is sometimes thought that evidence of mismeasurement in current statistics confirms the mismeasurement hypothesis, but that is not so. The mismeasurement hypothesis is a hypothesis not just of mismeasurement but of differential mismeasurement. The hypothesis requires that errors in economic measurements are larger in the more recent period, that output was measured better before 1973 than it is today. For mismeasurement to explain the productivity slowdown, there must be more mismeasurement after 1973 than there was before 1973. For example, the Boskin Commission found upward bias in the current CPI but did not make any judgment about whether the CPI had less bias before 1973. The BLS research CPI index (Stewart and Reed, 1999) provided a post-1978 CPI that incorporated twenty years of measurement improvements. Some of those improvements raised the CPI. For example, an aging bias to the CPI rent index caused it to understate the true increase in rent (this has been shown repeatedly in research; the best reference is Randolph, 1987). Correction of the aging bias raised

28

Jack E. Triplett

the rate of growth in the CPI. Other improvements lowered the CPI’s rate—for example, substitution of geometric means for arithmetic means in calculating CPI elementary indexes (Stewart and Reed, 1999). On balance, methodological improvements since 1978 lowered the rate of growth in the CPI. However, measurement improvements in the CPI do not mean that the pre-1973 CPI was more accurate than the post-1978 one, as the mismeasurement hypothesis requires. Indeed, some of the adjustments incorporated into the research CPI correct for long-standing price index measurement problems (e.g., aging bias in housing), which remain uncorrected in the pre-1973 CPI, so in these cases the earlier data may contain more mismeasurement.9 If the same error affects the pre-1973 and post-1973 CPI, the mismeasurement bias might exist in the productivity numbers for both periods but not be a factor in the productivity slowdown. Similarly the recent upward revisions to GDP, and to nonfarm business output, have raised the rate of growth of labor productivity for the post-1973 period. The same adjustments were not carried back very far before 1973. Some would likely have raised the pre1973 productivity growth rate. The new GDP treatment of software (which raises the estimate of output growth) of course has a larger impact on the post-1973 period because there was far less software investment before 1973. Differential mismeasurement implies one or more of several things: that in compiling price statistics, agencies do a poorer job today adjusting for quality change and measuring the hard-to-measure services than they did in the past, or that the amount of quality change is greater than it used to be (this is often asserted, but not documented, as pointed out in section 1.2), or that measuring services is for some reason more difficult than it used to be (perhaps because the nature of services has changed), or that the sectors where mismeasurement exists have become more important than they were before 1973. Thus we face not only the difficult task of assessing the degree of mismeasurement today, which in itself requires much research, but also determining the size of the bias that existed 25 to 50 years ago.10 Abrupt Changes The mismeasurement hypothesis implies that measurement changes, if they occurred, must have been abrupt because the productivity


29

slowdown was abrupt. Though there is some debate about whether productivity started to slow in 1973, or whether signs of it were visible in the United States even earlier, around 1968, the slowdown was not a gradual reduction in the rate of productivity improvement. If mismeasurement is to account for the productivity slowdown, then we must find some fairly abrupt change in measurement practices, or an abrupt increase in measurement problems, or an abrupt increase in the size of the poorly measured sectors of the economy. So far as I know, none of these factors changed abruptly in 1973. International The productivity slowdown affected most industrialized economies at about the same time as the United States and in roughly the same magnitudes. Perhaps the U.S. Bureau of Labor Statistics (which publishes the U.S. price indexes and productivity measures) and the U.S. Bureau of Economic Analysis (the compiler of GDP) did things better in the ‘‘old days,’’ as seems to be implied by the views of some U.S. economists who subscribe to the mismeasurement hypothesis. But could economic statisticians in all countries all ‘‘forget’’ in concert?11 1.4

Evidence on Mismeasurement and Productivity

I lack sufficient knowledge to review the statistics of very many OECD countries, so I will focus on those of the United States. Strategy The U.S. national accounts distinguish approximately 200 components of consumption expenditures, and more than 800 for investment. Components of real final demand are estimated and deflated separately. For example, consumer expenditure on automobiles is deflated by a price index for automobiles. A list of deflators for the different components of health care expenditures appears in McCully (1999). Some measurement errors occur at the aggregate level, and affect aggregate price or output measures directly. An aggregate level measurement error is substitution bias, which arises in the CPI because it uses a fixed weight, Laspeyres index number. The Boskin

30

Jack E. Triplett

Commission estimated the substitution bias in the CPI at 0.15 index points per year. The substitution bias does not, however, affect economy wide U.S. output and productivity measures because deflation occurs at the detailed level, as already noted, and aggregations of real output in the U.S. national accounts and in the major U.S. productivity measures use superlative index numbers.12 Superlative indexes are free of substitution bias (Diewert 1976 introduced the concept of superlative indexes). Most other measurement errors are specific to particular products and industries. For example, the growth of discount food stores may bias food price indexes (the Boskin Commission ‘‘guesstimated’’ this bias at 0.1 index point per year), but it probably does not affect the automobile price index; conversely, overlooked or inadequately adjusted quality change may bias the automobile and electronics products price indexes, but probably not the price index for bananas. When measurement errors are specific, the effects of mismeasurement show up in the trends for individual commodities or industries that are mismeasured. One approach, then, is to look for anomalies in the behavior of commodities or industries or final demand categories where mismeasurement is thought to be most severe. In the following section I examine trends in final demand categories of consumption. Measurement Errors in U.S. Output and Productivity Statistics: Real Consumption Consumption is a large proportion of GDP (around 70 percent). Consumption also comprises a large share of the output of the private nonfarm business sector, which corresponds to the U.S. productivity measures in table 1.1. If measured real consumption growth is too low, this will bias estimates of aggregate productivity growth downward. Consumption Growth Rates13 U.S. real per capita consumption growth slowed after 1973, in tandem with productivity, but by a smaller amount.14 Real per capita consumption grew 2.6 percent each year from 1949 to 1973 (3.0 percent each year from 1959 to 1973); see table 1.3. In the previously available consumption data from the U.S. national accounts, real per


31

Table 1.3 Average annual percentage growth rates: Productivity, compensation, and consumption

Productivity (output per

hour)a

1949–73

1959–73

1973–96

2.8

2.9

1.4

Real hourly compensationa

2.8

2.4

0.7

Real per capita disposable personal income Real per capita consumption (old data)

2.8

3.3

1.7

2.6

3.0

1.7

Real per capita consumption (new data)

2.6

3.0

1.9

Sources: Productivity data: U.S. Department of Labor, Bureau of Labor Statistics, 1999a. Real hourly compensation data: U.S. Department of Labor, Bureau of Labor Statistics, 1999a. Real per capita disposable personal income data: U.S. Department of Commerce, 1999, file 807 (NIPA table 8.7). Real per capita consumption old data (1949, 1959, and 1973) data: U.S. Department of Commerce, Bureau of Economic Analysis, 1998, table 8.3. Real per capita consumption old data, 1996 data: U.S. Department of Commerce, Bureau of Economic Analysis, 1999a, table 8.3. Real per capita consumption new data: U.S. Department of Commerce, 1999, file 203 (NIPA table 2.3) and file 807 (NIPA table 8.7). a. Nonfarm business sector.

capita consumption grew 1.7 percent a year from 1973 to 1997 (table 1.3). The growth rate declined by around 35 percent after 1973, in the former data. In October 1999 BEA revised the U.S. national accounts. Post-1973 per capita consumption growth rates were revised substantially upward, to 1.9 percent a year; pre-1973 consumption growth was revised only trivially. In the revised data, the real per capita consumption growth rate declined after 1973 by around 25 percent, using the 1949 to 1973 interval for comparison, and not by 35 percent as the former data showed. So consumption growth was more rapid than was previously thought. Around a third of the post-1973 decline in per capita consumption growth has been revised away. This ‘‘natural experiment’’ can be used to examine implications of the mismeasurement hypothesis. We have in this instance revised data that have less mismeasurement than the unrevised data, although the revised data may still contain measurement error. Improving the measurement raised the aggregate post-1973 consumption growth rate, which is in the direction that the mismeasurement hypothesis predicts. As a result there should be fewer signs of mismeasurement, or fewer signs of the implications of mismeasurement, in the revised data than in the unrevised data.

32

Jack E. Triplett

Aggregate measurement error in real consumption is made up from specific errors that affect particular components: it can be an error in the estimation of the current-price consumption expenditures for a particular component, or an error in the individual price index used for deflating that component. If the overall growth rate of real consumption is too low because of a measurement error, as the mismeasurement hypothesis suggests, the measurement error ought to show up in slower growth rates for consumption components that are difficult to measure (e.g., high-technology electronic products, and services) compared with components that have fewer measurements problems (e.g., apples and bananas). This argument will be considered more fully in the following subsection, but here we need first to examine the trends in the major subaggregates of consumption. Unless otherwise stated, I will use the most recent (revised) data for the post-1973 comparisons. Growth rates for nearly every major consumption category— durables, nondurables, and services—declined after 1973, no matter which data are used (table 1.4). The only two exceptions to this statement are other durables and other nondurables. Other durables is a category that includes computers and electronic equipment, which have experienced rapid quality changes. Other services (i.e., nonhousing) are thought to be difficult to measure. Using 1959 to 1973 rates for comparison, post1973 growth rates in these two components declined by 13 percent and 31 percent, respectively, in the old data—by 5 percent and 20 percent, respectively, in the revised data. So the hard-to-measure components showed more growth relative to (smaller slowdowns than) total consumption. Among the major subcomponents, food and housing15 are relatively easy to measure. Both of these components experienced per capita consumption increases at half the rate after 1973, using the old data. In the revised data, food consumption growth was relatively close to the pre-1973 rate. Housing consumption was also revised downward; its growth rate for the most recent period is around a third of its pre-1973 rate of growth. These revisions are perplexing. However, in neither the old data nor the new are the easy-tomeasure components growing relative to the growth rate of total consumption, as the mismeasurement hypothesis implies. Thus, whether by new or old data, the post-1973 per capita consumption growth rate slowdown is actually smaller for the harder to

PCE durables

PCE nondurables

PCE services

Total PCE

All durables

Motor vehicles

Other durablesa

All nondurables

Food

Other non- All durablesb services

Housing

Other servicesc

1947–73 1959–73

2.57 3.00

4.10 5.34

5.38 5.51

2.73 5.10

1.39 1.89

0.97 1.13

1.98 2.97

3.29 3.39

4.13 3.52

2.98 3.33

1973–97 (old)

1.80

3.07

1.40

4.43

0.92

0.44

1.41

2.12

1.69

2.29

1973–97 (new)

2.06

3.06

1.34

4.82

1.30

0.81

1.86

2.38

1.62

2.67

1973–84 (old)

1.72

2.51

1.97

3.09

0.83

0.54

1.14

2.27

2.10

2.33

1973–84 (new)

1.91

2.60

2.02

3.43

1.01

0.68

1.42

2.49

2.05

2.68

1984–97 (old)

1.87

3.55

0.91

5.57

1.00

0.35

1.64

2.00

1.35

2.25

1984–97 (new)

2.19

3.45

0.77

6.01

1.55

0.92

2.23

2.29

1.27

2.67

Sources: Old 1973–1996 data: U.S. Department of Commerce, Bureau of Economic Analysis, 1998, table 2.3. Old 1997 data: U.S. Department of Commerce, Bureau of Economic Analysis, 1999a, table 2.3. New data: U.S. Department of Commerce, 1999, file 807 (NIPA table 8.7). a. Fewer motor vehicles. b. Less food. c. Less housing.


Table 1.4 Average annual growth rates: Real per capita personal consumption expenditures

33

34

Jack E. Triplett

measure components than for the easy-to-measure components, which is inconsistent with the implications of the mismeasurement hypothesis. Elasticities The mismeasurement hypothesis can also be examined by looking at income elasticities, or their empirical counterparts, real expenditure elasticities (the elasticity of a component of consumption with respect to the growth of total real consumption expenditure). In demand analysis it is customary to use total expenditures rather than measured income to avoid the difficulties in measuring saving. Suppose that the true structure of expenditure elasticities was unchanged before and after 1973, in particular, that the true marginal expenditure elasticities were unchanged. Suppose further that real consumption growth is understated (strictly, differentially understated) after 1973 in the hard-to-measure components, as the mismeasurement hypothesis suggests, but is measured more or less accurately in the easier-to-measure components. Then easyto-measure components ought to show higher measured marginal elasticities after 1973, and the measured marginal elasticities of the hard-to-measure components ought to fall after 1973, relative to their true marginal elasticities. To show this relationship more exactly, let YT designate the growth rate of the true value of real consumption. This is composed, approximately, of YT ¼ a þ ðPi yTi wi =w1 Þ w1 ðPj yTj wj =w2 Þ w2 þ ei ;

ð1Þ

where a is a nonhomotheticity parameter that plays no role in what follows, and yTi and yTj are the true growth rates for sets of consumption components, i and j, for which wi and wj are index number P P wj ¼ w2 ). The shares of well-measured weights ( wi ¼ w1 and and poorly measured components are w1 and w2 , respectively, and ei is a residual term, necessary because the U.S. national accounts are Fisher quantity indexes, which are not consistent in aggregation.16 The growth rate of real consumption as measured, that is, YM , contains measurement errors in components j. Thus YM ¼ a þ ðPi yTi wi =w1 Þ w1 ðPj yMj wj =w2 Þ w2 ;

ð2Þ

where the components i are measured without error (or else, their error is the same both before 1973 and after) and the components j


35

are measured with error after 1973 but not before (or else, their measurement error is greater after 1973). Then, of course, YT > YM

ð3Þ

because the mismeasurement hypothesis implies that consumers are experiencing greater aggregate real consumption growth than the data show.17 Easy-to-Measure Components The mismeasurement hypothesis implies that the marginal expenditure elasticities of the easy-to-measure components should rise after 1973 with respect to the measured growth rate of total consumption. Before 1973 the true (and measured) marginal budget share of, say, food is wf ¼ qpf qf =YT , where q designates a growth rate (recall that YT is already a growth rate). If the mismeasurement hypothesis is true, and aggregate real consumption is understated after 1973, consumers will spend part (equal to the true marginal elasticity) of the unmeasured real income on the easy-to-measure components, like food. Since food is measured accurately (or relatively accurately), the measured marginal budget share of food ðwf ¼ qpf qf =YM Þ rises after 1973 because the denominator of the share (measured total consumption, YM ) is too low. Moreover the increase in the measured elasticity for the wellmeasured components estimates the proportionate mismeasurement in the aggregate. The marginal expenditure elasticity for a consumption component is approximately its growth rate over the growth rate of total PCE. These ratios are tabulated in table 1.5 for two relatively easy-tomeasure subaggregates (food and housing), for two subaggregates that contain hard-to-measure components (durables less vehicles and services less housing), and for the residual categories (motor vehicles and other nondurables). Food In the old BEA data the marginal elasticity for food consumption was generally falling (from 0.38 to 0.31 to 0.19; see the first row of table 1.5). Note that all these elasticities are consistent with Engel’s law. The falling pattern, however, is contrary to the expectation derived from the mismeasurement hypothesis, in which measured elasticities of easy-to-measure components should rise if mismeasurement exists in aggregate consumption.

36

Table 1.5 Approximate marginal expenditure elasticities, hard to measure and easy to measure consumption components pre- and post-1973 1947–73

1959–73

1973–97 (old)

1973–97 (new)

1973–84 (old)

1973–84 (new)

1984–97 (old)

1984–97 (new)

Food

0.38

0.38

0.24

0.39

0.31

0.35

0.19

0.42

Housing

1.61

1.17

0.94

0.79

1.22

1.07

0.72

0.58

Durables, less vehicles

1.06

1.70

2.46

2.34

1.80

1.80

2.98

2.75

Services, less housing

1.16

1.11

1.27

1.30

1.36

1.40

1.20

1.22

Motor vehicles

2.09

1.83

0.77

0.65

1.15

1.06

0.49

0.35

Nondurables, less food

0.77

0.99

0.79

0.90

0.67

0.74

0.88

1.02

Source: Computed from data in table 1.4. Note: Computed are relative growth rates (component/PCE).

Jack E. Triplett


37

The new BEA data contain a big upward revision in per capita food expenditure for the most recent period. As a result the marginal elasticity for food is now higher than it was for the pre-1973 period (0.42 compared with 0.38). The behavior of food consumption in the new BEA data is now consistent with the mismeasurement hypothesis,18 where it was not consistent in the old data. Before interpreting these changes as evidence for the mismeasurement hypothesis, note a counterfactual: food is supposedly an easy to measure commodity. These big national accounts revisions suggest that it may not be so easy. To some extent this apparent anomaly in the food consumption data originates from restaurant meals (the category named ‘‘purchased meals and beverages,’’ in the U.S. national accounts). It includes, of course, fast food meal providers as well as conventional restaurants. Restaurant meals have a substantially larger marginal elasticity than grocery store sales of food (in the national accounts, ‘‘food purchased for off-premises consumption,’’ which from the title ought to include takeout meals purchased from restaurants, coffee shops, fast food providers, and the like but probably does not). In the most recent BEA data, the restaurant meals elasticity is 1.00, for the 1984 to 1997 period, which, if anything, seems low. On the other hand, the restaurant meals elasticity was only 0.47 from the 1947 to 1973 data, which seems far too low. Nevertheless, much of the rise in the elasticity for food comes from the contribution of a rising elasticity for restaurant meals, and a substantially increased share of total food consumption. Housing Housing is a different case, with different results. In the CPI a substantial change in the treatment of owner-occupied housing occurred in 1983, but its treatment in the national accounts has been consistent throughout.19 Putting the measurement of housing consumption in the easy-tomeasure class does not imply that there are no measurement difficulties at all. Randolph (1987) showed that rent price indexes have an inherent downward ‘‘aging’’ bias, because of imperceptible deterioration in the units priced, combined with the practice of ‘‘linking out’’ rent changes that accompany renovations that restore the unit to its original condition. The aging bias has been corrected in the CPI, and the aging adjustment was applied retrospectively back to

38

Jack E. Triplett

1978 by Stewart and Reed (1999). This aging adjustment has also been applied to the national accounts deflators for housing in the recent 1999 GDP revision. This means in effect that before 1978 the deflator for housing was downward biased (Randolph’s results have been replicated in other research extending back many years), so the growth of housing consumption was upward biased before 1978. The aging adjustment, however, is not sufficiently large to account for the substantial decline in the expenditure elasticity for housing in table 1.5. Housing research indicates that this is a commodity with approximately unitary income elasticity, or it is slightly income elastic. Either case corresponds to the elasticities in table 1.5 for 1959 to 1973 and, in the former data, for 1973 to 1984. However, even the former BEA data shows that the elasticity for housing declined markedly in the 1984 to 1997 period. The revisions have pushed housing elasticity down even further, so it is no longer that much greater than the elasticity for food (0.58 compared to 0.42). Housing provides no evidence that is consistent with the mismeasurement hypothesis. However, the revised BEA data imply per capita housing consumption that does not seem plausible. Hard-to-Measure Components By reasoning similar to that above, we can expect the marginal expenditure elasticities of hard-to-measure commodities to fall after 1973, relative to their true elasticities. In this case, wj ¼ qpj qj =YM ; both the numerator and the denominator are measured with error (the error may arise from mismeasurement of either pj or qj ). But the (downward) error in the numerator is proportionately larger because the denominator contains the accurately measured components of consumption as well as the mismeasured ones.20 Hard-to-measure categories include other durables and other services. Services are always hard to measure. Other durables includes electronics, which pose particularly difficult problems for price and output measurement because of rapid quality change. Services In table 1.5 services are mildly income elastic, and the elasticities are approximately constant over the whole postwar period, especially in the revised BEA data. The mismeasurement hypothesis suggests that measured services elasticities should fall. That is only true for the intervals preceding and following 1984, where the


39

elasticity falls from 1.40 (1973 to 1984, new data) to 1.22 (1984 to 1973). Services provide no support for the hypothesis that the post1973 slowdown was caused by mismeasurement. Other Durables This category is an anomaly. If these products were mismeasured, their marginal elasticity should have fallen after 1973. Instead, it rose tremendously (to 2.75, in the new BEA data). This shift is far more than can be accounted for by increased demands for computers. Electronic goods, including radios and TV’s, were present in the former, pre-1973 data, after all. Motor Vehicles Before leaving this topic, a discussion of motor vehicles is necessary. Automobiles fit most economists’ notion of a product that is hard to measure. I have not put them into that category because of a judgment that they are relatively well measured in the U.S. national accounts compared with other products of similar complexity that pose similar problems of quality change. In the U.S. price indexes, automobiles are subject to a complex quality adjustment procedure that uses data from manufacturers on the cost of quality improvements. The cost-based quality adjustments in the U.S. CPI and PPI automobile price indexes are very detailed, and they apply adjustments for even very minor quality improvements reported by manufacturers. For example, automobiles have been made more rust resistant through the use of superior metals, antirust coatings, and improved body construction; the extra cost of coated metal and similar changes are reported to the BLS by manufacturers, and the cost of the improvement is adjusted out of the price of the improved car.21 The procedure is described in Triplett (1990), which contains a table showing the annual values of cost-based quality adjustments made in both indexes. For example, the average price of cars in the United States increased by $536 in 1981; the quality adjustment in the same year was $530. In most years, however, the adjustments have been smaller. The BLS cost-based quality adjustment procedure has been in place since about 1959 or 1960. That means that a substantial methodological change occurred in automobile price indexes around 1959.22 Looking at the entries for motor vehicles in table 1.5, however, the 1959 methodological change does not seem to matter very much. The

40

Jack E. Triplett

big change in elasticies occurred, not in 1959 but in post-1973, and again post-1984. Indeed, the slowdown in real per capita expenditures on new automobiles has been the largest slowdown for any durable goods category (data not shown in the tables). This is puzzling for three reasons. First, it seems likely that for various reasons manufacturers overstate the costs of quality changes (Triplett 1990). This would artifically push upward the real quantity indexes. Second, through the entire post-1973 period, until just this last year, BLS counted government-mandated air pollution equipment as an improvement in automobile transportation, and adjusted it out of the indexes (for criticism of this approach, see Triplett, 1990, and the Boskin Commission). This again makes the automobile price index rise too slowly, real consumption rise too rapidly, and the major error arises post-1973 (table 7.1 in Triplett 1990 tabulates the cost adjustments for mandated smog and safety devices in the CPI and PPI). Third, all of the U.S. hedonic price indexes for automobiles that have been published in the last 30 years have suggested that U.S. price indexes for autos rise too slowly, and not too rapidly. All three of these considerations suggest that automobile price indexes were understated after 1973, so real expenditures on cars were overstated. The results for the PCE vehicle components in table 1.5 suggest that the research results are wrong, that quality change in cars is greater than the hedonic indexes show, and that it is greater than the quality adjustments that have been made in the CPI and PPI. Although this does not mesh with my own judgment on auto price indexes, it raises a further issue whose full exploration cannot be undertaken here. Summary The revised BEA data show that the U.S. real per capita consumption slowdown is not concentrated in the hard-to-measure areas of consumption. They are thus inconsistent with the mismeasurement hypothesis, as were the former BEA data (see Triplett 1997). Indeed, the declines in expenditure elasticities for housing, and in motor vehicles, the two components that show the greatest post-1973 consumption growth slowdown, are particularly large, but the increase in the food elasticity is suspect, though it appears to be going in the right direction for the mismeasurement hypothesis. Elasticities for the hard-to-measure components either rise after 1973 (other durables)


41

or remain constant (other services). Taken together, these elasticities are consistent, if anything, with overstatement of real consumption growth after 1973 (or with understatement of it before 1973). BEA’s natural experiment has improved the measurement of consumption but not lowered the marginal elasticities of the wellmeasured components nor raised the marginal elasticities of the poorly measured components of consumption. So, once again, the data go in the wrong direction to support the implications of the mismeasurement hypothesis. The revisions to the rate of total consumption growth should have lowered the measured marginal elasticities of the well-measured components, by reducing the amount of mismeasurement in the denominator. They did not. While my calculations do not prove that the mismeasurement hypothesis is wrong, they indicate that consumer demand data are not consistent with one set of implications of the hypothesis. Several qualifications might be considered. Perhaps the mismeasurement exists in the ‘‘easy-to-measure’’ components, so that the test is in some sense mis-specified. Or alternatively, BEA improved the measurement, but they did it in the components that were already relatively well measured, and they did not do it where most economists’ priors put the highest probability for mismeasurement (services and electronics), where quality change is thought to be most rapid and new products are most frequent. Or perhaps the demand specification employed in this chapter is too simple. A more realistic specification of consumer demands for detailed components of PCE could yield results more in keeping with the mismeasurement hypothesis. For example, the mismeasurement hypothesis pertains to the difference between the true elasticities and the measured elasticities. The test used the assumption that the true elasticities were constant after 1973, which might not have been the case. On the other hand, there are only six goods in tables 1.4 and 1.5; at this fairly aggregate level, constancy of elasticities seems broadly consistent with findings in the demand literature. Exploration of any of these alternatives is beyond the scope of this chapter. 1.5

Conclusions

Inevitably any conclusions have elements of subjectivity in them. Do I think there is mismeasurement in the data? Of course I do.

42

Jack E. Triplett

Economic measurement is seldom precise, and other evidence suggests serious mismeasurement in the output data in the services sectors (Triplett and Bosworth 1999). Does the mismeasurement affect aggregate output and productivity? Again, the answer must be yes. Our measures of productivity are less accuarate than they need to be, particularly in the most dynamic parts of the economy. Does the mismeasurement reduce the level of productivity or its rate of growth? Again, the evidence suggests that the current measures of productivity understate to some unknown extent the true rate of productivity improvement. Is mismeasurement the cause, or a major cause, of the post-1973 productivity slowdown, consistent with the mismeasurement hypothesis? I doubt it. The evidence for mismeasurement is persuasive, but the evidence for differential mismeasurement, which is what the mismeasurement hypothesis requires, is not persuasive. The productivity slowdown was real. The problems of measurement limit our understanding of technical change in the economy of the twenty-first century. This is the reason for improving the measures, and this goal should guide the research agenda, not some hypothesis that the post-1973 slowdown would go away if economic measurement were improved. Notes 1. The labor productivity data incorporate 1999 revisions in GDP as discussed below. 2. The aggregate nongoods, or services, U.S. productivity number is not yet published. Through the entire postwar period U.S. farm productivity advanced faster than nonfarm productivity so productivity gains in the private business sector (also computed by BLS) exceed those for the nonfarm business sector. 3. The Solow productivity paradox refers to Robert Solow’s (1987) now famous aphorism: ‘‘You can see the computer age everywhere but in the productivity statistics.’’ 4. This section is adapted from section VII in Triplett (1999a). 5. Reservations might be expressed about this interpretation of the number of products in supermarkets. 6. I believe the supposed historical analogy to be a false one. See Triplett (1999a). 7. Eldridge presents no overall estimate. The 0.6 point number comes from adding her 0.19 point estimate (from her table 4) for the replacement of arithmetic mean estimators with geometric means for CPI basic components, 0.08 point for outlet substitution bias (e.g., price reductions from discount stores; Eldridge 1999, p. 42), and 0.32 point for quality change and new products (Eldridge 1999, p. 43).


43

8. Potential errors from current price measures of consumption are reviewed in Triplett (1997). 9. The post-1978 adjustment for CPI housing aging bias raises the post-1973 CPI relative to the pre-1973 CPI, and lowers output growth, post-1973, relative to pre-1973. But it is the higher pre-1973 output and productivity growth rates that are in error. 10. Upward bias in the CPI before 1973 was suggested in the Stigler Committee Report (1961). Unlike the Boskin Commission, the Stigler Committee made no point estimate of CPI bias. However, four of the five sources of bias discussed by the Boskin Commission were featured prominently in the Stigler Committee report, which added another bias—lack of probability sampling—in the CPI (this was subsequently implemented in the CPI in 1978). See Price Statistics Review Committee (1961). 11. Interestingly in Australia, which like the United States has enjoyed an improvement in productivity growth in the 1990s, it has been alleged that a productivity speed up might have been caused by measurement error (see Productivity Commission 1999, vol. 1, p. xxxviii). Subsequent to the Productivity Commission’s report, revisions to Australian GDP lowered the rate of Australian productivity improvement, the opposite effect from the U.S. GDP revisions, discussed above. 12. Most other OECD countries still use Laspeyres index formulas, but a growing number of them use chain Laspeyres indexes for real GDP. So long as relative price changes are not too great (a crucial qualification), chain Laspeyres indexes have relatively small substitution biases at the aggregate level. 13. In an earlier paper Triplett (1997), I examine the slowdown in U.S. real per capita consumption growth. This section updates some of the data in that paper and extends the analysis. 14. Although the growth of real per capita consumption slowed after 1973, U.S. living standards did not decline, contrary to numerous articles that have appeared in the press, and to statements by some economists who perhaps have not examined the data. From 1973 to 1997, real per capita consumption as measured in the national income and product accounts advanced by more than 50 percent in the unrevised data, and by 63 percent in the new data. I am unable to comprehend the origin of the completely mistaken idea that the U.S. has had stagnant living standards since 1973. 15. Housing is measured as rental equivalence in national accounts of nearly every country. Because housing in the U.S. consumer price index is also measured by rental equivalence (not the case in the CPI’s of many countries), there is a CPI measure of homeowner rental equivalence available for deflation in the U.S. national accounts. Housing prices are measured by taking month-to-month changes in identical rental units. Even though I have put it in the easy to measure category, that does not mean that housing poses no measurement issues. See Randolph (1987) and Stewart and Reed (1999). 16. In a Fisher index, higher-level aggregates cannot be written as multipicative combinations of lower level aggregates. However, the linear logarithmic form of equation (1) will approximate the aggregate Fisher index with only a small residual. 17. In (2) the weights will also differ from those in (1) if the measurement errors are in current price consumption rather than in the deflators. I neglect the weighting effects partly because they are so small and partly because they do not affect what follows. An extended discussion of the current price measures of consumption is contained in Triplett (1997).

44

Jack E. Triplett

18. In the mismeasurement hypothesis this post-1973 increase in the marginal elasticity of food is consistent with more measurement error before 1973 than before; it is also consistent with more measurement error in the revised data than in the unrevised data. For the reasons noted, this is perhaps overinterpreting the statistics. 19. In the revised CPI the price measure for owner-occupied housing was shifted to a rental equivalence approach. Formerly the CPI had included elements, such as the price of houses and mortgage interest cost, that greatly biased upward the measure of CPI housing costs in the inflationary period of the late 1970s. This CPI change is not an issue for the present chapter because the U.S. national accounts has always used rental equivalence for owner-occupied housing. 20. I am neglecting relative price effects and substitution elasticities here because they are small at this level of aggregation. Their sizes can be estimated precisely for this purpose by comparing growth rates of fixed-weight real consumption with the growth of the Fisher indexes of real consumption that are now published by BEA. Over the interval 1992 to 1997, these growth rates differed by only 0.2 percent. 21. The Boskin Commission thought that such improvements were missed in the BLS cost-based quality adjustment process, but this was apparently a miscomprehension of what is done. 22. Used car quality adjustments, however, were introduced much later. See Stewart and Reed (1999).

References Berger, A. N., and L. J. Mester. 1997. Efficiency and productivity change in the US commercial banking industry: A comparison of the 1980s and 1990s. Federal Reserve Bank of Philadelphia Research Working Paper 97/05, May. Bonds, B., and T. Aylor. Investment in new structures and equipment in 1992 by using industries. Survey of Current Business, vol. 78, December 1998. Court, A. T. 1939. Hedonic price indexes with automotive examples. In The Dynamics of Automobile Demand. New York: General Motors Corporation, pp. 99–117. CPI Commission, Advisory Commission to Study the Consumer Price Index. 1996. Toward a more accurate measure of the cost of living. Final Report to the Senate Finance Committee. Available at http://www.senate.gov/@finance/cpi.pdf. David, P. A. 1990. The dynamo and the computer: An historical perspective on the modern productivity paradox. American Economic Review 80(2): 355–61. Diewert, W. E. 1976. Exact and superlative index numbers. Journal of Econometrics 4(2): 115–45. Diewert, W. E. 1993. The early history of price index research. In W. E. Diewert and A. O. Nakamura, eds., Contributions to Economic Analysis: Essays in Index Number Theory Vol. 1, New York: North-Holland. Eldridge, L. P. 1999. How price indexes affect BLS productivity measures. Monthly Labor Review 122(2): 35–46. Gordon, R. J. 1990. The Measurement of Durable Goods Prices. National Bureau of Economic Research Monograph series. Chicago and London: University of Chicago Press.


45

Griliches, Z., ed. 1992. Output Measurement in the Service Sectors. National Bureau of Economic Research. Studies in Income and Wealth 56. Chicago and London: University of Chicago Press. Griliches, Z. 1994. Productivity, R&D, and the data constraint. American Economic Review 84(1): 1–23. Griliches, Z. 1997. The Simon Kuznets Memorial Lectures. Unpublished draft manuscript, October. Griliches, Z. Forthcoming. Comments on ‘‘What’s Different about Health? Human Repair and Car Repair in National Accounts and in National Health Accounts.’’ National Bureau of Economic Research Studies in Income and Wealth. Chicago: University of Chicago Press. Gullickson, W., and M. J. Harper. 1999. Possible measurement bias in aggregate productivity growth. Monthly Labor Review 122(2): 47–67. Hausman, J. A. 1997. Valuation of new goods under perfect and imperfect competition. In T. Bresnahan and R. Gordon, eds., The Economics of New Goods. Conference on Research in Income and Wealth, Studies in Income and Wealth 58. University of Chicago Press for the National Bureau of Economic Research, pp. 209–32. McCully, C. 1999. The treatment of medical care expenditures in the national income and product accounts. Paper prepared for the Brookings Program on Output and Productivity Measurement in the Service Sector. Workshop on Measuring Health Care. Washington, DC, December 17. Mokyr, J. 1997. Are We Living in the Middle of an Industrial Revolution? Federal Reserve Bank of Kansas City Economic Review 82(2): 31–43. Murray, R. 1992. Measuring public-sector output: The Swedish report. In Z. Griliches, ed., Output Measurement in the Service Sector. National Bureau of Economic Research Studies in Income and Wealth 56. Chicago: University of Chicago Press, pp. 517–42. Price Statistics Review Committee. 1961. The Price Statistics of the Federal Government. U.S. Congress, Joint Economic Committee. Government Price Statistics, Hearings, Part 1. 87th Congress, 1st Session, 1961; also published as: National Bureau of Economic Research, General Series 73. Productivity Commission. 1999. Microeconomic Reforms and Australian Productivity: Exploring the Links. Commission Research Paper. AusInfo, Canberra. Raff, D. M. G., and M. Trajtenberg. 1997. Quality-adjusted prices for the American automobile industry: 1906–1940. In T. F. Bresnahan and R. J. Gordon, eds., The Economics of New Goods. National Bureau of Economic Research Studies in Income and Wealth 58. Chicago: University of Chicago Press. Randolph, W. C. 1987. Housing depreciation and aging bias in the consumer price index. Bureau of Labor Statistics Working Paper 166. April. Solow, R. M. 1987. We’d better watch out. New York Times Book Review: 36, July 12. Stewart, K. J., and S. B. Reed. 1999. Consumer price index research series using current methods, 1978–98. Monthly Labor Review 122(6): 29–38. Triplett, J. E. 1990. Hedonic methods in statistical agency environments: An intellectual biopsy. In E. R. Berndt and J. E. Triplett, eds., Fifty Years of Economic Measurement:

46

Jack E. Triplett

The Jubilee of the Conference on Research in Income and Wealth. Chicago: University of Chicago Press for the National Bureau of Economic Research. Triplett, J. E. 1997. Measuring consumption: The post-1973 slowdown and the research issues. Federal Reserve Bank of St. Louis Review 79(3): 9–42. Triplett, J. E. 1999a. The Solow productivity paradox: What do computers do to productivity? Canadian Journal of Economics 32(2): 309–34. Triplett, J. E. 1999b. Economic statistics, the new economy, and the productivity slowdown. Business Economics 34(2): 13–17. Triplett, J. E., and B. Bosworth. 1999. Productivity in the service sector. Washington, DC: Brookings Institution. U.S. Department of Commerce, Bureau of Economic Analysis. 1998. National Income and Product Accounts of the United States, 1929–94. Washington, DC: Government Printing Office. April. United States Department of Commerce, Bureau of Economic Analysis. 1999a. GDP and related data. Selected NIPA tables. Available at http://www.bea.doc.gov/bea/dn/ nipatbls/maintext.htm. Accessed September 1, 1999. United States Department of Commerce. Bureau of Economic Analysis. 1999b. Initial results of the 1999 comprehensive revision of the national income and product accounts. Survey of Current Business 79(11): 1–42. U.S. Department of Commerce. 1999. Stat-USA Web site: http://www.stat-usa.gov. State of the Nation. State of the Nation Library. NIPA Information. Selected NIPA Tables 1959–1999: III (SELTABS.EXE). Accessed November 23, 1999. U.S. Department of Labor, Bureau of Labor Statistics. 1999a. Major sector productivity and costs index. Available at http://146.142.4.24/cgi-bin/dsrv?pr. Accessed November 23, 1999. U.S. Department of Labor, Bureau of Labor Statistics. 1999b. Major sector multifactor productivity index. Available at http://146.142.4.24/cgi-bin/dsrv?mp. Accessed November 23, 1999. Yuskavage, R. E. 1996. Improved estimates of gross product by industry, 1959–94. Survey of Current Business 76(8): 133–155.

Comments on Chapter 1 Dietmar Harhoff

It is common sense to take a method and try it. If it fails, admit it frankly and try another. But above all, try something. (Franklin D. Roosevelt)

Introduction The innocent and modest view of measurement holds that this activity requires little more than common sense and—maybe—some basic counting skills. A somewhat less modest, though still naive view, is to say that measurement may require complex statistics, but again, no economic theory is required, no complex and elaborate construction of hypotheses or constructs like utility or welfare—just mere counting in some simple or complex fashion. Economists are by no means naive individuals. Yet keeping away from measurement issues has been something of a tradition among them. Even the slightest glance at the real world of economic phenomena will teach a student that the attempt of measuring them is difficult, frustrating, and badly paid as a professional activity. Coming up with smart models of the counterintuitive sort is more respected than measuring economic phenomena with care and diligence. Measurement is to many economists a yawn-inducing topic with little academic appeal. Increasing the precision or appropriateness of economic measurement will gladly be left to people working at statistical agencies or other such institutions. Only when measurement issues start to touch the distribution of real(?) money, the problems will move to center stage. This is where they belong. And this is where Jack Triplett puts them in his chapter on the mismeasurement hypothesis and the productivity slowdown. To summarize his results briefly, he thinks that there are real difficulties in measuring productivity growth, and that

48

Dietmar Harhoff

these problems are particularly acute in sectors of the economy with a large degree of innovation. Does this make the measured productivity slowdown an artifact of our measurement procedures? Not quite, argues Triplett, there may after all be something to the slowdown. ‘‘The problems of measurement limit our understanding of technical change of the twenty-first century,’’ he argues. Therefore toward a better understanding of technical change, we need better measurement and more research on how to improve measurement. His view of the productivity slowdown makes this an almost natural phenomenon—as the rate of new products and their improvement is no longer maintaining some postwar level, productivity growth must decline. The following section discusses these arguments in turn. The subsequent one briefly turns to the German statistical system and comments on the extent of measurement problems in that context. The ‘‘New Economy’’ View The first argument that Jack Triplett addresses concerns the ‘‘new economy view’’ of the problem. In essence, this perspective is a position emerging from thorough arm-chair reasoning. It states that there must be measurement error in official statistics, since the great and unprecedented improvements that we have witnessed over the last two or three decades (i.e., the microprocessor, the personal computer, the Internet and the World Wide Web) surely must have contributed to notable gains in productivity. If the statistics fail to show this effect, they must be wrong. This view is often bolstered by an allusion to comparisons of the variety of products observed now and observed some time before. Jack Triplett’s answer to this view is twofold. First, while he personally is willing to agree that there may be more new products than, say, thirty years ago, he is not sure that the incremental number of new products is large when objectively compared to some time period in the past. Second, in order to support constant productivity growth, the rate of emergence of new products would have to be constant, assuming that each product yields the same contribution to productivity growth. Here he is extremely skeptical that this has been the case. Moreover it is not clear that the marginal contribution of past innovations has been any less than the marginal contribution of recently observed innovations, such as the World Wide Web. Thus, absent any quantitative support, the ‘‘new economy view’’ is currently a mere hypothesis, although pos-

Comments on Chapter 1

49

sibly more interesting than Jack Triplett suggests. He focuses in his chapter on the implications of a growing number of new products, and on differences between a constant growth rate and adding a constant number of products to the existing ones. But growth implications may also arise from changing interactions between products, such as increasing degrees of complementarity. Note that a development of this kind also implies that it may have become more complex to measure contributions to productivity correctly (at least at the micro level). Moreover, the argument that recent innovations such as the internet are truly unprecedented cannot be rejected easily, just as it probably cannot be supported empirically at this point. A Detailed Look at the Mismeasurement Hypothesis Is there evidence supporting the notion that mismeasurement is at the root of the apparent productivity slowdown? At first sight there is an appealing chain of logic. Almost all economists, even in Europe, are now aware of the discussion surrounding the report of the Boskin Commission. Hence, if conventional price measurement has yielded an overstatement of real prices in the past, then deflated (real) output will be understated. Indeed, the recent changes introduced by the Bureau of Economic Analysis (BEA) and the Bureau of Labor Statistics (BLS) appear to demonstrate that such an effect is present and substantial. However, it does not ‘‘explain away’’ the slowdown. As Jack Triplett notes, the mismeasurement hypothesis has to be a hypothesis about differential mismeasurement—about a phenomenon that became particularly strong in the early 1970s. Pre1973 measurement must have been more precise, he suggests. The most recent adjustments introduced into productivity measurement have been applied roughly to the time period after 1973. Would they also show the same effect when applied to pre-1973 data? To date, there does not appear to be a firm answer on this question. On his way to showing that there may be something real behind the productivity slowdown, Triplett also notes that the productivity slowdown has set in abruptly in 1973. Hence, if some form of mismeasurement is the underlying cause of the apparent slowdown, then it must have emerged equally abrupt. But there are no candidates, such as sudden changes in measurement techniques. A change of this kind, when limited to the U.S. statistical system, would also cause some embarrassment. How should one explain productivity

50

Dietmar Harhoff

slowdowns in other countries that occurred roughly at the same time as the one in the United States? What is therefore left is the hypothesis that by some coincidence of technological development and standard measurement techniques, a statistical artifact called the productivity slowdown may have emerged. It is here where Jack Triplett devises a number of ways to study the problem in detail, but he finds little support in favor of the mismeasurement hypothesis. For example, he shows that the slowdown in per capita consumption growth rates after 1973 is more pronounced for the easier to measure components than for the statistically difficult ones. This is not consistent with a hypothesis according to which measurement problems in the harder to measure components of consumption led to statistical artifacts. Expenditure elasticities for the hard to measure components tend to rise after 1973 or remain constant. This evidence would support the suggestion that real consumption growth was overstated after 1973, and not understated. Triplett also takes a close look at industry-level productivity measures. The services industries are the most interesting sectors to study, since they are the dominant receiving sectors for information technology products and since their output is extremely difficult to measure. Moreover measured productivity in these industries has been declining! But even if it is admittedly difficult to study the impact of IT on this sector’s output or if the industry-level productivity measures are plainly wrong, any productivity effects that are present would show up in aggregate measures, provided that the output of the downstream industries using the services is measurable in a reliable way. This is due to the fact that most of the output of these service industries is intermediate. To conclude, Jack Triplett provides a thorough assessment of the (possible) relationship between the productivity slowdown and measurement problems. He shows that there is reason to believe that solving a number of measurement problems would reduce the observed slowdown somewhat. However, Triplett argues that it will not let the phenomenon disappear. Of course these answers are not fully satisfactory, with the important service industries being the darkest region of the landscape. At this point we cannot fully rule out or confirm that a technology-driven effect is at work here, namely that our measurement techniques have not fully coped with the emergence of particular products and services for which output


51

is hard to measure. Jack Triplett is skeptical with respect to such an explanation, but there is no firm evidence against it at this point, either. We know very little, but the full scope of our ignorance becomes clear when one realizes that we know much more about the U.S. economy than about the situation in many other countries. Mismeasure of the German Economy? A brief comment on the case of Germany may therefore be a helpful complement to Jack Triplett’s analysis of the U.S. situation. Countries differ considerably by the speed at which the statistical machinery begins to experiment and learn new techniques. Some convergence can be observed, but as in other areas it is slow. Unlike the U.S. case, there are virtually no academic studies of product prices in Germany prior to 1997. Some experimentation with hedonic indexes did occur, even within the Federal Statistical Office which does not have a research branch, contrary to similar U.S. and French institutions. These experiments were not widely published, and thus it is virtually unknown to non-German and German economists that the German Federal Statistical Office had implemented a group that studied hedonic price indexes for personal computers by 1989 already (Gnoss and von Minding 1990). The authors of this study computed hedonic indexes for PCs in Germany and found them to show much larger decreases in the quality-corrected price index than the official index did. The conclusion of the study was to describe hedonic techniques as appropriate for computing quality corrections in price indexes. However, given the difficulty of obtaining data, the dubious legal state of data not collected by officials from the Statistical Office (an important consideration given the legal status of official statistics), and the problems in determining the correct specifications, it advised against their use. Although several academic studies followed suit (e.g., for PCs and for PC software), the issue of quality biases in price indexes was not a major topic till 1998 when a Bundesbank study by Hoffmann (1998) suggested that the German consumer price index displays similar distortions as the U.S. CPI. The initial lack of such a debate may look strange to U.S. observers. However, one should point out that few transfer or other payments are indexed in Germany. The use of indexing is actually restricted by a paragraph in the Currency Stabilization Law (§3.2 Wa¨hrungssicherungsgesetz) which requires all

52

Dietmar Harhoff

indexing clauses in contracts to be admitted by the Bundesbank. In 1997, 55,038 requests for the admission of such clauses were filed with the Bundesbank. Nonetheless, indexing is considerably less important in Germany than in the United States. The Bundesbank report caused some embarrassment for the Statistical Office, which till then had argued that the German consumer price index is merely a price index and not a cost of living index for which household utility is kept constant. Furthermore the Statistical Office had argued that there was no serious problem of quality biases. The debate also showed that statisticians in the Federal Statistical Office and in academe preferred not to introduce any economic concepts—like utility or welfare—into the debate, making measurement a clean exercise free from concepts like the ones mentioned. Following the publication of the Hoffmann report, a lively debate started. It culminated in a 1999 workshop held at the Bundesbank (Deutsche Bundesbank 1999). There is by now considerable evidence that inflation is overstated in Germany, just as it appears to be in many other countries. The studies at hand indicate that for PCs and software similar quality distortions arise as in other countries (Harhoff and Moch 1997). Hoffmann’s report contains an analysis for a number of household goods, such as washing machines, refrigerators, and freezers. The results typically resemble those of Gordon (1990), with minor exceptions not being discussed here. An interesting deviation from U.S. results is apparent for automobiles. It is surprising that by 1999, there had been no hedonic study of automobile prices in Germany, given the importance of the industry—used and new cars account for about 7 percent of total household expenditures in Germany. The results from an own study (Harhoff 1999) indicate that the differences between official and hedonic price indexes for automobiles in Germany is on the order of 0.8 percentage points. The figure may appear large, but a study by Blow and Crawford (1998) for automobiles in Britain yields very similar results. What is amazing here is the direction of the deviation—many U.S. studies have shown that the respective U.S. index rises much slower than hedonic indexes, while the reverse appears to hold in Germany. Presumably this is due to the extensive quality corrections that U.S. price statisticians apply to car prices on the basis of detailed cost information. Many European countries have by now instituted working groups or panels to study the implications and repercussions of the report


53

issued by the Boskin Commission. To name but one example, the British Statistical Office has embarked on a three-year research program on these issues (Baxter and Camus 1999). Again, Germany has been a laggard in this development. If the debate in Germany has shown anything, it is this in my view: statistical agencies must maintain a research division in order to keep up with advances in statistics and economics. A naive quest for measurement without theory is doomed to fail. The fact that German economists have largely abandoned this research area has allowed developments in the theoretical treatment of economic measurement to be neglected. There are encouraging signs now that there will be more consideration given to experiments in the realm of price measurement. Experiments are certainly not a sufficient condition for making progress in price measurement, but they are a necessary first step. As Franklin D. Roosevelt said, ‘‘above all, try something.’’ References Baxter, M., and D. Camus. 1999. Three year research programme on RPI methodology. Economics Trends 543: 25–29. Blow, L., and I. Crawford. 1998. A quality-constant price index for new cars in the UK, 1986 to 1995. IFS Working Paper W98/12. Institute for Fiscal Studies, London. Boskin, M., et al. 1996. Towards a more accurate measure of the cost of living. Final Report to the Senate Finance Committee from the Advisory Commission to Study the Consumer Price Index. Washington, DC. Deutsche Bundesbank. 1999. Zur Diskussion u¨ber den Verbraucherpreisindex als Inflationsindikator—Beitra¨ge zu einem Workshop in der Deutschen Bundesbank. Diskussionspapier 3/99. Volkswirtschaftliche Forschungsgruppe der Deutschen Bundesbank. Gnoss, R., and B. V. Minding. 1990. Neue Ansa¨tze zur Berechnung von Preisindizes. Ausgewa¨hlte Arbeitsunterlagen zur Bundesstatistik 13. Wiesbaden: Statistisches Bundesamt. Gordon, R. J. 1990. The Measurement of Durable Goods Prices. Chicago: Chicago University Press. Harhoff, D. 1999. A Hedonic Price Index for Automobiles in Germany. Unpublished discussion paper. Ludwig-Maximilians-Universita¨t Mu¨nchen. Harhoff, D., and D. Moch. 1997. Price indexes for PC database software and the value of code compatibility. Research Policy 26: 509–20. Hoffmann, J. 1998. Probleme der Inflationsmessung, Diskussionspapier 1/98 der Volkswirtschaftlichen Forschungsgruppe der Deutschen Bundesbank. Deutsche Bundesbank, Frankfurt a.M.


2

Information Technology, Organizational Transformation, and Business Performance Lorin M. Hitt and Erik Brynjolfsson

2.1

Introduction

How can computers contribute to business performance and economic growth? A natural place to start would seem to be by looking at what computers do best. From their inception, computers have provided very accurate computation. During World War II, the U.S. government generously funded research into tools for calculating the trajectories of artillery shells. The result was the development of some of the first digital computers with remarkable capabilities for calculation—the dawn of the computer age. The underlying technology of computers has advanced at a breathtaking rate since the Mark I (1939) and the ENIAC (1943),1 but even today most people think of tasks like rapidly multiplying two large numbers, or for that matter, millions of large numbers, when asked to identify the computer’s strength. However, most problems in the world are not numerical problems. Ballistics, code breaking, parts of accounting, and bits and pieces of other tasks involve lots of calculation. The everyday work that most managers, professionals, and information workers involves other types of thinking. If computers were only used for number crunching, their impact on the business and the economy would be fairly limited. Fortunately computers are not fundamentally number crunchers, they are symbol processors. We may say that digital computers can only understand everything in terms of ones and zeros, but that is only one way of interpreting their inner workings. During World A shorter version of this chapter appeared as ‘‘Beyond Computation: Information Technology, Organizational Transformation and Business Performance,’’ Journal of Economic Perspective (Fall 2000).

56

Lorin M. Hitt and Erik Brynjolfsson

War II, Alan Turing was laboring on a much more general conception of computers. He had the critical insight that the on–off states of computers could just as easily be interpreted in more general, nonnumeric ways, such as the true–false states of any formal system. He proved that a basic symbol-manipulating machine could in fact simulate any arbitrary formal system with just four types of operations. This insight opened up a broad new range of potential applications for computer processing, and the related technologies of data storage and communications have also advanced at exponential rates. Because the same basic technologies can be used to organize and manipulate any types information that can be digitized—numbers, text, video, music, speech, programs, and engineering drawings, to name a few—the array of applications addressed by information technology (IT) now extends well beyond computation.2 As a result IT is best described not as a traditional capital investment but as a general-purpose technology (Bresnahan and Trajtenberg 1995). General-purpose technologies have always been the source of a disproportionate share of economic growth because they not only contribute directly to greater productivity but also indirectly by enabling complementary innovations. Earlier generalpurpose technologies, such as the telegraph, the steam engine, and the electric motor, initiated a series of complementary innovations and eventually led to dramatic productivity improvements. Some of these innovations were purely technological, such as ‘‘wireless’’ telegraphy developed by Marconi. However, some of the most interesting and productive developments were organizational innovations. For example, a major effect of the telegraph was to make possible the formation of geographically dispersed enterprises (Milgrom and Roberts 1992). Similarly the electric motor permitted more flexibility in the placement of machinery in factories, thus enabling industrial engineers to improve manufacturing productivity dramatically by redesigning work flow (David 1990). The steam engine was at the root of a whole array of technical and organizational changes that helped ignite the first industrial revolution. Like earlier general-purpose technologies the capabilities of IT are widely applicable across a range of functions and industries. While early IT applications such as accounting and payroll were successful in lowering the cost of these functions, much larger effects of computers were realized as managers redesigned their operations to take advantage of low-cost information processing and communications

IT, Organizational Transformation, Business Performance

57

(Malone, Yates, and Benjamin 1987; Hammer 1990; Davenport and Short 1990). Companies are pursuing new ways of organizing work internally, new structures for interacting with customers and suppliers, and new types of products and services. These complementary innovations are often essential to the success of IT investments. One way in which computers can act as a general-purpose technology is their ability to lower the cost of intra- or interorganizational communication. Organizations and markets can be thought of as information processors (Galbraith 1977; Simon 1976; Wyner and Malone 1996; Sah and Stiglitz 1986). As advances in computer and communications technologies enable us to speed up more and more steps of the information processing chain, advances in organizational technologies are need to avoid the bottlenecks that emerge in systems that were tailored for the use of earlier technologies. In the information age, the one thing that is much less scarce is information. However, as Herbert Simon (1976) has noted, ‘‘information consumes attention and the information processing bottleneck is worst at the tops of hierarchies.’’ Due to the wide ranging effects of general purpose technologies, their benefits and costs can often be difficult to quantify. When computers enable Dell Computer, Walmart stores, or Capital One Financial to change the way they do business, much of the resulting innovation in quality and customer service goes uncounted in standard economy- or industry-level productivity calculations. Saving customer time, raising service quality levels, and bringing new products to market create first-order welfare gains to the economy, but these benefits are largely ignored in our standard measures of economic performance, a fact that is becoming increasingly understood (e.g., see Boskin et al. 1997; Bresnahan and Gordon 1997). To the extent that the key benefits of computerization are not well counted, studies that rely on traditional productivity data will systematically underestimate the benefits of IT. At the same time successful IT strategies almost inevitably involve simultaneous ‘‘investments’’ in organization change, innovative business strategies, and employee’s human capital, not to mention software development and data acquisition. Although these complementary investments create valuable assets for individual firms and the economy as a whole, they are largely intangible and invisible to accountants, statisticians, and researchers. If indeed these complementary factors are large (as we will argue), suggestions that com-

58


puters are too small a factor input to have any economic significance (e.g., see Sichel 1996) may need to be reconsidered. While there is no silver bullet for assessing the value of the intangible costs and benefits of computers, one strategy that is likely to help is conducting the analysis at a reasonably fine level of detail, such as the individual firm. Even when we can’t directly observe the contributions of quality and innovation, firms that create both tangible and intangible benefits for their customers are likely to be more successful in the marketplace, and that outcome is observable. Equally important, if organizational change and other intangible investments have a strong influence on value of IT investment, these effects are best observed by examining these investments at the level that they are being made: the firm. In this review we will examine a variety of empirical studies (including studies published or otherwise available through yearend 1998) to argue that IT improves productivity by enabling complementary organizational innovations. 3 We first consider case studies to describe the role of IT in transforming firms, supply chains, and customer relationships. We then explore large sample statistical evidence on the relationship between IT and productivity levels, productivity growth, and stock market valuation to show the general performance effects of IT, and to provide indirect evidence of potential organizational complements. Next we examine studies that provide direct measurement of organizational complements and their interrelationships with IT investments and firm performance. Finally we explore the implications these results for assessing the contribution of IT to overall economic growth. While there are strengths and weaknesses in all the individual studies we consider, collectively they paint a coherent picture: organizational innovations play a critical role in realizing value from IT and that IT and its complements play a major role in economic growth. 2.2

Case Examples

There are numerous examples of how companies have employed IT to change the way they conduct business either internally or externally. A common theme throughout these stories is that investment in IT is complementary to changes in other aspects of the organization. Complementarity exists when increasing the amount of one factor raises the marginal benefit of another factor. This would suggest that


59

complementary factors will tend to appear together (Milgrom and Roberts 1990; Holmstro¨m and Milgrom 1994) in ‘‘systems’’ or ‘‘clusters’’ rather than acting in isolation. A further implication is that firms that adopt the entire complementary system together will typically show the greatest performance. These arguments also have implications in understanding the resistance to or difficulty of implementing IT-enabled organizational change (e.g., see Clemons, Thatcher, and Row 1995). Even when a new system of practices is substantially more productive than an existing system, changing incrementally (one factor at a time) from one complementary system to another can often make the firm substantially worse off than retaining the old system, at least until the firm has completed the transition (Brynjolfsson, Renshaw, and vanAlstyne 1997). The idea that changes between complementary systems must be ‘‘all or nothing’’ is part of the logic that underlies many organizational reengineering efforts, hence the slogan ‘‘Don’t Automate, Obliterate’’ (Hammer 1990). It also suggests why there is a reportedly high failure rate of reengineering and other large scale IT implementations (Kemerer and Sosa 1991). Many of the organizational practices that have been successful over the last century and have persisted today are a direct result of the high cost of information processing. For example, hierarchical structures are known to be efficient at minimizing communications costs (Radner 1993; Radner and VanZandt 1992; Malone 1987). Producing simple, standardized products is the most efficient way to utilize scale-intensive and inflexible manufacturing technology. However, as the cost of automated information processing has fallen over 10,000-fold since the 1950s, it is unlikely that the work practices that emerged in the previous era also happen to be the same practices that best leverage the value of cheap information and flexible production. Indeed, Milgrom and Roberts (1990) construct a model in which firms’ transition from ‘‘mass production’’ to flexible, computer-enabled, ‘‘modern manufacturing’’ is driven by exogenous changes in the price of IT. More generally, a variety of industries have been transformed in fundamental ways by combining information technology with changes in strategy, firm structure, or work practices. In this section we discuss the case evidence on three aspects of this transformation: the transformation of the firm, the transformation of supplier relations, and the transformation of the customer relationship.

60


Transforming the Firm The need to match organizational structure to technology capabilities and the challenges of making the transition to an IT-intensive production process is concisely illustrated by a case study of MacroMed, a large medical products manufacturer (Brynjolfsson, Renshaw, and van Alstyne 1997). In a desire to provide greater product customization and variety, MacroMed made a large investment in computer-integrated manufacturing in the early 1990s. These investments also coincided with a management edict for sweeping changes in incentive systems (the elimination of piece rates), job responsibilities (workers gained authority for scheduling machines, and for dealing more closely with customers and suppliers), information flows (increased lateral communication and teamwork), ‘‘line rationalization’’ (optimizing work flows by an engineering team), and literally dozens of other organizational practices. Despite the recognition by senior management that a complete system of changes needed to be implemented and the use of a detailed implementation plan, the new system fell well short of their expectations for greater flexibility and responsiveness. Further investigation revealed that line workers still retained many elements of the old work practices. Typically this did not reflect any conscious effort to undermine the change effort. On the contrary, the workers understood which practices were important for success in the past and sought to apply this knowledge. For example, one worker explained that ‘‘the key to productivity is to avoid stopping the machine for product changeovers.’’ While this was a valuable heuristic with the old equipment, it negated the flexibility of the new machines and created large work-in-process inventories. Ironically the new equipment was so flexible that the workers succeeded in getting it to work much like the old machines! Eventually management concluded that the best approach was to introduce the new equipment in a ‘‘greenfield’’ site with a handpicked set of young, motivated employees who were relatively unencumbered by knowledge of the old practices. This approach succeeded, and while a number of ‘‘bugs’’ in the new work flow needed to be worked out, major performance improvements ultimately materialized.4 Other types of systems investments also show strong complementarities with some aspects of internal organization. For example, in


61

consulting firms the implementation of ‘‘knowledge management’’ systems, designed to capture and share expertise throughout the firm, has often met with substantial resistance until incentives and other organizational practices were changed to support the use of the system (Orlikowski 1992; McKinsey and Company 1994; KPMG 1998). In addition the cost of making these organizational changes can be substantial. For every dollar an organization spends on implementing enterprise resource planning software (ERP) such as SAP R/3, the firm typically spends $3 to $5 on consultants to integrate the system into the organization, and a similar amount on internal expenses such as retraining, designing new business processes, and managing the implementation.5 Changing Interactions with Suppliers Historically, because of a lack of efficient supply markets and difficulty in coordinating with external suppliers, large firms tended to produce many of their required inputs in-house. For instance, the growth of the Lucky-Goldstar chaebol began a result of their inability to find an adequate supplier of plastic caps for cosmetic cream. This set off a chain of diversification and vertical integration decisions that led them into plastics, oil refining, and semiconductors (Milgrom and Roberts 1992, p. 542). Recently many firms are finding that the benefits of integration are increasingly outweighed by the disadvantages of internal production such as an inability to realize economies of scale or specialization, agency costs arising from a lack of market discipline, and inflexibility leading to late adoption of new production processes (Williamson 1975; Chandler 1977). Even companies whose success was tied to vertical integration, such as General Motors, have since reversed course, divesting large internal suppliers. As one industry analyst stated, ‘‘What was once the greatest source of strength at General Motors—its strategy of making parts in-house—has become its greatest weakness’’ (Schnapp 1998). One of the earliest forms of IT-enabled interorganizational communication is electronic data interchange (EDI). Through a standard format, firms can place orders and receive confirmations from suppliers. This saves the cost of preparation and delivery of paper purchase requests and invoices and enables ordering systems to be tied directly to production systems eliminating costly, time-consuming,

62


and error-prone manual data entry (Johnston and Vitale 1988). Although the cost savings from adopting EDI can be large, even greater savings can be achieved when EDI is combined with other methods of supply chain optimization. A particularly successful early example of an interorganizational system is the Baxter ASAP system, which enables hospitals to order supplies directly from wholesalers (Vitale and Konsynski 1988; Short and Venkatraman 1992). Originally the system was envisioned as a way of reducing data entry cost for the 40,000þ purchase orders per year that Baxter’s field sales representatives completed by hand for each hospital they served. However, once ordering was computerized and data were available on levels of hospital stock, Baxter increasingly took responsibility for the entire supply operation—designing stock room space, setting up computer-based inventory systems, and providing entirely automated inventory replenishment. The combination of the technology and the new supply chain organization created substantial efficiency gains both for Baxter (no paper invoices, predictable order flow) and for the hospitals (elimination of stockroom management tasks, inventory optimization, and reduction of stockouts). The later versions of the ASAP system also included the facility to order from other suppliers, creating an electronic marketplace in hospital supplies. One of the most sophisticated and modern examples of computerbased supply chain integration is present in consumer-packaged goods. Traditionally sales of products such as soap and laundry detergent were heavily driven by off-invoice promotions, in which manufacturers offered discounts, rebates, or even cash payments to retailers to stock and sell their products.6 These promotions often created substantial production inefficiencies. Because many consumer products can be warehoused for considerable periods of time without deterioration, retailers tended to make massive purchases during promotional periods (a practice known as ‘‘forward buying’’) increasing volatility in manufacturing schedules and increasing costs (e.g., rush orders for inputs, inefficient production run length). In response to the problem of forward buying, manufacturers increased the speed of packaging changes (thus enabling older product to be easily identified) or developed internal audit departments that monitored retailers’ purchasing behavior to detect contractual abuse. Moreover manufacturers who relied on sales data for planning received a distorted picture of the final demand characteristics of their


63

markets, a problem exacerbated by retailers’ tendencies to guard consumer data closely to improve their bargaining position with manufacturers (see Clemons 1993 for a discussion of the problems that promotion-based selling creates for manufacturers). To eliminate these efficiencies, Procter and Gamble (P&G) along with their larger customers such as Walmart pioneered a program called efficient consumer response (ECR). Implementing ECR requires changes in a whole collection of work practices, information systems, and other aspects of the retailer-manufacturer relationship. These include direct transmission of retailers’ checkout scanner data to the manufacturer, fully automated ordering through EDI, automated payments and invoicing, elimination of off-invoice promotions, and a shift to ‘‘everyday low price’’ (EDLP) pricing, and continuous replenishment of products on a daily basis. Manufacturers also became increasingly involved in inventory decisions and began a move toward ‘‘category management,’’ where a lead manufacturer would take responsibility for an entire retail category (e.g., laundry products) determining stocking levels their own and other manufacturers products as well as complementary items. This collection of changes reduced substantially the operational costs for both manufacturers and retailers and also removed the incentive distortions that created many of the inefficiencies in the first place. Since there were large efficiency gains from this system, manufacturers could offer retailers substantially lower prices to induce them to share data. Computers were a critical component of this new form of supply chain organization. In particular, without the direct computer-computer links to scanner data and the electronic transfer of payments and invoices, there would be no way to have the levels of speed and accuracy needed to implement such a system. Computer-enabled supply chain redesign is now being actively pursued in a wide variety of markets (Fisher 1997). In addition there has recently been a much greater reliance on subcontracting, outsourcing and buyer-supplier partnerships at least partially enabled by the low-cost interchange of complex information enabled by IT. Changing Customer Relationships Even in seemingly mature industries with massive overcapacity, some companies have managed to use information, information technology, and organizational innovation to deliver greater value for

64


the customer while earning extraordinary profits. Capital One Financial Corporation, previously the credit card division of Signet Bank, was formed to exploit the fact that most credit card issuers offered inflexible standardized pricing to all of their customers, despite the problem that this created enormous variations in customer profitability (Clemons and Thatcher 1997). Low-cost customers were effectively being overcharged to subsidize unprofitable accounts. This situation went unrecognized for many years, primarily because the information systems infrastructure in place at most credit card issuers was directed at minimizing processing cost, not identifying the factors that affect profitability. Furthermore, even if the information could be obtained, little could be done with it; their work practices and supporting information systems were often too inflexible to offer customized terms to specific customers. Through a strategy called ‘‘test and learn,’’ Signet Bank offered a wide range of product designs (annual fees, security deposit, interest rate, rebates, etc.) to different subsets of potential customers identified through publicly available sources. Through these experiments, some of which proved to be fairly costly, they amassed an enormous database in the early 1990s that related observable customer characteristics, product choice, and long-term customer profitability. Using these data, they identified several highly profitable product designs that were then aggressively promoted to a suitable target group.7 In addition to having large databases and flexible systems that allow analysis of ongoing customer information, they also supported the use of this information with a skilled staff and organizational innovation that emphasizes high levels of employee discretion. They claimed to employ more Ph.D.’s in their firm than the rest of the credit card industry combined. In their telephone support area, skilled customer representatives were empowered to negotiate product features with customers, supported by data on the impact of changes on customer profitability, and incentives based on the future value of accounts they retain. Competitors who lacked the systems or the staff to implement these programs were at a severe disadvantage. In the four years following its spin-off from Signet Bank, the price of Capital One stock rose over fivefold. Meanwhile some large, previously successful card issuers such as AT&T have left the business, and other issuers who lack the ability to micro-segment the market are under substantial profit pressure.


65

In addition to IT innovations in operational systems leading to dramatic improvements in customer service, convenience or product variety, the Internet has opened up a new range of possibilities for enriching interactions with customers. Dell Computer, a mail-orderbased PC manufacturer, has had enormous success in attracting incremental customer orders and improving service by placing full configuration, ordering, and technical support capabilities on the World Wide Web (see www.dell.com). This change in the front end was coupled with systems and work practice changes that emphasize just-in-time inventory management, build-to-order production systems, and tight integration between sales and production planning. This transformation of the traditional build-to-stock model of selling computers through retail stores to a consumer driven buildto-order model has been estimated to give Dell as much as a 10 percent cost advantage over their rivals in production cost and erects a substantial barrier to entry to other firms that would like to ‘‘be direct.’’ Summary The case evidence suggests that successful IT investment is often coupled with redesign of internal organization, a shift in how firms interact with suppliers, and the pursuit of new sources of value for customers such as service, variety, or convenience. In all the cases discussed above, it is not merely the idea that IT lowers the cost of conducting business as usual through automation, nor a ‘‘silver bullet’’ solution where the technology alone creates the value, but the collection of organizational, strategy, and systems changes that lead to substantial benefits to th firm. 2.3

Results from Firm-Level Analyses

While the case evidence provides many examples of a strong link between the value of information technology and investments in complementary organizational practices, it is difficult to know how general these effects are. Do the cases we describe represent the idiosyncratic situations of a few ‘‘leading edge’’ firms, or are the lessons more widely applicable? To understand general trends, it is necessary to examine these effects across a wide range of firms and

66


industries. In this section we will explore the results from largesample statistical analysis. First, we examine studies that focused on the direct relationship between IT investment and business value to understand the general impact of computers on performance as well as to provide some indirect evidence of complementary factors. We then consider studies that directly measured these organizational factors and their correlation with IT use, as well as the few initial studies that have linked this relationship to productivity increases. IT and Productivity at the Firm Level Early studies of the relationship between IT and productivity or other measures of performance were generally unable to determine the value of IT conclusively. Loveman (1988) and Strassmann (1990), using different data and analytical methods, both found that the performance effects of computers were not statistically significant. Barua, Kriebel, and Muhkadopadhyay (1995), using the same data as Loveman, found evidence that IT improved some internal performance metrics such as inventory turnover but could not tie these benefits to improvements in overall firm productivity. Although these studies had a number of disadvantages (small samples, noisy data) that yielded imprecise measures of IT effects, this lack of evidence combined with equally equivocal macroeconomic analyses by Steven Roach (1987) implicitly formed the basis for the ‘‘productivity paradox.’’ Later studies, using larger data sets covering more recent time periods, found substantially different results. Using data for about 300 large firms over multiple years (1988–1992), Brynjolfsson and Hitt (1993, 1996) and Lichtenberg (1995) estimated production functions (F) that relate the amount of output that a firm produces to the inputs it consumes and variables identifying the firm (i), industry ( j), or time (t):8 Qit ¼ FðKit ; Lit ; Mit ; Cit ; Sit ; i; j; tÞ; where Q is a measure of gross output or value added, K is ordinary capital, L is non-IT labor, M is materials (when the dependent variable is gross output), C is computer capital, and S is IT labor. Assuming a standard form (Cobb-Douglas) for the production function FðÞ, and taking logarithms, we have the following estimating equation9


67

log Qit ¼ log Aði; j; tÞ þ b k log Kit þ bl log Lit þ bm log Mit Qit ¼ þ bc log Cit þ b s log Sit þ eit ; where AðÞ generally includes dummy variables for time, industry, and sometimes the firm. The coefficients of interest in these models are bc and b s , which represent the output elasticity10 of computer capital and IS labor respectively. Contrary to previous studies, Brynjolfsson and Hitt and Lichtenberg found both of these elasticities were positive and significant. Estimates of the elasticity of computer capital range from 0.01 to 0.04 by Brynjolfsson and Hitt (1993, 1996) to 0.10 by Lichtenberg (1995) when total output is used as the dependent variable. When compared to the factor share of computer capital, these output elasticities imply returns that sharply exceed the user cost of computer capital. Estimates of the contribution of IS labor range from slightly over a dollar to six dollars for every dollar invested. These results were corroborated by Dewan and Min (1997) and others (e.g., Rai, Patnayakuni, and Patnayakuni 1997) using similar data and methods. Several studies have also examined the returns to IT using data on the use of various technologies rather than the size of the investment. Greenan and Mairesse (1996) matched data on French firms and workers to uncover a relationship between a firm’s productivity and the fraction of its employees who report using a PC at work. Although they only observe a small subset of the employees of any given firm, they use econometric techniques to calculate the true estimate of the elasticity of computers by comparing the results of firms that have different numbers of reported employees in the data. Their estimates are remarkably similar to earlier estimates of the computer elasticity. Other studies have also been conducted at a more micro-level and corroborate these firm-level studies. These studies generally use measures of technology use, rather than investment quantity, and tend to focus on the effects of mechanization rather than the output contributions of general purpose computing. Nonetheless, the studies provide additional support for the idea that computers and computer-controlled equipment have a substantial impact on productivity. Kelley (1996) found that plants that use computer-controlled machinery are substantially more productive than other metalworking plants. Black and Lynch (1997) used data on the percentage of em-

68


ployees who use computers and relate this to plant productivity; they find that the contribution of computers is positive and significant. Doms, Dunne, and Troske (1997) utilize data on the adoption of different numbers of advanced manufacturing technologies (AMTs) and find that plants that use more AMTs (many of which are computer based) have higher productivity and wages. Other studies have shown similar results at even finer levels of detail; for example, computerization has been found to increase productivity in government activities such as toll collection and postal sorting (Muhkopadhyay, Rajiv, and Srinivasan 1997; Muhkopadhyay, Lerch, and Mangal 1997; see also Lichtenberg 1998 for a study of the productivity of IT across multiple government organizations). The recurring findings of a correlation between IT and productivity in recent studies contrasts with the findings of research in the 1980s and early 1990s. This can be attributed to a several factors. First, most of these studies used recent data where the real quantity of computers in use is often an order of magnitude larger than studies that focused on earlier time periods. This makes it easier to identify the contribution of computers and to distinguish it from other factors and random measurement error. Second, the sample size in these studies was often much larger (e.g., about 300 firms and over 1,000 observations for Brynjolfsson and Hitt 1996 compared to 43 business units and less than 200 observations for Loveman 1988) enabling much more precise measurement of the potential contribution of computers. In fact, as noted by Brynjolfsson (1993), the previous literature may be better characterized as a ‘‘shortfall of evidence rather than evidence of a shortfall’’ in computer productivity. Finally the trend toward using models derived from economic theory rather than ad hoc correlations with performance ratios can also explain the apparent ‘‘reversal’’ of findings in some cases (Hitt and Brynjolfsson 1996; Alpar and Kim 1991). The Role of Intangible Outputs One of the most serious problems in productivity measurement is accurately accounting for quality change. It is typically much easier to count the number of units produced by a firm than to assess their intrinsic quality. Harder still is knowing whether the product is delivered to the specific customer who values it most. As a result a significant fraction of value of quality improvements, greater timeli-


69

ness, customization, and customer service is overlooked (and implicitly treated as nonexistent) in official statistics. Brynjolfsson and Hitt (1996, 1998) argue that firm-level studies, unlike industry- or economy-level studies, may also be able to capture some intangible value (e.g., customer service, convenience, product variety) that is created by IT, even when such value cannot be directly measured. To illustrate, suppose that a firm invests in IT to improve product quality and that these benefits are recognized and valued by consumers. If other firms do not make similar investments, any difference in quality will lead to differences in the equilibrium prices for each firm. When high-quality and low-quality firms are combined together in industry data, both the IT investment and the difference in revenue will be averaged out. Unless statisticians account for the change in product quality, little net correlation of productivity with IT across different industries will be detected. However, when the analysis is conducted across firms, variation in quality will contribute to differences in firm-level output and thus will be measured as increases in the output elasticity of computers. This difference may also account for at least part of the divergence between firm-level productivity studies and studies conducted at a higher level of aggregation. For example, a number of studies using industry data on ‘‘high-tech’’ capital investments on productivity found little evidence of benefits utilizing industry-level data that ran through 1988 (Morrison and Berndt 1990; Berndt, Morrison, and Rosenblum 1992). While some of the differences between these analyses and the work at the firm-level (and below) may also be due to time period, similar results also appear in a subsequent analysis that runs through 1991 (Morrison 1997), suggesting that the differences may be more systemic. Zvi Griliches (1994) noted that much of the current evidence for a productivity ‘‘slowdown’’ is concentrated in industries where output is poorly measured. This point can be illustrated by selected industry-level productivity growth data over different time periods as listed in table 2.1. According to the figures in the table, a bank today is only about 80 percent as productive as a bank in 1977, a health care facility is only 70 percent as productive, and a lawyer is only 65 percent as productive as he or she was 1977. This suggests the following thought experiment: Would you prefer to have a 1977 bank, hospital, or

70


Table 2.1 Annual productivity growth for selected industries over different time periods Industry

1948–67

1967–77

1977–96

Depository institutions (e.g., banks)

0.03%

0.21%

1.19%

Health services

0.99%

0.04%

1.81%

Legal services

0.23%

2.01%

2.13%

Source: Partial reproduction from Gordon 1998, table 3.

lawyer? In 1977 all banking was conducted at the teller window at branch locations during regular ‘‘bankers’ hours.’’ In 1996 only 55 percent of banking transactions occured face to face, and customers now have access to a network of 139,000 ATMs11 (Osterberg and Sterk 1997) which provide 24-hour service, 7 days a week in a wide range of locations. Computer-controlled medical equipment has enabled more successful and much less invasive medical treatment, many procedures that previously required extensive hospital stays can now be performed on an outpatient basis, and medical tests that involved surgical procedures can now be done using noninvasive imaging devices such as MRI or CT scanners. A lawyer today has access to a much wider range of information through online databases (e.g., Lexis/Nexis) and an increased ability to manage large numbers of legal documents through image processing and document retrieval systems. In addition some basic legal services, such as the design of a simple will, can now be performed by consumers using standard software packages. Nolo press has sold over 500,000 copies of WillMaker, a PC-based will writing tool. All of these examples suggest that there may be substantial value, at least part of which is attributable to computers, that has gone uncounted in industry-level and economy-level productivity statistics. Interestingly Siegel (1997) found that the measured effect of computers on productivity was substantially increased when he used econometric techniques designed to address various types of measurement error in industry-level data. One of the most important types of unmeasured benefits arise from new goods. New goods provide a first-order contribution to consumer welfare, yet statistical agencies do not incorporate them into price indexes until many years after their introduction. For example, the VCR was not incorporated into the consumer price index


71

until 1987, about a decade after VCRs began selling in volume. While this methodology may correctly capture product improvements and price declines for the VCR after 1987, the contributions to consumer welfare and productivity growth before then were missed. This methodological problem may understate the value of computers in two ways. First, the computer industry itself has one of the fastest rates of new product introduction in history. While it may be possible to incorporate improvements in processors or memory into official output statistics, new applications such as personal digital assistants and voice recognition systems will go unmeasured.12 Second, new goods like computers enable even more new goods to be developed, produced, and managed. For instance, the number of new product introductions in supermarkets has grown from 1,281 in 1964 to 1,831 in 1975 and then to 16,790 in 1992 (Nakamura 1997); the data management requirements for managing this large number of products would have overwhelmed the computerless supermarket of two decades ago. Bar code scanners, inventory systems, and computer-based supply chain integration enable this new way of doing business. A series of surveys conducted by Brynjolfsson and Hitt (1997) also provide evidence that the production of intangible outputs is an important consideration for IT investments. In multiple surveys of IS managers, conducted in 1993, 1995, and 1996, they found that customer service and sometimes other aspects of intangible output (specifically quality, convenience, and timeliness) ranked higher than cost savings as the motivation for IS investments (see figure 2.1). Moreover firms that place greater emphasis on these intangible outputs have higher productivity and profitability, while firms that emphasize cost cutting have, at best, average productivity. Brooke (1992) also found that IT was associated with increases in product variety using measures of trademark filings. This collection of results suggests that IT may be associated with increases in the intangible component of output. Because these intangibles are poorly measured, this can lead to systematic underestimates of the contributions of computers. Unfortunately, even firm-level data will not fully capture the value quality improvements or other intangible benefits that are commonplace across an industry (e.g., the convenience value of ATMs is not limited to a single bank) because there will be no source of interfirm variation.

72


Figure 2.1 Reasons given by managers for investments in information technology. Source of data: Brynjolfsson and Hitt (1996).

Excess Returns? In addition to finding a positive and significant effect of computers, many of these firm-level productivity studies also showed that the net benefits of computers, after capital costs were subtracted, were substantially greater than zero in both economic and statistical significance. Because economic theory would predict that the net benefits of any investment, including IT, should be zero in equilibrium (absent market imperfections or barriers preventing the adoption or diffusion of IT), the evidence of ‘‘excess returns’’ is somewhat more perplexing. One possible explanation is that these high returns reflect reverse causality: firms with unexpectedly high sales quickly spend some of their windfall on computers, thus leading to the illusion that IT is causing increased output. However, across multiple studies, attempts to correct for this bias using various techniques, such as including only the predetermined components of inputs13 which further increases the estimated IT coefficients (e.g., see Brynjolfsson and Hitt 1996, 1998). Thus, if reverse causality is a problem, it may be better explained by IT investment being less procyclical than other types of investments. That is, firms with an unexpected increase in free cash flow invest in other factors, such as labor, before they change their IT spending. Alternatively, the higher coefficients in instrumental vari-


73

ables regressions could be attributed to errors in measurement (which would tend create downward bias; see Brynjolfsson and Hitt 1997, 1998, for a discussion of this issue) but does not explain why there may be ‘‘excess returns’’ implied by coefficient estimates from a variety of regression approaches. As noted by Brynjolfsson and Hitt (1996), an alternative explanation is that the output elasticities for IT are about right but that we are underestimating the input quantities. This would lead to an overestimation of the rate of return for IT. If computers are associated with unmeasured complementary investments, then estimates of the benefits of IT may really represent the benefits of an overall system of IT and other complementary factors. However, when these benefits are compared to the measured costs, which include only the IT capital (and in some analyses IT labor costs), they will appear to be too large, since the substantial costs of these unmeasured complements are omitted. Thus the presence of unmeasured, complementary organizational investments may explain a substantial portion of the excess returns (a more formal analysis of this issue appears in Hitt 1996 and Brynjolfsson and Hitt 1998). The Role of Differences among Firms in Understanding IT Value The studies cited previously focused entirely on the average return across a wide variety of potentially heterogeneous firms. However, even when the regressions control for industry and year, this average may mask substantial variation across firms in the productivity, IT investment, and adoption of other organizational practices. This is most clearly illustrated in figure 2.1, which plots firm-level IT investment against productivity (reproduced from Brynjolfsson and Hitt 1998). While the upward-sloping regression line through these points indicates that greater IT investment is associated with higher productivity,14 there is also substantial variation around this line. Some firms spend heavily on IT but have very low productivity, while others with similar IT investments are very successful. To probe this variation across firms, Brynjolfsson and Hitt (1995) estimated a firm effects specification. This regression method can be interpreted as dividing firm-level IT benefits into two parts: a part common to all firms due to variation in investment over time, and a part due to specific firm characteristics. An added advantage of this specification is that to the extent that sources of reverse causality are

74


constant over time (e.g., management skill), they will not influence the results (Griliches 1984 first made this argument in the context of R&D productivity). In this specification, Brynjolfsson and Hitt found that, the coefficient on IT was reduced by about 50 percent as compared to the results of an ordinary least squares regression, while the coefficients on the other factors (capital and labor) were only slightly changed. One interpretation of this difference is that it reflects the extent to which unmeasured and slowly changing organizational practices affect the returns to IT investment. Using different data, which included more firms (600þ) and a longer time series (1987–1994), Brynjolfsson and Hitt (1998) examined the effects of IT on productivity growth rather than productivity levels, which had been the emphasis of most previous work. Their most striking result is that the effects of IT are substantially larger when measured over longer time periods. When one-year differences in IT are compared to one-year differences in firm productivity, the measured benefits of computers are approximately equal to their measured costs. However, in a variety of econometric specifications, the measured benefits rise by a factor of 2 to 8 as longer time periods are considered. While part of this effect could be explained by measurement error (which tends to average out somewhat over longer time periods), the size of the change is too large to be attributed solely to specification problems. One interpretation of these results is that the benefits of IT are increased over time because organizational complements are not always present at exactly the moment that IT investments come on line. The short-term returns represent the direct effects of IT investment, while the longer-term returns represent the effects of IT when combined with related investments in organizational change. Further analysis by Brynjolfsson and Hitt (1998) based on earlier results by Schankermann (1986) suggested that if these omitted factors were erroneously included in ordinary capital and labor, there would be little impact on the results. In particular, simulation results indicated that a failure to fully account for the productivity benefits of IS labor could not explain the high measured returns to IT capital. This corroborates earlier work which found that the returns to IT capital were largely unaffected when an additional term for IS labor was explicitly included in a regression (Brynjolfsson and Hitt 1996). To explain the ‘‘excess returns,’’ the complements to IT capital must therefore be productivity enhancing ‘‘assets’’ that do not appear on a


75

firm balance sheet or on the expense side of the income statement. Natural candidates include software, data, training, and new organizational processes. In each case the expenses associated with the creation of these ‘‘assets’’ are typically written off in one year, although they continue to provide value for multiple years. Another perspective on these results can be obtained by looking at alternative metrics of firm performance such as stock market valuation. Brynjolfsson and Yang (1997) developed a model based on the literature on empirical measurement of Tobin’s q. They relate the stock market value of the firm to the various capital assets it owns. The basic assumption underlying their analysis is that in equilibrium the market value of a firm, as measured by the financial markets, should be equal to the sum of the values of its fixed assets (e.g., see the asset revelation theorem of Hall 1999). Their analysis suggests that while a dollar of installed ordinary capital on the firm’s balance sheet is valued at approximately a dollar, a dollar of installed IT capital appears to be correlated with about $10 of additional stock market value for the firm. An explanation for these results is that IT capital is disproportionately associated with other intangible assets which affect firm valuation but do not appear on the balance sheet as a capital asset.15 For example, the cost of developing new software, populating a database, implementing a new business process, acquiring a more highly skilled staff, or undergoing a major organizational transformation represent investments that have long-term benefits, yet they all go uncounted in firm financial statements. In this interpretation, for every dollar of IT investment, the firm also has on the order of $9 of additional intangible assets. A related explanation is that substantial ‘‘adjustment costs’’ must be incurred before IT is effective. These adjustment costs drive a wedge between the value of a computer sitting on the loading dock and one that is fully integrated into the organization. When these studies are considered together, even though complementary organizational factors are not always measured directly, we are able to understand a number of the properties that they must have. They are large in the sense of potentially being several multiples of the measured IT investment. They are unmeasured in the sense that they do not appear as a capital asset or as another component of firm input, although they do appear to be unique characteristics of particular firms (as opposed to industry effects). Finally

76


their effect is greater in the long term than the short term, suggesting that multiple years of adaptation and investment are required before their influence is maximized. Direct Measurement of the Interrelationship between IT and Organization If indeed there exist other types of organizational investments that are complementary to IT investment, then this implies two possible observable conditions. First, these factors are likely to be correlated with increased IT investment in a sample of firms. Second, firms that combine complementary factors (e.g., IT and certain organizational practices) should have higher performance than firms that adopt one but not the other (see Holmstrom and Milgrom 1994 and Athey and Stern 1997 for a discussion of the empirical assessment of complementarities relationships). Measuring these effects in practices is only possible if there is some source of exogenous variation due to external factors or firm mistakes. If there are no barriers to adopting the optimal level of all factors immediately and all firms face the same cost and benefit drivers of IT investment and organizational practices, then there will be no data variation: all firms will immediately adopt the same, optimal set of practices. If either of these conditions is relaxed, then different firms will adopt different levels of the various practices and may show differences in productivity. This could arise if some managers do not fully understand all the complementarities, if some firms make mistakes in their adoption levels, if some firms face adjustment costs or other barriers preventing them from optimizing, or if work practice adoption is partially random, due to chance or conscious experimentation. In these cases we may be able to observe both correlations between the use of various complementary factors of production and productivity differences between firms that adopt the complementary system as a whole and those that adopt practices piecemeal (e.g., IT investment without organizational redesign, and vice versa). However, without a full structural model specifying the production relationships as well as the demand drivers for each factor, statistical methods alone cannot prove that these practices are complements (see more discussion of this issue in Athey and Stern 1997) when there is not a clear alternative hypothesis. However, description and


77

refutation of possible alternative explanations can often leave complementarities as the most plausible explanation for observed relationships between IT and organizational factors (see Bresnahan, Brynjolfsson, and Hitt 1998 for an example in the context of investigating IT-skill complementarities). Correlational Results Hitt and Brynjolfsson (1997) surveyed approximately 400 large firms to obtain information on several aspects of internal organizational structure: allocation of decision rights, workforce composition, and investments in human capital. Across a wide variety of measures of information technology they find that greater levels of IT are associated with increased delegation of authority to individuals and teams, greater levels of skill and education in the workforce, and an increased emphasis on pre-employment screening for education and training. In addition they find that these individual work practices are themselves correlated with each other, suggesting that they do not act independently but are part of an overall complementary work system. Adopting a more structured demand framework, Brynjolfsson and Hitt (1998) found that firms that adopt this work system have not only higher demand for IT but also appear to adopt IT at a greater rate over time. Overall these results show a consistent pattern that IT investment is greater in organizations that are decentralized and have a greater demand for human capital in the workforce. Some other studies have also considered the relationship between work organization and IT use. Kelley (1994) finds that programmable automation is correlated with several aspects of human resource practices. Although the interaction of organizational practices and IT was not often the focus, evidence in many of the IT-productivity studies conducted at a micro-level (e.g., see Muhkadopadhyay et al. 1995; Black and Lynch 1996) suggests that the influence of IT on productivity is affected by various measures of worker characteristics or work organization. The human capital results are also corroborated by industry-level studies on the demand for skills and education. Berndt, Morrison, and Rosenblum (1992) found that high IT firms had a greater demand for education and skilled workers in data on two-digit manufacturing industries. Berman, Bound, and Griliches (1994) discovered that computer-using industries tended to have a greater demand for skilled workers. Autor, Katz, and Krueger (1997) found that high

78


IT industries had faster growth in the share in the proportion of college-educated workers they employ than other industries. Similar results were found by Bresnahan, Brynjolfsson, and Hitt (2002) at the firm level when they examined the relationship between IT use and demand for skills; firms that had high levels of IT use also tended to investment more in training and screening for education. Again, this is consistent with the idea that increasing use of computers is associated with a greater demand for human capital. Several studies have also considered the effect of IT on macroorganizational structure such as firm size, vertical integration, and diversification. Brynjolfsson et al. (1994) found that increases in the level of IT capital in an economic sector (e.g., durable manufacturing) were associated with a decline in average firm size. In addition these effects were most pronounced after a two-year lag. They argue that this is consistent with IT leading to a reduction in vertical integration. Hitt (1999) analyzed firm-level data and examined the relationship between IT capital stock and direct measures of firm’s participation in multiple industries: vertical integration and diversification. He found that IT was strongly correlated with decreased vertical integration, as well as a small correlation with increased diversification. These corroborate earlier case analyses and theoretical arguments (Malone, Yates, and Benjamin 1987; Gurbaxani and Whang 1991; Clemons and Row 1992; Clemons, Reddi, and Row 1993) that suggested that IT would be associated with a decrease in vertical integration and a shift toward market governance as a result of the lower costs of coordinating externally with suppliers engendered by IT. Altogether, these studies are consistent with the idea that IT is complementary to certain types of organizational characteristics; firms that use more computers (broadly defined) tend to use increasing amounts of delegated decision making to individuals and teams, have higher-skilled staff, invest more in training and education, and adopt corporate structures that involve less vertical integration. However, while this is consistent with complementarities, it may also be due to other effects. For example, suppose that some firms have a tendency to overspend on both organizational or technological innovation: they adopt new ideas without a thorough economic justification. This could lead IT and ‘‘modern’’ work organization to be correlated even it they were not complements. To distinguish spurious correlations from combinations of factors motivated by


79

economic concerns, it is useful to perform productivity analyses. If combining IT with these types of organizations is economically justified, then firms that adopt these practices as a system should outperform those that fail to combine IT investment with appropriate organizational structures. Productivity Effects of Combining IT and Organizations Initial results suggest firms that adopt the new work system have a higher contribution of IT to productivity (Brynjolfsson and Hitt 1998). Firms that are the most decentralized (top quartile on the measure of decentralization) have a 20 percent greater investment in IT and a 25 percent greater IT elasticity that the average firm. As an alternative way of viewing the data, they show that firms that are in the top half of both IT investment and decentralization are on average 5 percent more productive than firms that adopt only IT or organization alone. Similar results also appear when performance is measured as stock market valuation. As hypothesized earlier, organizational decentralization does behave as an intangible asset: firms that are in the top third of decentralization have a 6 percent higher market value after controlling for the levels of measured assets. Moreover a dollar of IT capital is valued between $2 and $5 more in decentralized firms than decentralized firms (Brynjolfsson, Hitt, and Yang 1998). This supports two elements of our previous arguments: first, that organization is indeed an asset with a positive value (or at least a positive cost), and second, that these types of organizations increase the value of IT investment. While these results are preliminary, they are relatively consistent with the collection of previous case and empirical work. These types of organizational practices may represent at least one component of an overall system of IT-enabled work practices, although there are undoubtedly others yet to be identified in future work. 2.4

Implications for the Macro Economy

While there is increasing evidence that IT, at least over recent time periods, has created substantial value for the firms that have invested in it, it can be a challenge to link these benefits to overall economic performance. One cannot necessarily add up value of IT to each of the firms in the economy to get the total value for the whole country. The true value may be greater or less than this sum because

80


of the private value of an investment does not necessarily equal its social value. More important, the traditional growth accounting techniques focus on the (relatively) observable aspects of investment, such as the price and quantity of computer hardware in the economy, and neglect the much larger intangible investments in developing complementary new products, services, markets, business processes, and worker skills. Nonetheless, standard growth accounting techniques have been applied to this question by several researchers and provide a useful benchmark for the contribution of IT to economic growth. A good place to start is to note that the nominal value of purchases of IT hardware in the United States in 1997 was about 1.4 percent of GDP. Since the quality-adjusted prices of computers decline yearly by about 30 percent, this means that we could now spend less than 1.1 percent of GDP and buy the same amount of computers as we did last year. This translates into a direct increase in productivity of over 0.3 percent. (The fact that Americans collectively choose to spend more on computers each year does not negate this fact. It merely means that they have found new uses for these ever-cheaper computers and that the actual contribution is somewhat higher.) This value is very close to the 0.3 percent contribution of computers in 1987 that Brynjolfsson (1996) found when he estimated of computer demand more precisely.16 One can also look at the contributions of IT to total output growth. Growth accounting calculations by Jorgenson and Stiroh (1995) suggest that computers contributed an average of 0.38 percent to output growth from 1985 to 1992, while Oliner and Sichel’s estimates range from 0.16 percent to 0.38 percent for the 1970 to 1992 period, depending on which assumptions they use. How do we interpret these numbers? In recent years total measured productivity growth in the United States has been about 1 percent, which is down from over 2 percent in the 1948 to 1967 period. This slowdown is correlated with increased computerization of the economy. However, correlation is not causality. The analysis above suggests that the implied contribution of computers has grown from close to zero to over 0.3 percent. This is actually relatively large compared to the contribution of all other types of capital equipment. Without computers it appears that productivity growth would have been even lower, presumably due to non-computerrelated factors.


81

While this analysis is intriguing, the productivity contribution of computers suffers from at least three shortcomings. First, it presumes that the buyers of computers are ‘‘getting their money’s worth,’’ since it is based on the quantities purchased and not on actual measures of value created. Most economists are comfortable with this assumption. Who better to judge if a computer is worth at least its price than the consumers and businesses who buy them? While some will be disappointed, others will be delighted beyond their expectations. On balance there seems to be no basis for presuming that buyers have been systematically wrong about their own preferences year after year for what has now been several decades. Nonetheless, it should be stressed that such analyses cannot, by definition, determine whether consumers or firms are getting a ‘‘good deal’’ on their current computer investments. Second, and more troubling, purchasing decisions are made based on the expected private benefits, and not the social benefits. If private returns exceed the social returns, then the approach above will overestimate the contribution of computers. Conversely, if private returns are less than social returns, then the true contribution will be underestimated. Either situation is possible. On one hand, various types of market imperfections may allow firms to invest in IT to redistribute rents from their customers, suppliers, or competitors without improving total welfare (Hitt and Brynjolfsson 1996). IT can be used to price discriminate (charging different prices to different customers), thereby capturing rents from consumers, or in enabling price shopping among vendors, thereby capturing rents from suppliers. For instance, at least some of the value created by Capital One in credit cards results from customers shifting from one credit card provider to another, and a major goal of ECR programs and vendor managed inventory is to discourage purchase of competitors products and reduce the incentives for the development of lower-cost private label products. What’s more, IT can also build switching costs, foreclosing competition and enabling price increases. Many firms that have heavily invested in vendor-specific interorganizational systems find themselves ‘‘locked-in’’ to specific suppliers, a situation that is often not resolved until a the systems become obsolete or are supplanted by a substantially more productive technology. On the other hand, firms often fail to capture the full value created by their investments. Some of the surplus may spill over to cus-

82


tomers, especially in industries where competitors can rapidly imitate innovators. Some may also spill over to other firms in the supply chain or firms that make complementary products. For instance, when banks use IT to lower the costs of transaction processing, customers and firms in dozens of industries benefit, even if they make no direct investments themselves. Historically only a small fraction of the benefits from innovations have been captured by the innovator and the same may hold true for IT-related business innovations. As mentioned earlier, if IT is a general-purpose technology like electricity or the internal combustion engine, the spillover benefits to the economy may be far larger than the direct benefits to individual investors. The computer-enabled revolution in the organization of manufacturing and retailing, the spread of ATMs in banking, and the eagerness of firms to copy ‘‘best practices’’ from each other is supportive of this view. This brings us to the third, and most serious, problem with applying the standard growth accounting approach to computers: it ignores that main argument of this paper that IT is only a small fraction of a much larger system. As discussed in the previous sections, very large investments in human capital, ‘‘organizational capital’’ and ‘‘strategic capital,’’ not to mention software, data acquisition, communication equipment, and peripherals, are typically required to make computerization successful for a firm. Thus it is a mistake to focus entirely on the computer hardware portion of the investment and not consider the large changes in complementary factors associated with computers. To the extent that an innovation is a critical enabler of other changes, it will have a large impact on output and productivity, even if its factor share is small. An extreme example of this is the power of ideas: the share of GDP investment devoted to, say, codifying the principles of mass production is quite small, but that would be a poor way to estimate the contribution of mass production to productivity growth.17 One could get a better estimate by considering all the complementary investments, such as new factory equipment and structures, but even that would be an underestimate because these new investments are vastly more productive in their new uses than in their old uses. Many technologies are valuable precisely because they embody ideas. If we only measure the ‘‘physical’’ aspects of them, we are sure to underestimate their role in growth. The motion picture camera,


83

while a relatively small share of GDP by itself, directly led to the creation of the movie industry; the telegraph and the internal combustion engine led to new ways of organizing firms and society as a whole. All of these innovations contributed much more to economic welfare than would be suggested by multiplying their share of GDP by their annual price declines.18 Returning to our ATM example, the physical capital stock of ATMs is only about $10B as compared to over $70B for retail bank branches,19 yet they as much as triple the hours of bank availability and increase the number of banking locations by over 150 percent. Like these earlier general purpose technologies, IT is an extremely general and flexible machine. It has been an important carrier of new ideas to a remarkably diverse constellation of industries and applications. A good example is the new method of retailing pioneered by Amazon.com for books. IT in general, and the Internet in particular, are obviously essential enablers of this model. However, the dollar value of the IT used by Amazon is dwarfed by it’s market value, and Amazon’s market value may well underestimates its impact on economic welfare, since many of the benefits of greater convenience and product selection accrue to consumers. Furthermore some of the benefits also accrue to imitators of Amazon’s business model. The business impact of computers on firms such as Dell, Walmart, and Capital One also stem mainly from the way they have reorganized their strategies and structures. It would be unrealistic to quantify the economic contribution by simply multiplying their computer hardware stock by some ‘‘normal’’ rate of return. Without more detailed data on the nature of the changes enabled by IT, it is impossible to estimate the overall impact on the economy. However, the central lesson that emerges from our analysis of the firm-level studies is that these complementary changes are not small. Any realistic attempt to estimate the overall contribution of IT cannot begin by assuming the value of complements is zero merely because they are difficult to measure. 2.5

Summary and Conclusion

After two decades of research into IT value, we have substantially improved our understanding of the relationship between information technology and economic performance. The shift to using firm-

84


Figure 2.2 Productivity versus IT stock of large firms. Reproduced from Brynjolfsson and Hitt (1998).

level data for studying the effects of IT has enabled researchers to identify both the outputs (e.g., intangible benefits) as well as the relevant inputs (e.g., complementary organizational investments) of an information and information technology–based economy. Overall the data suggest that there is no evidence of a productivity paradox, at least at the firm level. In addition, in examining a combination of case evidence and statistical studies, we find that there is a strong indication that organizational complements have a major effect on the contribution of IT. These complementary investments may be as much as an order of magnitude larger than the investments in the technology, suggesting that the real contribution of computers to the economy may be much larger than has been previously believed. The recurring message in the research is that in most cases where IT adds real value, it is only a small part of a much larger system of innovations. Measuring all the components of such systems is never easy, but if researchers and managers at least understand the importance of the intangible costs and benefits of computers, an assessment of their magnitude need not be beyond computation.


85

Notes 1. The Mark I was the first modern computer; the ENIAC was the first electronic computer that had no moving parts. 2. The French word for computers, ordinateur, comes closer capturing to this broader set of uses. 3. For a more general treatment of the literature on IT value see reviews by Attewell and Rule (1984), Brynjolfsson (1993), Wilson (1995), and Brynjolfsson and Yang (1996). 4. The transition at Macromed is documented in Brynjolfsson, Renshaw, and VanAlstyne (1997). 5. Representatives from EDS, Siebel Systems, and SAP have corroborated these figures. 6. This is above and beyond the costs of direct promotions to consumers through couponing and advertising. 7. For example, they created a balance transfer card that offered lower interest rates to customers who transfer a large existing credit balance. This product accomplishes a form of second-degree price discrimination to identify high-profitability ‘‘revolvers,’’ customers who borrow large sums on credit cards but pay back slowly. The lower interest rate is not attractive to customers who do not intend ever to pay since some effort is required to transfer a balance. The lower rate is also not attractive to ‘‘transactors’’ who pay off their balances in full every month and tend to also be unprofitable. Thus the lower rate for transfers attracts customers with relatively low operations cost but very high interest-accruing balances—the most profitable segment of the business. Other product designs opened up the opportunity to serve new customer segments or increase utilization of credit cards by underserved groups. For example, school teachers were offered the ability to defer payments during the summer when they are not drawing a salary. Capital One even offers a secured card program that allows a credit card to be obtained by high credit risk consumers, even those with previous bankruptcies. 8. A similar model was used by Loveman (1986, 1994). 9. In general, using different functional forms, such as the transcendental logarithmic (translog) production function, has little effect on the measurement of output elasticities. The standard method of estimation is to include dummy variables for time and industry as indicators of multifactor productivity (A). 10. The output elasticity of an input is roughly the percentage increase in output for a 1 percent change in that input, holding other factors constant. If a input has a ‘‘normal’’ rate of return, then the output elasticity should equal its share of total inputs. 11. This is in addition to approximately 85,000 bank branches in 1996, including banks, thrift offices, and bank offices in supermarkets. 12. For instance, Greenstein (1997) estimates the consumer surplus from extending the capabilities of mainframes and supercomputers exceeds $13B per year. 13. This is done via the technique of instrumental variables. 14. In figure 2.1 productivity is defined as MFP ¼ log O b c logðKÞ bL logðLÞ Time and industry controls

86


where the coefficients on the factors ðbÞ and the time and industry controls are estimated by regression. 15. Note that the simple explanation that IT capital is simply worth more than other types of capital would be inconsistent with economic rationality, since computer hardware can be freely purchased. 16. A related approach is to look at the overall effect of IT on the GDP deflator. Gordon (1998) calculates that ‘‘computer hardware is currently contributing to a reduction of U.S. inflation at an annual rate of almost 0.5% per year, and this number would climb toward one percent per year if a broader definition of IT, including telecommunications equipment, were used.’’ Since Gordon is referring to the GDP deflator, this means that real output is being increased by about 0.5 percent annually by computers, for the same amount of nominal spending. This implies that computers are adding on the order of 0.5 percent to productivity. 17. However, there are some ideas with sufficient intellectual property protection that it is possible to fully capitalize them. 18. Unless, of course, one includes the ‘‘virtual price’’ that consumers would have been willing to pay for the very first unit of these goods, as Hicks (1940) suggests. 19. This number is based on an average branch capital cost of $1m, as reported in Osterberg and Sterk (1997). The ATM figure is based on $75,000 capital cost per ATM.

References Alpar, P., and M. Kim. 1991. A microeconomic approach to the measurement of information technology value. Journal of Management Information Systems 7(2): 55–69. Athey, S., and S. Stern. 1997. An empirical framework for testing theories about complementarities in organizational design. Mimeo. MIT. Attewell, P., and J. Rule. 1984. Computing and organizations: What we know and what we don’t know. Communications of the ACM 27(12): 1184–92. Autor, D., L. Katz, and A. Kreuger. 1997. Computing inequality: Have computers changed the labor market? Mimeo. Harvard University. Barua, A., C. H. Kriebel, and T. Mukhopadhyay. 1995. Information technologies and business value: An analytic and empirical investigation. Information Systems Research 6(1): 3–23. Bender, D. H. 1986. Financial impact of information processing. 3(2): 22–32. Berman, E., J. Bound, and Z. Griliches. 1994. Changes in the demand for skilled labor within U.S. manufacturing industries. Quarterly Journal of Economics 109 (May): 367– 98. Berndt, E. R., and C. J. Morrison. 1995. High-tech capital formation and economic performance in U.S. manufacturing industries: An exploratory analysis. Journal of Econometrics 65: 9–43. Berndt, E. R., C. J. Morrison, and L. S. Rosenblum. 1992. High-tech capital, economic performance and labor composition in U.S. manufacturing industries: An exploratory analysis. MIT Working Paper 3414EFA.


87

Black, S. E., and L. M. Lynch. 2001. How to compete: The impact of workplace practices and information technology on productivity. Review of Economics and Statistics 83 (August). Boskin, M., J. Dulberger, E. R. Gordon, R. J., Z. Griliches, and D. Jorgenson. 1997. The CPI commission: Findings and recommendations. American Economic Review 87(2): 78– 83. Bresnahan, T., and R. Gordon. 1997. The Economics of New Goods. Chicago: University of Chicago Press. Bresnahan, T. F., and M. Trajtenberg. 1995. General purpose technologies: ‘‘Engines of growth’’? Journal of Econometrics 65: 83–108. Bresnahan, T. F., E. Brynjolfsson, and L. M. Hitt. 2002. Information technology, workplace organization and the demand for skilled labor: A firm-level evidence. Quarterly Journal of Economics 117: 339–76. Bresnahan, T. F. 1986. Measuring spillovers from technical advance: Mainframe computers in financial services. American Economic Review 76(4). Brooke, G. M. 1992. The economics of information technology: Explaining the productivity paradox. MIT Sloan School of Management Center for Information Systems Research Working Paper 238. April. Brynjolfsson, E. 1993. The productivity paradox of information technology. Communications of the ACM 35(12): 66–77. Brynjolfsson, E. 1996. The contribution of information technology to consumer welfare. Information Systems Research 7(3): 281–300. Brynjolfsson, E., and L. Hitt. 1997. Breaking boundaries. Informationweek (September 22): 54–61. Brynjolfsson, E., and L. Hitt. 1993. Is information technology spending productive? New evidence and new results. Proceedings of the 14th Annual International Conference on Information Systems. Orlando, FL. December. Brynjolfsson, E., and L. Hitt. 1995. Information technology as a factor of production: The role of differences among firms. Economics of Innovation and New Technology 3(4): 183–200. Brynjolfsson, E., and L. Hitt. 1996. The customer counts. Informationweek (September 8): 38–43. Brynjolfsson, E., and L. Hitt. 1996. Paradox lost? Firm-level evidence on the returns to information systems spending. Management Science 42(4): 541–58. Brynjolfsson, E., and L. Hitt. 1998. Beyond the productivity paradox. Communications of the ACM 41(8): 49–55. Brynjolfsson, E., and L. Hitt. 1998. Computing productivity: Are computer pulling their weight? Mimeo. MIT and Wharton. Brynjolfsson, E., and H. Mendelson. 1993. Information systems and the organization of modern enterprise. Journal of Organizational Computing 3: 245–55. Brynjolfsson, E., and S. Yang. 1996. Information technology and productivity: A review of the literature. In M. Zelkowitz, ed., Advances in Computers, vol. 43.

88


Brynjolfsson, E., and S. Yang. 1997. The intangible benefits and costs of computer investments: Evidence from financial markets. In Proceedings of the International Conference on Information Systems, Atlanta, GA. Brynjolfsson, E., A. Renshaw, and M. V. Alstyne. 1997. The matrix of change. Sloan Management Review 38 (Winter). Chandler, A. 1977. The Visible Hand. Cambridge, MA: Belknap. Clemons, E. K., and M. C. Row. 1992. Information technology and industrial cooperation: The changing economics of coordination and ownership. Journal of Management Information Systems 9(2): 9–28. Clemons, E. K., S. P. Reddi, and M. Row. 1992. The impact of information technology on the organization of production: The ‘‘move to the middle’’ hypothesis. Decision Sciences Department, University of Pennsylvania Working Paper. Clemons, E. K., M. E. Thatcher, and M. C. Row. 1995. Identifying sources of reengineering failures: A study of the behavioral factors contributing to reengineering risks. Journal of Management Information Systems 12(2): 9–36. Clemons, E. K. 1993. Reengineering the sales function: Reengineering internal operations. Teaching Case. Wharton School. Clemons, E. K., and M. E. Thatcher. 1997. Capital One. Proceedings of the Hawaii International Conference on Systems Sciences, IEEE. Davenport, T. H., and J. E. Short. 1990. The new industrial engineering: Information technology and business process redesign. Sloan Management Review 31(4): 11– 27. David, P. A. 1990. The dynamo and the computer: A historical perspective on the modern productivity paradox. American Economic Review Papers and Proceedings 1(2): 355–61. Dewan, S., and C.-K. Min. 1997. Substitution of information technology for other factors of production: A firm level analysis. Management Science 43(12): 1660–75. Doms, M., T. Dunne, and K. R. Troske. 1997. Workers, wages, and technology. Quarterly Journal of Economics 112(1): 253–90. Fisher, M. L. 1997. What is the right supply chain for your product? Harvard Business Review 75(2): 105–16. Galbraith, J. 1977. Organizational Design. Reading, MA: Addison-Wesley. Gordon, R. J. 1998. Monetary policy in the age of information technology: Computers and the Solow paradox. Working Paper. Northwestern University. Greenan, N., and Mairesse, J. 1996. Computers and productivity in France: Some evidence. National Bureau of Economic Research Working Paper 5836. November. Greenstein, S. 1997. From superminis to supercomputers: Estimating surplus in the computing market. In T. F. Bresnahan and R. M. Gordon, eds., The Economics of New Goods. Chicago: University of Chicago Press. Griliches, Z. 1994. Productivity, R&D and the data constraint. American Economic Review 84(2): 1–23.


89

Griliches, Z., and J. Mairesse. 1984. Productivity and R&D at the firm level. In Z. Griliches, ed., R&D, Patents and Productivity. Chicago, University of Chicago Press, pp. 339–74. Gurbaxani, V., and S. Whang. 1991. The impact of information systems on organizations and markets. Communications of the ACM 34(1): 59–73. Hall, R. E. 1999. The stock market and capital accumulation. NBER Working Paper 7180. June. Hammer, M. 1990. Reengineering work: Don’t automate, obliterate. Harvard Business Review (July–August): 104–12. Hicks, J. R. 1940. The valuation of social income. Economica 7: 105–40. Hitt, L. 1996. Economic analysis of information technology and organization. Unpublished doctoral dissertation. MIT Sloan School of Management. Hitt, L., and E. Brynjolfsson. 1996. Productivity, business profitability, and consumer surplus: Three different measures of information technology value. MIS Quarterly 20(2): 121–42. Hitt, L. 1999. IT and firm boundaries: Evidence from panel data. Information Systems Research 10(9): 134–49. Hitt, L., and E. Brynjolfsson. 1997. Information technology and internal firm organization: An exploratory analysis. Journal of Management Information Systems 14(2): 81– 101. Holmstrom, B., and P. Milgrom. 1994. The firm as an incentive system. American Economic Review 84(4): 972–91. Johnston, H. R., and M. R. Vitale. 1988. Creating competitive advantage with interorganizational information systems. MIS Quarterly 12(2): 153–65. Jorgenson, D. W., and K. Stiroh. 1995. Computers and growth. Journal of Economics of Innovation and New Technology 3: 295–316. Kelley, M. R. 1994. Productivity and information technology: The elusive connection. Management Science 40(11): 1406–25. Kemerer, C. F., and G. L. Sosa. 1991. Systems development risks in strategic information systems. Information and Software Technology 33(3): 212–23. Lau, L. J., and I. Tokutsu. 1992. The impact of computer technology on the aggregate productivity of the United States: An indirect approach. Unpublished working paper. Stanford University. August. Lichtenberg, F. R. 1995. The output contributions of computer equipment and personal: A firm-level analysis. Economics of Innovation and New Technology 3: 201–17. Loveman, G. W. 1988. An assessment of the productivity impact of information technologies. MIT Management in the 1990s Working Paper 88–054. July. Malone, T. W. 1997. Is ‘‘empowerment’’ just a fad? control, decision-making, and information technology. Sloan Management Review 38(2): 23–35. Malone, T. W., J. Yates, and R. I. Benjamin. 1987. Electronic markets and electronic hierarchies. Communications of the ACM 30(6): 484–97.

90


Malone, T. W. 1987. Modelling coordination in organizations and markets. Management Science 33(10): 1317–32. Milgrom, P., and J. Roberts. 1990. The economics of modern manufacturing: Technology, strategy, and organization. American Economic Review 80(3): 511–28. Milgrom, P., and J. Roberts. 1992. Economics, Organization and Management. Englewood Cliffs, NJ: Prentice-Hall. Morrison, C. J., and E. R. Berndt. 1990. Assessing the productivity of information technology equipment in the U.S. manufacturing industries. National Bureau of Economic Research Working Paper 3582. Morrison, C. J. 1996. Assessing the productivity of information technology equipment in U.S. manufacturing industries. Review of Economics and Statistics 79(3): 471–81. Mukhopadhyay, T., F. J. Lerch, and V. Mangal. 1997. Assessing the impact of information technology on labor productivity—A field study. Decision Support Systems 19(2): 109–22. Mukhopadhyay, T., S. Rajiv, and K. Srinivasan. 1997. Information technology impact on process output and quality. Management Science 43(12): 1645–59. Nakamura, L. I. 1997. The measurement of retail output and the retail revolution. Paper presented at the CSLS Workshop on Service Sector Productivity and the Productivity Paradox, Ottowa, Canada. April. Oliner, S. D., and D. E. Sichel. 1994. Computers and output growth revisited: How big is the puzzle?’’ Brookings Papers on Economic Activity: Microeconomics (2): 273– 334. Orlikowski, W. J. 1992. Learning from notes: Organizational issues in groupware implementation. In J. Turner and R. Kraut, eds., Conference on Computer Supported Cooperative Work. Toronto: Association for Computing Machinery, pp. 362–69. Osterberg, W. P., and S. A. Sterk. 1997. Do more banking offices mean more banking services? Economic Commentary (Federal Reserve Bank of Cleveland): 1–5. Radner, R. 1993. The organization of decentralized information processing. Econometrica 62: 1109–46. Radner, R., and T. Van Zandt. 1992. Information processing in firms and returns to scale. Annales d’Economie et de Statistique 25/26: 265–98. Rai, A., R. Patnayakuni, and N. Patnayakuni. 1997. Technology investment and business performance. Communications of the ACM 40(7): 89–97. Roach, S. S. 1988. America’s white-collar productivity dilemma. Manufacturing Engineering (August): 104. Roach, S. S. 1987. America’s technology dilemma: A profile of the information economy. Morgan Stanley Special Economic Study. April. Sah, R. K., and J. E. Stiglitz. 1996. The architecture of economic systems: Hierarchies and polyarchies. American Economic Review 76(4): 716–27. Schankerman, M. 1981. The effects of double-counting and expensing on the measured returns to R&D. Review of Economics and Statistics 63: 454–58.


91

Schnapp, J. 1998. An old strategy is backfiring at G.M. New York Times (July 12): section 3: 12. Short, J. E., and N. Venkatraman. 1992. Beyond business process redesign: Redefining Baxter’s business network. Sloan Management Review 34(1): 7–20. Sichel, D. 1996. The Computer Revolution: An Economic Perspective. Washington, DC: Brookings Institution Press. Siegel, D., and Z. Griliches. 1991. Purchased services, outsourcing, computers and productivity in manufacturing. National Bureau of Economic Research Working Paper 3678. Siegel, D. 1997. The impact of computers on manufacturing productivity growth: A multiple-indicators multiple-causes approach. Review of Economics and Statistics 79(1): 68–78. Simon, H. A. 1976. Administrative Behavior, 3rd ed. New York: Free Press. Strassmann, P. E. 1990. The Business Value of Computers. New Canaan, CT: Information Economics Press. Vitale, M., and Konsynski, B. 1988. Baxter Healthcare Corp.: ASAP Express. Harvard Business School Case 9-188-080. Williamson, O. 1975. Markets and Hierarchy: Analysis and Antitrust Implications. New York: Free Press. Wilson, D. D. 1995. IT investment and its productivity effects: An organizational sociologist’s perspective on directions for future research. Economics of Innovation and New Technology 4(3): 235–51. Wyner, G. M., and T. W. Malone. 1996. Cowboys or commanders: Does information technology lead to decentralization? Proceedings of the International Conference on Information Systems, Columbus, OH, December 15–18.


Comments on Chapter 2 Jacques Mairesse

Eric Brynjolfsson, with his coauthor Lorin Hitt, launched some ten years ago the new field of empirical economic research that we call IT econometrics.1 Their work started investigations of the relations between IT and productivity, or the so-called Solow paradox. Since then, they have endeavored successfully to maintain their lead, while a rapidly growing number of economists have entered this new field. In their contribution here, they provide an inspiring review of the evidence and of the understanding that has been gained, largely from their own work, on how computers contribute to business performance and economic growth. The general argument is that IT improves productivity by enabling complementary organizational innovations, and the conclusion is that the Solow paradox does not exists anymore, if indeed it ever did at all. I will organize my comments from the perspective of one who has always been sceptical about the Solow paradox. I will present my main arguments for not believing in it at the onset and indicate how they relate to the major points raised by Hitt and Brynjolfsson in their survey. I will show that about every word in Robert Solow statement, ‘‘We can see the computer age everywhere but in the productivity statistics,’’ is largely controversial, although the paradox itself has had a very positive influence in stimulating research in the development and performance of the digital economy.2 The Paucity of Statistics on IT My initial scepticism about the Solow paradox (as well as part of my initial interest in it) arose from the fact that I had been working for years on the issues of the productivity of Research and Development (R&D). I wondered immediately why we were (and could be) fairly

94

Jacques Mairesse

successful in the challenge of assessing the productivity of investment in R&D and why we could not be at least as successful, or more, as concerns investment in IT.3 For various reasons this challenge should be, a priori, much more difficult to meet in the case of R&D than for IT. There is typically much uncertainty in R&D outcomes, and the time lags between the expenses and the benefits when they come can be very long; the returns to R&D are also highly skewed. Moreover, the knowledge resulting from R&D has characteristics of a public good: it is difficult to fully control and to appropriate despite the existence and enforcement of intellectual property rights. R&D is thus a source of positive externalities or spillovers, and generates social returns which can be much larger than the private returns in most industries. All these properties concur in making the task to measure the productivity of R&D a formidable one. But none apply to IT, or when they do, they apply to a relatively modest extent.4 In fact it is difficult not to regard that IT investment, in comparison to R&D in general, is more like an ordinary investment in equipment. Starting from such a premise, the most important explanation for a supposedly Solow paradox seemed quite clear. We simply did not see computers in productivity statistics because we did not see them at all in statistics. So to speak there were no statistics on computers! For lack of readily available information on computer use and more generally on IT investment, as might have been provided by specific surveys, econometricians had not tackled the problems of estimating the productivity of IT. This situation was very different from that of R&D, in which the existence of regular R&D surveys in all economically advanced countries had been fundamental for the development of econometric analyses. It seemed safe enough to predict that the paradox will be dwindling away, as soon as econometricians will manage to gather data, even scanty, on IT and will apply their ingenuity to consider seriously the issue. This is indeed what begun to happen at the Charleston Conference in 1993, the first one of many to come, which gathered economists on the specific theme of the impacts of IT on productivity. At this conference Brynjolfsson and Hitt produced estimates of significant positive returns to IT, which were based on a panel data sample of large U.S. manufacturing and service firms obtained from a private source (Computerworld data). Frank Lichtenberg presented confirming estimates based on similar data (again from Computerworld and also


95

from Informationweek). Nathalie Greenan and I also showed positive evidence for France based on matched data on firms and workers.5 Since the Charleston Conference, the number of econometric studies conducted at the firm level or at a more micro-level (plant, employee), using larger and richer data sets, and based on approaches often differing in precise techniques and detailed specification, has greatly grown. This has only been possible because of the development and active mobilisation of specific sources of information on IT through new or improved statistical surveys. Most of these studies do seem to corroborate the existence of a significant productivity of IT investment and positive returns, at least comparable to the returns to other forms of investment. As could be expected, many of the studies also started focusing on more precise questions on the effects of IT, such as the changing characteristics of products and services delivered, the changing relations with competitors, suppliers, and customers, the implications for job content and worker skills, the complementarity with new organizational practices, the specific effects of networking and the Internet. Many of these issues (as well as some of the enduring econometric problems of productivity analysis) have been considered by Brynjolfsson and Hitt in their own work, and they give an insightful point of view on them in their review here. Drawing Useful Evidence from Macro and Micro Data Besides the absence or extreme scarcity of IT statistics, the other major explanation for the Solow paradox was also quite clear. We did not see computers in productivity statistics, simply because we could not see them with the naked eye! For a whole host of reasons, this is not possible without using proper techniques and data (and modeling assumptions). One cannot really expect to see an apparent effect of the rise in IT investment on the evolution of productivity, without controlling for other factors and using statistical methods to do so. Even if we consider that the share of IT investment proper (in hardware and software) relative to total investment (and to GDP) is already quite large, and is much larger when one takes into account all the complementary investments (in training, organization, etc.), and even if its growth rate is very high when one takes into account the speed of technical improvements (by using hedonic methods to measure price changes), the expected impact on the evolution of

96

Jacques Mairesse

aggregate productivity could not be so great that it could not be confounded with or counterbalanced by the impacts of other factors and macro shocks.6 Furthermore, it also seemed extremely dubious (not to say again nearly impossible) that even if we had at the macroeconomic level good series on IT investment as well as on aggregate productivity (and other relevant variables) and if even we relied on the best of statistical techniques, we could assess the overall impact of IT on productivity with reasonable confidence. There are indeed too many potential problems of estimation and model specification. This is not, of course, to say that macroeconomic information, such as that elaborated by national accounts, is not important. It is essential to give a general picture of the economy and provide basic orders of magnitude, to assess the consistency of short-term to long-term forecasts, or to make comparisons of major economic characteristics across countries, industrial sectors, and time periods. To go beyond what can be learned from scrutinizing macroeconomic information, one must have access to the underlying microeconomic information. This is what Brynjolsson and Hitt did from the start in analyzing cross-sectional and panel data of firms. The careful econometric investigation of such data allows one to assess, with reasonable confidence, the existence and magnitude of correlations between variables of interest, controlling for the influence of other relevant variables. It thus becomes possible to analyze the clustering and co-evolutions of the main factors of firm productivity level and growth, to disentangle the effects of a given factor, say IT or R&D investment from those of other factors, and to estimate its absolute or relative importance. Such microeconometric analyses, however, also have their limits. The basic problem is that good measures, or even poor ones, or even rough proxies, of some relevant factors may not be available. These omitted factors may be the cause of biases in the estimated effects of the observed factors to the extent they are correlated with them. Measurement errors in variables and issues of simultaneity in their determination raise similar difficulties. Relying only on the time-series dimension of the information in the case of firm panel data by estimating within-firm correlations and regressions has in principle the advantage of controlling for all firm unobserved permanent characteristics (or ‘‘fixed effects’’). This


97

method, however, tends go too far in discarding potentially valuable cross-sectional information, and unfortunately contribute to strongly exarcebate most other sources of biases. As a result, in practice, the within-firm estimates are often fragile and unreliable.7 In their survey Hitt and Brynjolfsson are explicit about these difficulties and explain how they have tried to overcome them in their work with apparent success—even though at times one has the feeling that they are a bit too optimistic in judging this success. A major limitation that macro- and microeconometric analyses share, of which Brynjolsson and Hitt are well aware, is in confronting explicitly the issue of causality. This issue is crucial if one wants to justify policy recommendations and contribute to elaborate and calibrate them. There is not much that economic theory and econometrics alone can do to unambiguously clarify the confusion of effects and causes beyond providing a sophisticated assessment of the network of correlations. To gain an understanding of the main causal mecanisms and their webs of influence, it is necessary to rely on well chosen and rigorous case studies. Hitt and Brynjolfsson begin their survey by providing a few enlightening case studies. They usually try to confirm their own econometric results, and to ascertain their interpretation on the basis of these illustrative cases. Related Issues and beyond the Solow Paradox In its original formulation the Solow paradox remains vague about what definition and measure of productivity it refers to. It seems likely that at the time (in the late 1980s), in the context of the ongoing economic queries about the persistance of the productivity slowdown that had occured in the developed countries in the wake of the first (1973–75) oil shock, it simply meant labour productivity as measured at the macro level. Since then, however, many economists have also interpreted the paradox in terms of TFP (total factor productivity), while others like Brynjolfsson and Hitt, rightly decided that its validity, as we just discussed, could only be tested at the micro level. This ambiguity of interpretation raise in fact two additional issues both considered in Brynjolfsson and Hitt: excess returns to IT investment, and the construction of macro-level facts from the micro evidence. The latter corresponds to an old pipe dream of national accountants that remains far from being fulfilled because of

98

Jacques Mairesse

the many practical and analytical problems it entails. To a lesser extent the issue of excess returns to IT investment also relates to a kind of a dream of econometricians. For several technical and substantial reasons, finding that IT investment has an impact on labor productivity, with positive returns on average, is less liable to errors than testing that it also has an impact on TFP as traditionnally measured (i.e., assuming equality of the returns on all factors to the ‘‘normal returns’’), and assessing by how much returns to IT investment exceed normal returns. A major problem in trying to assess the magnitude of excess returns—and this is also a problem in trying to connect evidence at the micro and macro levels—is what Hitt and Brynjolfsson themselves consider as a main argument of their survey: ‘‘IT is only a small fraction of a much larger system.’’ IT goes along with complementary factors in the form of very large investments in human capital, ‘‘organizational capital’’ and ‘‘strategic capital.’’ Clarifying the definitions and the frontiers of these various forms of intangible investments and trying to measure them is extremely difficult and necessarily involves accounting conventions. As a result, on the one hand, it does not make much sense to consider the returns to IT investment strictly circumscribed to computers, sofware, and possibly communication equipment. On the other hand, a certain degree of arbitrariness and uncertainty is unavoidable in estimating the returns on IT investment in a wider but more meaningful acceptation such as might include investments in training and changes in organizational practices to adapt to advances in IT. A decade of econometric studies based on micro data is showing the Solow paradox about not ‘‘seeing computers’’ in the productivity statistics to be a nonissue. More precise and diverse topics of interest are demanding our attention concerning the various directions of the diffusion of both IT and CT (communication technology) and how firms combine these technologies with other productive factors, assets, and ressources. A new frontier of research lies in trying to characterize and ‘‘measure’’ the organization and in investigating how IT enables organizational change, and conversely, and how the combination of both contributes to firm productivity. Brynjolfsson and Hitt rightly target a growing body of work already accomplished along these lines. They remark on their own latest findings as showing strong complementarity between IT and organizational decentralization, and positive combined effects on productivity


99

growth and market value. Much more research remains to be done, of course, to confirm, extend, and enrich these promising first findings. Notes 1. We follow here the use by Brynjolfsson and Hitt of the simple acronym IT, but we view it in fact as they do, as an equivalent to the more precise one of ICT for information and communication technologies. In no way should this suggest that communication technologies are not extremely important in conjunction with information technologies. 2. Solow’s statement was actually a casual remark in a book review (Solow 1987). In combination with the prestige of its author, its repute came from its being quoted at an important OECD Conference on Technology and Productivity (OECD 1991) and also in an influential report by an MIT Commission on Indutrial Productivity (Dertouzos, Lester and Solow 1989). 3. For a detailed survey of the results and difficulties of econometric work (performed at the firm level) on the productivity of R&D, see Mairesse and Sassenou (1991). Although this survey is now about ten years old and numerous investigations have been performed since then (in particular, in attempting to estimate R&D spillovers), most of its conclusions remain largely relevant. 4. IT, likewise R&D, can give rise to pecuniary spillovers (especially in relation to network externalities). These are, however, different from knowledge spillovers. They can only affect the distribution of benefits between consumers and businesses, and among the various IT producing and IT using sectors. In no way are the overall benefits (or social welfare) increased. On the shortcomings of the growth accounting approach to the evaluation of the contribution of IT to economic growth, see the discussion by Brynjolfsson and Hitt. 5. In chapter 2 Brynjolfsson and Hitt briefly assess these three studies (see Brynjolfsson and Hitt 1995; Lichtenberg 1995; Greenan and Mairesse 1996) and later econometric work on firm data. There were actually earlier econometric investigations on diverse (mostly small and specific) data sets that tended to find positive evidence in micro-level analyses but not in macro or industry-level studies (see the detailed survey presented at the Charleston Conference by Diane Wilson 1995). Even the negative findings in these earlier studies, as Brynjolfsson and Hitt jokingly state, could also be better characterized as ‘‘shortfall of evidence rather than evidence of a shortfall in computer productivity.’’ 6. See Oliner and Sichel (1994) who were the first to develop this line of argument. 7. On the problems of panel data econometrics in the context of productivity analysis, for example, see Griliches and Mairesse (1998).

References Brynjolfsson, E., and L. Hitt. 1995. Information technology as a factor of production: The role of differences among firms. Economics of Innovation and New Technology 3(3–4): 183–99.

100

Jacques Mairesse

Dertouzos, M. L., R. K. Lester, and R. M. Solow. 1989. Made in America: Regaining the Productive Edge. MIT Press, Cambridge, Mass. Greenan, N., and J. Mairesse. 1996. Computers and productivity in France: Some evidence. National Bureau of Economic Research Working Paper 5836. (Published in 2000 in Economics of Innovation and New Technology 9(3): 275–315.) Griliches, Z., and J. Mairesse. 1998. Production functions: The search for identification. In S. Stro¨m, ed., Econometrics and Economic Theory in the 20th Century: The Ragnar Frish Centennial Symposium. Cambridge: Cambridge University Press, pp. 169–203. Lichtenberg, F. R. 1995. The output contribution of computer equipment and personnel: A firm level analysis. Economics of Innovation and New Technology 3(3–4): 201–17. Mairesse, J., and M. Sassenou. 1991. R-D and Productivity: A Survey of Econometric Studies at the Firm Level. Science-Technology Industry Review 8. Paris, OECD, pp. 9– 43. Oliner, S. D., and D. E. Sichel. 1984. Computers and output growth revisited: How big is the puzzle? Brookings Papers on Economic Activity: Microeconomics, Washington, DC, 2, pp. 273–334. Solow, R. M. 1987. We’d better watch out. New York Times Book Review, July 12, 36. Wilson, D. D. 1995. IT investment and its productivity effects: An organisational sociologist’s perspective on directions for future research. Economics of Innovation and New Technology 3(3–4): 235–51.

3

Innovation and Employment: A Critical Survey Vincenzo Spiezia and Marco Vivarelli

3.1

Introduction

Our purpose in writing this chapter is to throw some light on the relationship between ICT diffusion and employment levels, both from a theoretical and an empirical perspective. In so doing we will also provide a critical survey of the current economic literature devoted to studying the relationship between technology and employment. Currently the main concern among specialists, and in public opinion, is about ‘‘jobless growth’’ as a source of social exclusion. The fear is that ICT have weakened—or even eliminated—the positive correlation between growth and employment, which had characterized the Fordist ‘‘golden age.’’ In some instances this fear has been translated in future scenarios where ‘‘the end of work’’ is forecast because of a possible labor-saving outcome of ICT lowering the demand for labor. However, we aim to show that the relationship between technology and employment is a more complex issue, that the lasting effects of ICT cannot be inferred from the initial dismissal of workers due to labor-saving innovation. Historically, the fear of technological unemployment has always emerged in periods characterized by radical technological changes. For example, the striking response of the English workers to the first industrial revolution was the destruction of machines under the charismatic lead of Ned Ludd in the industrial areas and of Captain Swing in the countryside (see Hobsbawm 1968; Hobsbawm and Rude´ 1969). However, since its very beginning, the economic theory has pointed out the existence of economic forces that can spontaneously compensate for the reduction in employment caused by technological progress. In other words, since the classical debate, two ‘‘extreme interpretative views’’ have started to compete in deal-

102

Vincenzo Spiezia and Marco Vivarelli

ing with the employment consequences of technological progress. Using Ricardo’s words, the ‘‘working class opinion’’ was characterized by the fear of being dismissed because of technological change (see Ricardo 1951 [1821], p. 392), while the academic and political debate was mainly dominated by a ex ante confidence in the market compensation of dismissed workers. It is interesting that nowadays, mutatis mutandis, two similar ‘‘orthodoxies’’ can be singled out within the debate about the employment impact of the ICT revolution. We will take these two streams of literature and compare the relative explanatory power of the competing theories. The chapter’s plan is as follows: The next section is devoted to the classical ‘‘theory of compensation,’’ which helps in understanding current forms of technological change. Section 3.3 presents a critical discussion of the opposite orthodoxy dealing with the alleged ‘‘end of work’’ brought about by ICT. Section 3.4 then surveys the available empirical evidence on the relationship between ICT and employment. Section 3.5 sums up the main findings of the study. 3.2

The First Orthodoxy: The ‘‘Compensation Theory’’

While the ‘‘end of work’’ literature is traditionally biased toward a gloomy picture that is characterized by the global decline of the labor force (see section 3.3), the economics profession has tried, since its foundation, to dispel all concerns about the harmful effects of technical progress. These contrasting views sum up the economic outlook in the first half of the nineteenth century when, as Luddites were destroying the new machines, economists set about developing a theory that Marx later called the ‘‘compensation theory’’ (see Marx 1961 [1867], vol. 1, ch. 13, 1969 [1905–10], ch. 18). This theory sees different market compensation mechanisms triggered by technical change that counterbalance the initial labor-saving impact of process innovation (for an extensive analysis, see also Vivarelli 1995, chs. 2 and 3; Vivarelli and Pianta 2000, ch. 2). The Compensation Mechanism ‘‘via New Machines’’ In this view, the same process innovations that displace workers in the user industries create jobs in the capital sectors where the new machines are produced (e.g., see Say 1964 [1803], p. 87).

Innovation and Employment

103

The Compensation Mechanism ‘‘via Decrease in Prices’’ Process innovations involve the displacement of workers, but these very innovations lead to a decrease in the unit costs of production. In a competitive market this effect is translated into decreasing prices. In turn decreasing prices stimulate a new demand for products, and likewise additional production and employment. This mechanism was singled out at the very start of the history of modern economic thought: The introduction of machines is found to reduce prices in a surprising manner. And if they have the effect of taking bread from hundreds, formerly employed in performing their simple operations, they have that also of giving bread to thousands. (Steuart 1966 [1767], vol. 2, p. 256)

This line of reasoning became the cornerstone of compensation theory as Say’s law became the cornerstone of classical economic theory (see Say 1964 [1803], p. 87). The idea is that in a competitive world the supply of goods generates its own demand and technical change is seen as a part of this self-adjusting process. The compensation mechanism ‘‘via decrease in prices’’ was adopted by neoclassical economists at the beginning of the twentieth century (see Clark 1907, p. 270; Pigou 1962 [1920], p. 672) and perpetuated by modern theorists (see Neary 1981; Heffernan 1981; Stoneman 1983a, chs. 11 and 12; Hall and Heffernan 1985; Dobbs, Hill, and Waterson 1987; Nickell and Kong 1989; for a detailed analysis of the hypotheses, procedures, and conclusions of these models, see Vivarelli 1995, chs. 4 and 6). It should be noted that this mechanism is at the core of the competitive effect of new technologies. Indeed, international competitiveness is mainly driven by the implementation of new cost-saving technologies, and employment compensation is mainly obtained by increases in world market shares. The Compensation Mechanism ‘‘via New Investments’’ In a world where the competitive convergence is not instantaneous, it is observed that in the gap between the decrease in costs—due to technical progress—and the consequent fall in prices, extra profits may accrue to the innovative entrepreneurs. These profits are invested and so advance more production, and new jobs are created. First set forth by Ricardo (1951 [1821], vol. 1, p. 396), this view was

104


also perpetuated by neoclassicals like Marshall (1961 [1890], p. 542) and Douglas (1930, p. 936); in recent years we have the dynamic models of Hicks (1973), Stoneman (1983a, pp. 177–81, 1983b), and Zamagni (1984). The Compensation Mechanism ‘‘via Decrease in Wages’’ As with other forms of unemployment, the direct effect of laborsaving technologies may be compensated within the labor market through an appropriate price adjustment. In a neoclassical framework—with free competition and full substitutability between labor and capital—a decrease in wages leads to an increase in the demand for labor. The first theorist to apply this argument to technological unemployment was Wicksell (1961 [1901–1906], p. 137); he was followed by Hicks (1932, p. 56), Pigou (1933, p. 256), and Robbins (1934, p. 186). In modern times wage adjustment is a component of partial equilibrium models such as those by Neary (1981) and Sinclair (1981) and general equilibrium models such as those by Layard and Nickell (1985), Venables (1985), and Layard, Nickell and Jackman (1991 and 1994). The Compensation Mechanism ‘‘via Increase in Incomes’’ In direct contrast to the decrease in wages argument, this compensation mechanism embeds the Keynesian and Kaldorian interpretation of the Fordist mode of production whereby unions take part in the distribution of the fruits of technical progress. It therefore takes into account the view that a portion of the cost savings due to technical change can be translated into higher income and hence higher consumption. This increase in demand is said to lead to an increase in employment and thus compensates for the initial job losses due to process innovations (see Pasinetti 1981; Boyer 1988c, d, 1990). The Compensation Mechanism ‘‘via New Products’’ Technical change is not equivalent to process innovation, but it can assume the form of creation and commercialization of new products. In this regard, as new economic branches develop, additional jobs are created. Again, the labor-intensiveness of product innovations is foregrounded by classical economists (Say 1964 [1803], p. 88), and


105

even the most severe critic of compensation theory has admitted that positive employment benefits can derive from this kind of technical progress (Marx 1961 [1867], vol. 1, p. 445). In the current debate various studies (Freeman, Clark, and Soete 1982; Freeman and Soete 1987, 1994; Vivarelli and Pianta 2000, chs. 5 and 11) agree that product innovations have a positive impact on employment because they open the way to the development of entirely new goods or differentiation of existing goods. In the latter case the ‘‘welfare effect’’ (new branches of production) has to be compared with the ‘‘substitution effect’’ (displacement of existing products; see Katsoulacos 1984, 1986). This compensation mechanism is similar to that of a decrease of prices as concerns international competitiveness. Overall, technical change induces market forces that can counterbalance the initial labor-saving effect of process innovation. In addition the diffusion of entirely new products can have a positive effect on employment trends. Finally, decreasing prices due to process innovation and the diffusion of product innovation are attractive to foreign markets and so operate to induce additional international demand. However, compensation mechanisms are affected by inconsistencies that are often ignored or misinterpreted by neoclassical theorists. Using the same taxonomy as proposed above, the criticisms of the compensation theory have ranged as follows: 1. With few exceptions (see Hicks 1973), nowadays the compensation mechanism ‘‘via new machines’’ is largely dismissed on the basis of Marx’s critique. Here is what he said: . . . the machine can only be employed profitably, if it . . . is the (annual) product of far fewer men than it replaces. (Marx 1969 [1905–10], p. 552)

Moreover labor-saving technologies spread around in the capital goods sector, as well; so this compensation is an endless story that can be only partial (see Marx 1969 [1905–10], p. 551). Finally, Marx foregrounded the fact that the new machines can be implemented through additional investments or simply by scrapping obsolete ones. The latter is indeed the most frequent route, and so there is no compensation at all (e.g., see Freeman, Clark, and Soete 1982). 2. As originally noted by Sismondi (1971 [1819], p. 284), Malthus (1964 [1836], vol. 2; pp. 551–60), and Mill (1976 [1848], p. 97), the

106


first effect of labor-saving technology is to decrease aggregate demand, since the cessation of demand is associated with dismissed workers. So the compensation mechanism ‘‘via decrease in prices’’ is still valid only if it is able to more than counterbalance the initial decrease in the aggregate purchasing power. Now, as this mechanism relies on Say’s law it does not take into account that demand constraints might occur. Difficulties concerning the components of the ‘‘effective demand’’ (Keynes’s term) such as a low value of the ‘‘marginal efficiency of capital’’ (again, Keynes 1973 [1936], ch. 11) can delay expenditure decisions and lower demand elasticity. As this effect occurs, the compensation mechanism is hampered, and long-term technological unemployment results (for a full critique of compensation based on Say’s law, see also Standing 1984, pp. 131–33). Finally, the compensation mechanism ‘‘via decrease in prices’’ in fact depends on the hypothesis of perfect competition. In an oligopolistic regime the compensation mechanism is weakened because cost savings are not necessarily translated into decreasing prices (see Sylos Labini 1969 [1956], p. 160). 3. The compensation mechanism ‘‘via new investments’’ relies on the Say’s assumption that the accumulated profits due to technical change are fully translated into additional investments in the same period of time. Again, Marx’s and Keynes’s treatment of Say’s law can be used to argue against the effectiveness of this compensation mechanism. Clearly, if the new investments are capital-intensive, compensation can only be partial. The accumulation of capital, though originally appearing as its quantitative extension only, is effected, as we have seen, under a progressive qualitative change in its composition, under a constant increase of its constant, at the expense of its variable constituent (Marx 1961 [1867], vol. 1, p. 628).

4. The mechanism ‘‘via decrease in wages’’ conflicts with the Keynesian notion of effective demand. Although a decrease in wages can induce firms to hire more workers, the decrease in aggregate demand will lower employers’ expectations, so they will not be able to hire the additional workers. Another criticism to this mechanism is that it does not take into account the cumulative and irreversible nature of modern technical change (see Rosenberg 1976; Dosi 1988). In this view, science and technology have their own rules: ‘‘localized technical progress’’


107

occurs along a technological trajectory and so gives rise to ‘‘locked-in’’ technologies (see Stiglitz 1987, p. 128; Freeman and Soete 1987, p. 42). If the cumulative and localized nature of technical change is taken into account, the notion of perfect substitutability between capital and labor assumed by neoclassical models can be regarded as arbitrary. Thus, according to Freeman and Soete (1987, p. 46), [T]here is inherent plausibility in the Hicks inducement theory, biasing the long term direction of technical change in a labour-saving direction. Attempts to generate a reversal of this trend by temporary small reductions in the relative price of labor are extremely unlikely to be effective.

5. In the 1950s and 1960s, the golden age of the Fordist mode of production, wages were not regulated by a competitive labor market; rather, workers were allotted a portion of the productivity gains accruing from technical progress. In turn, increased real wages meant increased mass consumption, and this stimulated investments, leading to further productivity gains through technical progress and scale economies (Boyer 1988a, b). Labor-saving technologies were introduced on large scale, but the Kaldorian ‘‘virtuous circle’’ enabled the compensation mechanism ‘‘via new incomes’’ to operate. Nowadays the Fordist mode of production is over for many reasons that cannot be discussed here (see Boyer 1988a, 1990). The distribution of income follows different rules, and the labor market has once again become competitive. Overall, the compensation mechanism ‘‘via increase in wages’’ has been weakened because of a new institutional context. 6. Development of new products is still the most powerful way to counterbalance labor-saving process innovations. Yet different ‘‘technological paradigms’’ (see Dosi 1982) are characterized by different groups of new products, which in turn have different impacts on employment. For example, in the 1950s and 1960s, the introduction of new automobile models had a more labor-intensive effect than today’s diffusion of the home computer. Indeed, in different historical periods and different institutional frameworks, the relative balance between the labor-saving effect of process innovations and the labor-intensive impact of product innovations can considerably vary. Overall, the different mechanisms of compensation theory form a complex picture of the many ways of counterbalancing direct laborsaving effects. But they are not without their drawbacks and hin-

108


drances. Economists sometimes forget that compensation can be partial and that it is dependent on particular historical and institutional circumstances, as the end-of-work literature has erroneously underestimated the opportunities of compensation (see next section). For a well-balanced and sensible conclusion about the compensation theory, we quote from Pasinetti (1981, p. 90): For the time being, we may draw the important conclusion that the structural dynamics of the economic system inevitably tend to generate what has rightly been called technological unemployment. At the same time, the very same structural dynamics produce counter-balancing movements which are capable of bringing macro-economic condition . . . towards fulfilment, but not automatically.

3.3

The Second Orthodoxy: A New Life without Work

The transition from a society based on mass employment in the private sector to one based on non-market criteria for organising social life will require a rethinking of the current world view. Redefining the role of the individual in a society absent of mass formal work is, perhaps, the seminal issue of the coming age (Rifkin 1995a, p. 235).

What is sometimes called the ‘‘end-of-work’’ literature is characterized by the following hypotheses: First, the end-of-work approach is neither academic nor concentrated within the limits of a single discipline. On the contrary, the different scenarios put forth by philosophers, sociologists, economists, and management scholars have been popularized through books that often have been best-sellers. Culturally, this tradition has two roots: one in France and one in the Anglo-Saxon world. In their respective cultural climates, the debate is very general in France and basically philosophical, while in the United States it is more social and pragmatic. Since the intent of this study is to concentrate on economic issues, the Anglo-Saxon tradition will be examined more closely, and the main propositions of the French position will be only briefly considered. To begin, in France authors like Andre´ Gorz and Guy Aznar have underscored what they believe to be the fundamental features of the postindustrial societies: the decline of the Fordist way of production, jobless growth, and opportunities to enlarge leisurely and selfabsorptive activities (see Touraine 1971; Gorz 1989; Aznar 1993). In their view the huge increase in the rate of the productivity growth


109

leaves time for nonmarket concerns, and consequently recreational interests and social needs can be properly tended. In the near future it is said that people will perform paid work for only four hours a day and will spend the rest of the day pursuing their own interests (see Gorz 1993). Technological advances and productivity growth will make this possible. It has even been proposed that work associated with a normal wage will be paid by the firm, and self-directed activities will be compensated by a second check delivered by the state (see Aznar 1993, ch. 5). In France such considerations have opened the way to a popular optimistic expectation of liberation from work (termed ‘‘computopia’’). In the more Calvinistic Anglo-Saxon world the dominant concern is about the short-run labor-saving effects of new technologies. While the vast majority of economists, commentators, and policymakers have always, and strongly believed in the thaumaturgical properties of technical progress, an influential radical minority has emphasized the gloomy consequences of automatization (Noble 1983, 1984). This more pessimistic tradition dates back to the 1930s exile of the Frankfurt School in the United States (see Marcuse 1964). In its view, the so-called technical progress is never neutral. The introduction of new technologies is spurred by profit expectations and can have adverse effects on the worker both from a qualitative and from a quantitative point of view (for the possible de-skilling impact of technical change, see Braverman 1974). More precisely, as concluded by Noble (1983), the most persuasive view is that the diffusion of computer numerically controlled (CNC) technologies has enabled the control of the labor force in two ways: in being laborsaving, these technologies keep the rate of unemployment sufficiently high so as to lower the unions’ bargaining power; and by the de-skilling effect, and the re-organisation of the shop floor, the new machines allow stronger control of those workers who have not been dismissed. As a result Noble (1983) does not hesitate to revalue the Luddite movement and to propose protective legislation that would soften the negative impact of the new technologies on labor. In regard to the economic content of this neo-Luddite approach, it is easy to see that the associated literature has mainly focused on the labor-saving effects of ‘‘process innovations’’ brought about by scrapping and investments. The introduction of microelectronics is viewed as involving a continuous decrease in labor requirements and so a massive displacement of workers (mainly manual workers

110


in manufacturing). In this view, any Keynesian policy measure appears only as a palliative, since it is not able to stop the tendency toward more automated ways of production. However, ultimately the impact of such policy measures may be neutral as the demand for labor associated with the new investments becomes completely compensated by the labor-saving effect associated with the renewal of capital (see Jenkins and Sherman 1979, p. 36). With the help of case studies deriving from the big restructuring processes carried out within large manufacturing firms in the 1980s, this literature has almost exclusively dealt with the direct laborsaving effect of process innovation. The labor-intensive implications of product innovation have often been either neglected or undervalued (e.g., see Jenkins and Sherman 1979, p. 3; Rifkin 1995a, pp. 33–35). Likewise market compensation of the initial labor-saving effect has not been fully studied. The market compensation approach has been vastly simplified to treat only general characteristics of the economic arguments (e.g., see Jenkins and Sherman 1979, p. 35; Rifkin 1995a, pp. 15–17). The a priori conclusion is that compensation is dramatically disproportionate in comparison with the prevailing direct labor-saving impact of new technologies: For more than a century, the conventional economic wisdom has been that new technologies boost productivity, lower the costs of production, and increase the supply of cheap goods, which, in turn, stimulates purchasing power, expand markets, and generates more jobs. This central proposition has provided the operating rationale for economic policy in every industrial nation in the world. Its logic is now leading to unprecedented levels of technological unemployment, a precipitous decline in purchasing power, and the spectre of a worldwide depression of incalculable magnitude and duration (Rifkin 1995a, p. 15).

The catastrophic picture of modern times is often extended from the unemployment problem to the whole condition of the working class (Aronowitz and DiFazio 1994). In support of such a thesis, Rifkin (1995a) has reported in some detail on the loss of 250,000 jobs in General Motors from 1978 to 1993 (p. 130), of 100,000 jobs in US Steel from 1980 to 1990 ( p. 134), of 170,000 jobs in General Electric from 1981 to 1993 ( p. 138), of 179,800 jobs in AT&T from 1981 to 1988 ( p. 142). However, this kind of evidence is mainly derived from isolated events rather than from aggregate statistics or historical trends.


111

Overall, this literature is characterized by some common features: 1. The analysis is often confined to the study of the direct laborsaving effect of process innovations, and the supporting empirical evidence is anecdotal. For example, the case of General Motor or AT&T is taken to represent the entire economy, while the direct labor-saving effect is considered as the only consequence of technical change. Of course, labor-saving reorganizations in large manufacturing firms do not tell us the entire story: small firms and the compensation through the expansion of the service and public sectors have to be taken into account as well. Indeed Rifkin (1995a, b) presents anecdotal evidence also for the service sectors, but unfortunately, he does not discuss aggregate employment statistics for the entire U.S. economy. Here we may reflect on a crucial point: in a given economy, is the total amount of hours of work decreasing or not? In other words, does technical change—having developed its direct and indirect impacts—affect the total demand for labor in terms of work time? The literature being discussed in this section does not provide an answer to this basic question. Of course, this does not mean that the realistic descriptions put forward by these authors do not deserve a serious consideration. But it is one thing to consider a social issue of a certain community or the serious problem of retraining a certain workforce and another thing is to forecast a dramatic decrease of the total amount of work in an entire economic system. 2. These forecasts derive from a method of analysis that underestimates the indirect effects of technical change and the job opportunities that open up because of new products and new sectors. From a static perspective, it is easy to point out any adverse effects of labor-saving innovations, but from a dynamic perspective, all the indirect effects of innovation have to be taken into account. Technical change brings about new products and the development of entirely new fields of activity that generate a demand for labor. These issues are basic to an economy’s experience of technical change. 3.4 Available Empirical Evidence on the Employment Impact of Innovation The ‘‘classical taxonomy’’ discussed in section 3.2 can be applied to the present forms of technical change and particularly to the intro-

112


duction and diffusion of ICT. On the one hand, there is a recent concern that ICT can involve lower employment coefficients and so weaken the traditional positive correlation between growth and employment. On the other hand, the conclusion that can be driven from the previous sections is that a rigorous approach cannot be based on ex ante beliefs. Economic theory does not in fact have a clear-cut answer about the employment effect of ICT. For this reason attention should be turned to aggregate, sectoral and microeconomic empirical analyses in order to take into account the different forms of technical change, their direct effects on labor, and the various compensation mechanisms and their possible drawbacks. This is not an easy task. While theoretical economists may develop clear models about the employment impact of process and product innovation, applied economists have ‘‘to measure’’ technological change, the compensation mechanisms and the final employment impact of innovation. In this regard there are at least three main problems to contend with. First, technical change and ICT diffusion are not easy to measure. Traditional indicators like R&D (input indicator) or patents and important innovation (output indicators) are seldom fully available. They often do not adequately represent technical progress (e.g., the role of tacit knowledge and intangible investments in fostering ICT diffusion). Second, as discussed in section 3.2, the overall employment impact of innovation depends on institutional mechanisms that can be very different at the micro, meso, and macro level, and the statistics vary in different economic contexts, in different countries, and in different sectors within the same country. Third, it is difficult to determine the full impact of innovation on employment, since employment depends on many other factors: macroeconomic and cyclical conditions, labor market dynamics, trends in work time, and so on. Besides these general shortcomings, there are some problems that arise at the levels of analysis. Starting with the micro studies, an empirical analysis at the firm’s level can reveal how ICT new products generate new jobs and how ICT process innovation destroys old jobs. However, a big shortcoming is that this kind of analysis produces a ‘‘positive bias’’ because it tends to overemphasize the positive employment consequences of technical change. In fact, at the


113

level of the single firm, the empirical analysis tends to characterize innovative firms as providing more employment because of the market shares they gain as a result of innovation. Even where the innovation is labor-saving, these analyses generally show a positive link between technology and employment because they do not take into account the important effect on rivals that are crowded out by the innovative firms (business stealing). This bias can be corrected by conducting the empirical analysis at the sectoral level. In this case the observer takes into account the new hires by the innovative firms, the indirect effects on the competitors, and the number of employees at the end of the diffusion process. Here too the sectoral analysis can be tainted by a negative or a positive bias depending on the observation point of view (manufacturing versus services). Also, even at the sectoral level, the analysis cannot take into account all the direct and indirect effects of technological change. Only aggregate studies can assess (1) the direct negative labor displacement of process innovation in some economic sectors (mainly manufacturing), (2) the compensation effects that operate within those sectors (through decreasing prices and increasing investments) and in other sectors (through intersectoral flows of products and incomes), and (3) the positive employment impact of product innovation in other sectors (mainly services). However, aggregate empirical analyses are very difficult to carry out because of the three problems discussed above. Microeconomic Empirical Evidence In the 1990s some interesting results were obtained using panel data analysis (see Berman, Bound, and Griliches 1994; Dunne and Schmitz 1995; Johnson, Baldwin, and Diverty 1996; Machin 1996; Betts 1997; Doms, Dunne, and Trotske 1997; Machin and Van Reenen 1998). These studies were concerned with new technology–skill complementarities, and not with the impact of ICT on employment. So only the relative impact of innovation across employment skill groups was analyzed instead of any absolute effect. These studies are discussed more fully in chapter 5 of this book. Overall, the empirical evidence suggests that the skill bias is an important determining factor in the level and the internal composition of unemployment. Obviously ICT imply a reduction in

114


low-skilled manual employment, but so far in the recent stream of literature few studies have investigated whether the employment reduction is balanced by a rise of skilled workers. Among the studies that have explicitly tried to take into account the overall absolute employment impact of ICT at the firm’s level, Entorf and Pohlmeier (1990) proposed a simultaneous model of export, innovation, and employment. By way of this model, they showed that product innovation (but not process innovation) has significantly increased the demand for labor. Using the 1984 British Workplace Industrial Relations Survey, both Machin and Wadhwani (1991) and Blanchflower, Millward, and Oswald (1991) found a negative raw correlation between ICT adoption and employment. However, once they controlled for workplace characteristics and fixed effects, the same correlation turned out to be positive. In contrast to these two studies, Brouwer, Kleinknecht, and Reijnen (1993) found an overall negative relationship between technical change and employment in 771 Dutch manufacturing firms over the period 1983 to 1988. However, the Dutch firms focusing on ICT and product innovation showed a positive relationship between innovation and employment. Klette and Førre (1998) found a similar negative relationship between R&D intensity and employment growth in the Norwegian manufacturing plants employing more than 20 employees over the period 1982 to 1992. In particular, they found out that the most R&D intensive units had a worse employment dynamics relative to the rest of the manufacturing sector, and that this result held both at the industry and at the plant level. Indeed, in dealing with firm-level data, one has to control for omitted variables, lags, and correlated fixed firms effects (e.g., more efficient firms can be jointly characterized by a higher degree of ICT adoption and a higher employment growth). Using British firm-level panel data, Van Reenen (1997) compared innovation data drawn from the Science Policy Research Unit (SPRU) database on 4,378 innovations commercialized in Britain after the Second World War. His main finding was that technological progress was associated with a higher employment rate of growth at the level of the firm. Product innovation was more likely to increase the demand for labor than process innovation.


115

Finally, Smolny (1998) analyzed 2,405 West German manufacturing firms for the period 1980 to 1992. He found a significant positive employment correlation for product innovation, but this was less clear for process innovation. Notwithstanding the accuracy of recent microeconometric studies on the employment impact of ICT, the question whether these results are generalizable remains unanswered. Indeed, it appears that more innovative firms have better employment performances, but this risks being a tautology if one does not take into account the effects on rivals, the fact that innovators are more likely to survive, and the reality of technological diffusion. For instance, Greenan and Guellec (2000) examined data on 15,186 French manufacturing firms over the period 1986 to 1990 and found a positive correlation between innovation and employment at the firm’s level (both product and process innovation). At the sectoral level, however, their results confirmed the idea that only product innovation create jobs, while process innovation generate jobs within the innovative firm but at the expense of its competitors, leading to an overall negative effect at the sectoral level. Sectoral and Input-Output Empirical Evidence The sectoral dimension is particularly important in investigating the employment impact of ICT as in recent times the compensation mechanism ‘‘via new product’’ (discussed in section 3.2) has become compensation ‘‘via new services.’’ Indeed, ICT have reshaped the ways in which most of traditional services are produced, traded and delivered. Some immediate examples are the diffusion of credit cards revoluting purchasing, email, and telebanking and teleshopping (see Miles 1993; Petit and Soete 1996; Motohashi 1997). ICT have generated entirely new services in a number of different sectors. Here we consider, for example, the business opportunities on the Internet and new communication devices like teleconference, cellular phones, ISDN, and so on (see Miles 1996; OECD 1997). The ICT diffusion reflects Engle’s law, which infers both an expansion in the final demand for new services and a shifting of the existing demand from tangible goods to intangible, information and knowledge intensive services (see Evangelista and Savona 1998; Evangelista and Perani 1998).

116


In manufacturing, however, the ICT revolution seems to be characterized mainly by labor-saving process innovation. An important characteristic to note here is that ICT labor-saving technologies are continuously introduced through scrapping, and when laborsaving innovation are dominant (Freeman, Clark, and Soete 1982, p. 160): . . . scrapping of the older equipment will in general necessitate an investment expenditure greater than the replacement value of the scrapped capital if employment levels are to be maintained.

In manufacturing, although the mechanisms via ‘‘new machines’’ and via ‘‘new investments’’ can be severely weakened by the presence of a labor-saving embodied technical change, all the other compensation mechanisms keep on being important in the ICT era. Of course, manufacturing includes the production of new technologies, which started in the 1970s and has continued for three decades: home computers and hardware, fax, network infrastructures for new ICT services (thinks about optical fibres and ISDN), video recorders and video tapes, CDs, cellular phones, and so on. Thus, although the mechanism ‘‘via new products’’ mainly operates ‘‘via new services,’’ manufacturing is not immune to the positive employment impact triggered by the diffusion of new products. Nevertheless, the empirical evidence suggests that the labor-saving bias of ICT tends to dominate in manufacturing. Some time ago, Clark (1983, 1987) put forth a supply-oriented vintage model investigating UK manufacturing. The author found that the expansionary effect of innovative investments (the Keynesian multiplier) was dominant until the mid-1960s, when the rationalizing effect (due to investments and scrapping embodying labor-saving technologies) started to overcome the expansionary one. More recently Pianta, Evangelista, and Perani (1996) investigated the relationships among value added, employment, and productivity for 36 manufacturing sectors for the G-6 over the 1980 to 1992 period. In the aggregate they found a positive relationship between growth in value added and growth in employment; that is, growing industrial sectors were more likely to increase their employment. Nevertheless, especially in European countries, an important group of sectors appeared characterized by a marked labor-saving trajectory (restructuring sectors), with growth in production and a fall in employment.


117

In another study based on Italian data, Vivarelli, Evangelista, and Pianta (1996) showed that in Italian manufacturing a negative correlation between productivity growth and employment had occurred, and, in particular, that product and process innovation had opposite effects on the demand for labor. While positive employment effects were entirely due to the disembodied forms of technical change— namely design and engineering associated with product innovation—embodied process innovations revealed to have a strong labor displacing effect. These results are consistent with what one can find in Pianta (2000) where the cases of Germany, Denmark, the Netherlands, Norway, and Italy are studied. The analysis is based on data from 21 manufacturing sectors, using a new dataset derived from the European Community Innovation Survey (a standardized cross-sectional analysis carried out in 1992). In this study industrial value added and employment correlated positively significantly, while total innovation expenditure per employee negatively affected employment. Yet the share of R&D related to product innovation showed the opposite effect. Sectoral studies overcome the limits of microeconometric estimates and are able to assess the overall employment impact of technical change. Indeed, they include both winning firms (first-comers that innovate and gain market shares) and losing firms (losers that fail to innovate or late-comers that innovate too late and cannot compensate the labor-saving impact of new technologies). None of the studies take into account all the compensation mechanisms (see section 3.2) that operate outside the sector originally affected by a labor-saving innovation. For instance, the studies discussed above are mainly focused on manufacturing, where laborsaving process innovation are dominant. Yet it is perfectly plausible that the new increased incomes (profits and real wages, connected with the compensation mechanisms via new investments and via decrease in prices) due to technical change in manufacturing may be spent in service activities. Moreover process innovation can imply an increase in the demand for capital goods (compensation mechanism via new machines), product innovation can involve an increase in employment in all the sectors connected with the production of the new product (compensation mechanism via new products), and finally the mechanism via decrease in wages also operates at an intersectoral level.

118


Thus, if the aim is to measure the overall impact of ICT diffusion and assess the effectiveness of the different compensation mechanisms, attention has to be turned to the intersectoral, aggregate level. While purely aggregate empirical models will be discussed in the next section, some input-output estimates will be discussed here. The advantage of input-output technique is the possibility of studying the links between sectors and so following the different compensation mechanisms from the initial labor-saving impact until the ultimate counterbalancing effect. The disadvantage is that the empirical simulations can properly address only the impact of process innovation, since product innovation would require additional rows and columns in the input-output matrix. Leontief and Duchin (1986) used Leontief matrices to test the employment impact of ICT, assuming four different scenarios (characterized by different paces of technological change). Their simulations took future demand evolution as exogenous. While all four simulations led to an increasing employment trend, the study revealed a clear labor-saving bias of ICT technologies: in effect, more accelerated technical progress was accompanied by lower employment growth rates. Whitley and Wilson (1982, 1987) created a multisectoral dynamic model to study the employment impact of ICT using a compensation framework. In their first study they forecast employment levels in 1990 for most sectors of British economy, and in their simulation compensation mechanisms appeared able to more than compensate initial job losses due to process innovation. Among the compensatory forces the mechanism ‘‘via decrease in prices’’ appeared the more effective mechanism, accounting for more than 50 percent of compensation of the initial labor displacement. In their second study the simulation scenario was moved to the period 1985 to 1995, and they also took into account office automation and the public sector. In this case compensation turned out to be only partial, with an overall effect of ICT equal to 288,000 job losses within British economy. Compensation mechanisms proved effective in counterbalancing 280,000 initial job losses, and most effective mechanisms appeared to be those ‘‘via decrease in prices’’ and ‘‘via new investments.’’ Close in spirit to Whitley and Wilson’s model is the framework proposed by Kalmbach and Kurz (1990). Their simulation of the impact of ‘‘microelectronic-based best-practice techniques’’ on the West Germany economy showed compensation mechanisms at work but


119

was unable to fully compensate for the initial labor displacement due to ICT diffusion. Again, for West Germany, we have the input-output study by Meyer-Krahmer (1992). Using data from 51 sectors of the entire economy in the 1980s, the author emulated the employment reaction of German economy to ICT (in-house R&D spending and purchased R&D knowledge—spillovers). His econometric results support the view that technical progress implies overall labor-saving effects; yet important sectoral differences emerge: while purchased R&D involves job losses in industries like textiles, clothing, and electronic equipment, in-house R&D stimulates the demand for labor in sectors like chemicals and computer industries. Aggregate Empirical Evidence Among the aggregate empirical studies two streams of literature can be singled out. On the one hand, some econometric studies within the ‘‘compensation approach’’ (see section 3.2) tried to test the validity of (some) compensation mechanisms within a partial or general equilibrium framework. On the other hand, more recent studies turned the attention either to the direct relationship between growth and employment, or to aggregate macroeconomic models. Sinclair (1981), Layard and Nickell (1985), and Nickell and Kong (1989) belong to the first group of studies. In the first contribution Sinclair put forward a macro IS/LM scheme and concluded that a positive employment compensation can occur if the demand elasticity and the elasticity of factor substitution are sufficiently high. Using estimates based on U.S. data, the author found a strong evidence supporting the mechanism via decrease in wages but not the mechanisms via decrease in prices. Layard and Nickell (1985) derived a demand for labor in a quasi– general equilibrium framework. The crucial parameter was the elasticity of the demand for labor in response to a variation in the ratio between real wages and labor productivity. In the model technical change was found to increase labor productivity, and given an adequate elasticity, proportionally the demand for labor was increased. This was enough to fully compensate initial job losses. Using data for the UK economy, the authors estimated an elasticity coefficient equal to 0.9, and this was sufficient, in the authors’ opinion, to rule out technical change from the possible causes of British unemployment.

120


Nickell and Kong (1989) focused on the compensation mechanism operating ‘‘via decrease in prices’’ in nine UK two-digit industries. They set up a price equation where the cost-saving effects of laborsaving technologies were fully transferred into decreasing prices; the authors found that in seven sectors out of nine a sufficiently high demand elasticity implied an overall positive effect of technical change on employment. Turning now to the second stream of literature, according to the different nature of ICT diffusion (process or product innovation) and to the different effectiveness of the compensation mechanisms, growth can be more or less labor intensive. Boltho and Glyn (1995) elaborated data on OECD countries over subperiods from 1960 to 1993. Their main results from pooling estimates show that the employment–growth relationship is not so robust from a descriptive point of view, but it is confirmed by simple econometric estimates (univariate and contemporaneous). Interestingly the positive correlation between GDP growth and employment growth is also confirmed over the period 1990 to 1993 at odds with the notion of jobless growth as a result of ICT diffusion in the OECD economies (see also Glyn 1995). Pini (1997) obtained less optimistic results. He carried out estimates of the employment elasticities—both in aggregate and by economic sectors—for the G-6 plus Sweden over the period 1960 to 1995. The more recent results (1990–1995) showed negative elasticities (jobless growth) in Italy and Sweden, while all countries but Japan show a decrease in such elasticities in comparison with the 1980s. The findings are clear-cut in the sectoral analysis: all the countries show negative elasticities for manufacturing, but they also show positive elasticities in services (see also Pini 1995, 1996). Padalino and Vivarelli (1997) undertook an empirical study of the G-7 economies over the period 1960 to 1994. Their conclusions were mainly that (1) in the long run, significant job creation in North America contrasted with moderate employment creation in Europe, (2) in manufacturing the diffusion of ICT technologies meant jobless growth and negative employment elasticities in all countries but Japan; no similar clear evidence was detectable with regard to the whole economic system, and (3) long-run evolution has to be distinguished by short-run correlation. Indeed, despite the structural differences between North America and Europe in their job creation


121

capacity in the long run, both regions nevertheless kept on showing a strong and statistically significant short-run correlation between growth and employment. Of course, growth and employment are only the final outcomes of the complex relationship between ICT and employment and the many direct and indirect mechanisms affecting this relationship (see sections 3.2 and 3.3). In an earlier study one of the authors, Vivarelli (1995, chs. 7–9), examined the direct labor-saving effect of process innovation, the different compensation mechanisms, and job creation resulting from product innovation using a simultaneous equations model. He ran three-stage least-squares regressions using Italian and U.S. data, and concluded that the more effective compensation mechanism was that ‘‘via decrease in prices’’ in both cases; the other mechanisms appeared not to be as important. In addition the U.S. economy turned out to be more product oriented (so characterized by a positive relationship between technology and employment) than the Italian economy where different compensation mechanisms could not counterbalance the labor-saving effect of prevailing process innovation. Table 3.1 summarizes the empirical studies devoted to the direct test of the relationship between technology and employment. 3.5

Conclusions

According to the compensation theory, market forces should ensure complete compensation of the initial labor-saving impact of process innovations. In section 3.2 a critique of this approach was proposed; the general conclusion was that although compensation is always working, the complete reabsorption of dismissed workers cannot be assumed ex ante. On the opposite side, the ‘‘end of work’’ literature tends to overemphasise the labor-saving effects of ICT process innovations, generalizing dramatic anecdotal empirical evidence. The critique of this approach has underlined the strong bias of this literature that tends to neglect both the indirect compensation effects of technical change and the positive impact of product innovations. Both of these two strong orthodoxies do not hold up when subjected to a critical assessment. So the debate on the employment consequences of ICT risks degenerating to a stalemate. A possible

Table 3.1 Empirical studies on the employment impact of innovation Overall impact on employment

Level of analysis

Method

Data

Entorf and Pohlmeier (1990)

Firms

Cross-sectional model, simultaneous model

German firms, 1984

Positive (when originating from product innovation)

Machin and Wadhwani (1991)

Establishments

Cross-sectional model

British Workplace Industrial Relation Survey, 1984

Positive (after controlling for other workplace characteristics)

Blanchflower, Millward, and Oswald (1991)

Establishments


British Workplace Industrial Relation Survey, 1984

Positive (after controlling for other workplace characteristics)

Brouwer, Kleinknecht, and Reijnen (1993)

Firms

Cross-sectional with sample selection model

771 Dutch manufacturing firms survived, 1983–88 (out of 859)

Negative, but with a positive impact of product innovation

Klette and Førre, (1998)

Plants

Panel analysis

All Norwegian manufacturing firms with more than 20 employees, 1982–92

Negative

Van Reenen (1997)

Firms

Panel analysis

598 British manufacturing firms, 1976–82

Positive (after controlling for lags, firms fixed effects and endogeneity of innovations)

Smolny (1998)

Firms

Panel analysis

2,405 West German manufacturing firms, 1980–92

Positive, especially with regard to product innovation

Greenan and Guellec (2000)

Firms and manufacturing sectors


15,186 French manufacturing firms, 1986–90

At firm’s level positive for both product and process innovation; at sector’s level positive for product but negative for process innovation

122

Study A. Microeconomic evidence


Manufacturing

Vintage model

U.K. manufacturing, 1948–84

Negative

Pianta, Evangelista, and Perani (1996)

Manufacturing


36 manufacturing sectors in the G6, 1980–92

Negative

Vivarelli, Evangelista, and Pianta (1996)

Manufacturing


30 Italian manufacturing sectors, 1985

Negative with regard to process innovation, positive with regard to product innovation

Pianta (2000)

Manufacturing


21 manufacturing sectors in West Germany, Denmark, Netherlands, Norway, and Italy, 1992

Negative (but positive for product innovations)

Leontief and Duchin (1986)

All sectors

Input–output model

Entire U.S. economy (simulations, 1980–2000)

Negative

Whitley and Wilson (1982)

Most sectors of the economy


Entire U.K. economy (simulation, 1990)

Positive, after controlling for compensation mechanisms

Whitley and Wilson (1987)

All sectors


Entire U.K. economy (simulation, 1985–95)

Negative, after controlling for compensation mechanisms

Kalmbach and Kurz (1990)

All sectors


Entire West German economy (simulation, 2000)

Negative, after controlling for compensation mechanisms

Meyer-Krahmer (1992)

All sectors


Entire West German economy, 1980s

Negative, but with important sectoral peculiarities

123

Clark (1983, 1987)


B. Sectoral evidence

Study

Overall impact on employment

Level of analysis

Method

Data

Sinclair (1981)

Aggregate

IS/LM macroeconomic model

U.S. economy in the post war period, 1946–73

Mixed, according to the operating of different compensation mechanisms

Layard and Nickell (1985)

Aggregate

General equilibrium model

U.K. economy, 1954–83

Neutral, after controlling for compensation through demand for labor

Nickell and Kong (1989)

Manufacturing sectors

Partial equilibrium model

U.K. manufacturing, 1974–85

Positive in 7 sectors out of 9, after controlling for the compensation mechanism ‘‘via decrease in prices’’

Boltho and Glyn (1995) and Glyn (1995)

Aggregate

Pooling analysis

OECD economies, 1960–93

Positive (no evidence of jobless growth)

Pini (1997)

Aggregate

Time series

G6 þ Sweden, 1960–95

Mixed (some evidence of jobless growth in some countries and some subperiods)

Padalino and Vivarelli (1997)

Aggregate

Time series

G7 economies, 1960–94

Mixed (some evidence of jobless growth in Europe)

Vivarelli (1995)

Aggregate

Macroeconomic simultaneous equations model (3SLS)

Italy (1967–86) and U.S. (1966–86) economies

Mixed, accordingly to the different compensation mechanisms and the different institutional contexts

124

Table 3.1 (continued)

C. Aggregate evidence



125

solution is to carry out empirical studies that assess whether or not in the aggregate the labor force is really decreasing as a consequence of innovation. From the available empirical evidence there have emerged contrasting results according to the different levels of analysis. While most microeconometric studies show a positive correlation between ICT and employment (though this correlation is affected by a strong skill-bias), some questions can be raised about the generalizability of such micro studies. At the sectoral level the distinction between product innovation (in labor friendly growing sectors as ICT-related services) and process innovation (in labor-saving restructuring sectors) is particularly important. Different input-output simulations also can have opposite results according to the functioning of different ‘‘within and between sectors’’ compensation mechanisms. At the aggregate level, contrasting empirical results about the occurrence of jobless growth in different periods and different countries reflect the different balances between product and process innovation and the different degrees of effectiveness of compensation mechanisms. In sum, it appears that economists cannot offer precise account of the employment impact of ICT—neither theoretically nor empirically. The pragmatic diagnosis is that the complex relationship between ICT and employment cannot be entirely revealed by partial equilibrium models or apodictical hypotheses or by empirical generalizations. Instead of strong optimistic or pessimistic affirmations, it is necessary to keep an open minded theoretical perspective, to use reliable data, and then painstakingly to estimate all the various direct and indirect employment effects of technological change. References Aronowitz, S., and W. DiFazio. 1994. The Jobless Future: Sci-Tech and the Dogma of Work. Minneapolis: University of Minnesota Press. Aznar, G. 1993. Travailler moins pour travailler tous. 20 Propositions. Paris: Syros. Berman, E., J. Bound, and Z. Griliches. 1994. Changes in the demand for skilled labor within U.S. manufacturing industries: Evidence from the Annual Survey of Manufacturing. Quarterly Journal of Economics 109: 367–97. Betts, J. 1997. The skill bias of technological change in Canadian manufacturing industries. Review of Economics and Statistics 79: 146–50. Blanchflower, D., N. Millward, and A. Oswald. 1991. Unionisation and employment behaviour. Economic Journal 101: 815–34.

126


Boltho, A., and A. Glyn. 1995. Can macroeconomic policies raise employment? International Labour Review 134: 451–70. Boyer, R. 1988a. Technical change and the theory of re´gulation. In G. Dosi, C. Freeman, R. Nelson, G. Silverberg, and L. Soete, eds., Technical Change and the Economic Theory. London: Pinter, pp. 67–94. Boyer, R. 1988b. Formalizing growth regimes. In G. Dosi, C. Freeman, R. Nelson, G. Silverberg, and L. Soete, eds., Technical Change and the Economic Theory. London: Pinter, pp. 608–35. Boyer, R. 1988c. New technologies and employment in the 1980s: From science and technology to macroeconomic modelling. In J. A. Kregel, E. Matzner, and A. Roncaglia, eds., Barriers to Full Employment. London: Macmillan, pp. 233–68. Boyer, R. 1988d. Assessing the impact of RD on employment: Puzzle or consensus? Paper presented at the International Conference on New Technology: Its Impacts on Labour Markets and the Employment System, Berlin, December 5–7. Boyer, R. 1990. The capital labor relations in OECD countries: From the Fordist ‘‘golden age’’ to contrasted national trajectories. CEPREMAP Working paper 9020. Paris. Braverman, H. 1974. Labour and Monopoly Capital. New York: Monthly Review Press. Brouwer, E., A. Kleinknecht, and J. O. N. Reijnen. 1993. Employment growth and innovation at the firm level: An empirical study. Journal of Evolutionary Economics 3: 153–59. Clark, J. B. 1907. Essentials of Economic Theory. New York: Macmillan. Clark, J. 1983. Employment projections and technological change. In D. L. Bosworth, ed., The Employment Consequences of Technical Change. London: Macmillan, pp. 110– 25. Clark, J. 1987. A vintage-capital simulation model. In C. Freeman and L. Soete, eds., Technical Change and Full Employment. Oxford: Basil Blackwell, pp. 86–98. Dobbs, I. M., M. B. Hill, and M. Waterson. 1987. Industrial structure and the employment consequences of technical change. Oxford Economic Papers 39: 552–67. Doms, M., T. Dunne, and K. Trotske. 1997. Workers, wages, and technology. Quarterly Journal of Economics 112: 253–89. Dosi, G. 1982. Technological paradigms and technological trajectories. Research Policy 11: 147–63. Dosi, G. 1988. Source, procedure, and microeconomic effects of innovation. Journal of Economic Literature 26: 1120–71. Douglas, P. H. 1930. Technological unemployment. American Federationist 37(8): 923– 50. Dunne, T., and J. A. Schmitz. 1995. Wages, employment structure and employer sizewage premia: Their relationship to advanced-technology usage at US manufacturing establishments. Economica 62: 89–107.


127

Entorf, H., and W. Pohlmeier. 1990. Employment, innovation and export activities. In J. P. Florens, ed., Microeconometrics: Surveys and Applications. London: Basil Blackwell. Evangelista, R., and G. Perani. 1998. Innovation and employment in services: Results from the Italian innovation survey. Paper presented at the EAEPE conference, 5–8 November, Lisbon. Evangelista, R., and M. Savona. 1998. Patterns of innovation in services: The results of the Italian innovation survey. Paper presented at the VIII annual RESER conference, 8–10 October, Berlin. Freeman, C., J. Clark, and L. Soete. 1982. Unemployment and Technical Innovation. London: Pinter. Freeman, C., and L. Soete, eds. 1987. Technical Change and Full Employment. Oxford: Basil Blackwell. Freeman, C., and L. Soete. 1994. Work for All or Mass Unemployment? Computerised Technical Change into the Twenty-first Century. London: Pinter. Glyn, A. 1995. The assessment: Unemployment and inequality. Oxford Review of Economic Policy 11: 1–25. Gorz, A. 1989. Critique of Economic Reason. London: Verso. Gorz, A. 1993. Preface. In G. Aznar, Travailler moins pour travailler tous. 20 Propositions. Paris: Syros. Greenan, N., and D. Guellec. 2000. Technological innovation and employment reallocation. Labour 14(4): 547–90. Hall, P. H., and S. A. Heffernan. 1985. More on the employment effects of innovation. Journal of Development Economics 17: 151–62. Heffernan, S. A. 1981. Technological unemployment. Unpublished Ph.D. thesis. Oxford University. Hicks, J. R. 1932. The Theory of Wages. London: Macmillan. Hicks, J. R. 1973. Capital and Time. Oxford: Oxford University Press. Hobsbawm, E. J. 1968. Industry and Empire: An Economic History of Britain since 1750. Harmondsworth, Middlesex: Penguin Books. Hobsbawm, E. J., and G. Rude´. 1969. Captain Swing. Harmondsworth, Middlesex: Penguin. Jenkins, C., and B. Sherman. 1979. The Collapse of Work. London: Cox and Wyman. Johnson, J., J. R. Baldwin, and B. Diverty. 1996. The implication of innovation for human resource strategies. Futures 28: 103–19. Kalmbach, P., and H. D. Kurz. 1990. Micro-electronics and employment: A dynamic input-output study of the West-German economy. Structural Change and Economic Dynamics 1: 371–86. Katsoulacos, Y. S. 1984. Product innovation and employment. European Economic Review 26: 83–108.

128


Katsoulacos, Y. S. 1986. The Employment Effect of Technical Change. Brighton: Wheatsheaf. Keynes, J. M. 1973. The general theory of employment, interest and money. In The Collected Writings of John Maynard Keynes. London: Macmillan; 1st ed. 1936. Klette, T. J., and S. E. Førre. 1998. Innovation and job creation in a small open economy: Evidence from Norwegian manufacturing plants 1982–92. Economics of Innovation and New Technology 5: 247–72. Layard, R., and S. Nickell. 1985. The causes of British unemployment. National Institute Economic Review 111: 62–85. Layard, R., S. Nickell, and R. Jackman. 1991. Unemployment: Macroeconomic Performance and the Labour Market. Oxford: Oxford University Press. Layard, R., S. Nickell, and R. Jackman. 1994. The Unemployment Crisis. Oxford: Oxford University Press. Leontief, W., and F. Duchin. 1986. The Future Impact of Automation on Workers. Oxford: Oxford University Press. Machin, S. 1996. Changes in the relative demand for skills. In A. L. Booth and D. J. Snower, eds., Acquiring Skills: Market Failures, Their Symptoms and Policy Responses. Cambridge: Cambridge University Press, pp. 129–46. Machin, S., and S. Wadhwani. 1991. The effects of unions on organisational change and employment: Evidence from WIRS. Economic Journal 101: 324–30. Machin, S., and J. Van Reenen. 1998. Technology and changes in skill structure: Evidence from seven OECD countries, Quarterly Journal of Economics 113: 1215–44. Malthus, T. R. 1964. Principles of Political Economy. New York: M. Kelley; 1st ed. 1836. Marcuse, H. 1964. One Dimensional Man. Boston: Beacon. Marshall, A. 1961. Principles of Economics. Cambridge: Macmillan; 1st ed. 1890. Marx, K. 1961. Capital, Moscow, Foreign Languages Publishing House; 1st ed. 1867. Marx, K. 1969. Theories of Surplus Value, London, Lawrence & Wishart, first edn. 1905–10. Meyer-Krahmer, F. 1992. The effects of new technologies on employment. Economics of Innovation and New Technology 2: 131–49. Miles, I. 1993. Services in the new industrial economy. Futures 25(6): 653–72. Miles, I. 1996. Infrastructure and the delivery of new services. In OECD, Employment and Growth in the Knowledge-Based Economy. Paris: OECD. Mill, J. S. 1976. Principles of Political Economy. New York: M. Kelley; 1st ed. 1848. Motohashi, K. 1997. ICT diffusion and its economic impact in OECD countries. Science Technology Industry Review (OECD) 20: 13–45.


129

Neary, J. P. 1981. On the short-run effects of technological progress. Oxford Economic Papers 32: 224–33. Nickell, S., and P. Kong. 1989. Technical progress and jobs. Discussion paper 366. Centre for Labour Economics, London School of Economics. Noble, D. F. 1983. Present tense technology, democracy. Journal of Political Renewal and Radical Change 3(2): 8–24; 3(3): 70–82; 3(4): 71–93. Noble, D. F. 1984. Forces of Production: A Social History of Industrial Automation. New York: Knopf. OECD. 1997. Information Technology Outlook. Paris: OECD. Padalino, S., and M. Vivarelli. 1997. The employment intensity of economic growth in the G-7 countries. International Labour Review 136: 191–213. Pasinetti, L. 1981. Structural Change and Economic Growth. Cambridge: Cambridge University Press. Petit, P., and L. Soete. 1996. Technical change and employment growth in services: Analytical and policy challenges. Paper presented at the Technology, Employment, and Labour Markets conference, 16–18 May, Athens. Pianta, M. 2000. The employment impact of product and process innovations. In M. Vivarelli, and M. Pianta, eds., The Employment Impact of Innovation: Evidence and Policy. London: Routledge, pp. 77–95. Pianta, M., R. Evangelista, and G. Perani. 1996. The dynamics of innovation and employment: An international comparison. Science Technology Industry Review, Paris, OECD, 18: 67–93. Pigou, A. 1933. The Theory of Unemployment. London: Macmillan. Pigou, A. 1962. The Economics of Welfare. London: Macmillan; 1st ed. 1920. Pini, P. 1995, Economic growth, technological change and employment: Empirical evidence for a cumulative growth model with external causation for nine OECD countries, 1960, 1990. Structural Change and Economic Dynamics 6: 185–213. Pini, P. 1996. An integrated cumulative growth model: Empirical evidence for nine OECD countries, 1960–1990. Labour 10: 93–150. Pini, P. 1997. Occupazione, tecnologia e crescita: modelli interpretativi ed evidenze empiriche a livello macroeconomico. In Accademia Nazionale dei Lincei, Atti dei conveswi Lincei 139: Sviluppo tecnologico e disoccupazione: Trasformazione della societa`, 16–18 January, Rome: Accademia Nazionale dei Lincei, pp. 113–201. Ricardo, D. 1951. Principles of political economy. In P. Sraffa, ed., The Works and Correspondence of David Ricardo, vol. 1. Cambridge: Cambridge University Press; 3rd ed. 1821. Rifkin, J. 1995a. The End of Work: The Decline of the Global Labour Force and the Dawn of the Post-market Era. New York: Putnam’s. Rifkin, J. 1995b. After work: A blueprint for social harmony in a world without jobs. Utne Reader, May–June.

130


Robbins, L. 1934. The Great Depression. London: Macmillan. Rosenberg, N. 1976. Perspectives on Technology. Cambridge: Cambridge University Press. Say, J. B. 1964. A Treatise on Political Economy or the Production, Distribution and Consumption of Wealth. New York: M. Kelley; 1st ed. 1803. Sinclair, P. J. N. 1981. When will technical progress destroy jobs? Oxford Economic Papers 31: 1–18. Sismondi, J. C. L. 1971. Nouveaux principes d’ećonomie politique ou de la richesse dans ses rapports avec la population. Paris: Calmann-Levy; 1st ed. 1819. Smolny, W. 1998. Innovations, prices and employment: A theoretical model and an empirical application for West German manufacturing firms. Journal of Industrial Economics 46: 359–81. Standing, G. 1984. The notion of technological unemployment. International Labour Review 123: 127–47. Steuart, J. 1966. An Inquiry into the Principles of Political Economy. Chicago: Oliver and Boyd; 1st ed. 1767. Stiglitz, J. E. 1987. Learning to learn, localized learning and technological progress. In P. Dasgupta and P. Stoneman, eds., Technology Policy and Economic Performance. Cambridge: Cambridge University Press. Stoneman, P. 1983a. The Economic Analysis of Technological Change. Oxford: Oxford University Press. Stoneman, P. 1983b. New technology, demand and employment. In D. L. Bosworth, ed., The Employment Consequences of Technological Change. London: Macmillan, pp. 82– 96. Sylos Labini, P. 1969. Oligopoly and Technical Progress. Cambridge: Harvard University Press; 1st ed. 1956. Touraine, A. 1971. Post-industrial Society. New York: Random House. Van Reenen, J. 1997. Employment and technological innovation: Evidence from U.K. manufacturing firms. Journal of Labor Economics 15: 255–84. Venables, A. J. 1985. The economic implications of a discrete technical change. Oxford Economic Papers 37: 230–48. Vivarelli, M. 1995. The Economics of Technology and Employment: Theory and Empirical Evidence. Aldershot: Elgar. Vivarelli, M., R. Evangelista, and M. Pianta. 1996. Innovation and employment in Italian manufacturing industry. Research Policy 25: 1013–26. Vivarelli, M., and M. Pianta. 2000. The Employment Impact of Innovation: Evidence and Policy. London: Routledge. Whitley, J. D., and R. A. Wilson. 1982. Quantifying the employment effects of microelectronics. Futures 14(6): 486–95.


131

Whitley, J. D., and R. A. Wilson. 1987. Quantifying the impact of information technology on employment using a macroeconomic model of the United Kingdom economy. In OECD, Information Technology and Economic Prospects. Paris: OECD. Wicksell, K. 1961. Lectures on Political Economy. London: Routledge and Kegan; 1st ed. 1901–1906. Zamagni, S. 1984. Ricardo and Hayek effects on a fix wage model of traverse. Oxford Economic Papers (suppl.) 36: 135–51.


Comments on Chapter 3 Marie-Claire Villeval

Technological innovations are usually analyzed as reducing labor requirements and raising capital inputs. However, innovation adoption does not necessarily entail unemployment because of the existence of counterbalancing effects. Spiezia and Vivarelli offer a rich critical survey of the current state of the art in analyzing the links between technological change and employment from both theoretical and empirical perspectives. The authors oppose two strains of economic literature providing two extreme views. On the one hand, the traditional theories of compensation emphasize the importance of various market forces in offsetting the initial labor-saving impact of technological change. On the other hand, the ‘‘end of work’’ literature emphasizes dramatic, widespread, and long-lasting unemployment consequences of process innovation. Broadly speaking, the first strain puts forward a quasi-perfect compensation, whereas the second concludes to an impossible compensation. The authors convincingly argue on the failure of each stream to explain the current effects of ICT on employment. Some classical compensation mechanisms are no longer efficient. For example, with regard to the compensation via a decrease in wages, the reduction of labor costs is not a sufficient argument to employ people. In the United States the decline in the real wages of low-educated people has not permitted their full employment, while the demand for highskilled workers has increased despite their raising relative price. On the other side, the ‘‘end of work’’ strain1 grounds most of its conclusions on anecdotal evidence. Thus it necessarily fails to provide any reliable and convincing economic reasoning. Consequently compensation would be neither perfect nor impossible, but more simply imperfect.

134


However, this imperfection of compensation needs to be theoretically and empirically grounded. Purposely the authors report a complete survey of recent empirical works, conducted at the microeconomic, sectoral, or aggregate levels, considering various indicators of innovation. Each observation level enriches the understanding of the quantitative impact of innovation on employment. So far it contributes to shape the contours of imperfect compensation. Most econometric estimations conducted at the firm level outline a positive influence of innovation on the volume of employment. It confirms, to some extent, the working of compensation via both the improvement of innovators’ market performance and the development of product innovation. Sectoral-level analyses help understanding the extent of inter- and intrasectoral compensation mechanisms. While a compensation via the development of new products/services is established, the employment performance of innovative firms is partly offset by the increasing difficulties of followers, due to shifting market shares. Last, aggregate studies emphasize the operating of a compensation mechanism via a decrease in prices, through the analysis of both demand elasticity and factor substitution elasticity. However, the analysis of imperfect compensation requires one to identify the interdependencies between these various levels. The microeconomic determinants of the impact of technological change interact with its macroeconomic determinants, with the aggregate result being also partly influenced by the sectoral dynamics of reallocation. To take an analogy, the authors use three different lenses to observe the same phenomenon. But they do not establish clearly and definitely which elements are specific to each level and how they relate to each other. Indeed, the development of many empirical studies has enriched the analysis of the consequences of technological change on employment. Their conclusions differ according to the level of observation, the methodology, the indicators of technological change retained. These empirical differences inform on the imperfection of compensation. But stating and explaining the coherence of these results requires one to renew the theoretical framework able to explain the intermediate forces that diffuse and distort the impact of an initial technological shock on employment at various levels. Indeed, the quality of empirical estimations has greatly improved over the last decade, especially thanks to the availability of new databases. However, this improvement has not been accompanied yet by a theoretical reappraisal of the links between techno-


135

logical change and employment. Furthermore these recent studies argue no longer in terms of compensation theory. Thus, instead of merely applying the classical taxonomy of compensation mechanisms to the recent empirical results, one should also better use these results to reinvestigate the theory of the transmission processes. Consequently this approach does not allow the authors to answer a crucial question crossing the various levels, a question that one could name the ‘‘paradox of the productivity paradox.’’ The positive impact of innovation on employment is mainly attributed to product innovation, whereas process innovation would be labor-saving by nature. Meanwhile the productivity paradox is based on the idea that ICT are everywhere except in the productivity statistics. Thus, as evoked in the introduction of this book, if one considers ICT as process innovation, how could a limited impact of process innovation on productivity entail massive job destruction? The second line of comment refers to the limits of an arithmetic of compensation. Even if one job creation compensates for one job destruction in relation to technological change, it does not mean that compensation is perfect. The second dimension of the imperfection of compensation is a qualitative one. One especially needs to assess the relativity of compensation with respect to both the quality of jobs and the delay of adjustment processes, two aspects that are unfortunately not taken into consideration in this critical survey. When recent empirical studies aim at measuring the quantitative impact of innovation on employment, they generally do not provide any information on the comparative quality of created versus suppressed jobs. However, the skill content of jobs, the promotion perspectives and the stability of jobs for those who occupy them, their associated wage schemes, are probably not similar. Inversely, when studies focus on the skills bias of technological change, thus noting the shifting quality of labor, they do not assess the final impact of this bias on the total employment. Does the increasing demand for high-skilled labor induced by a technological shock compensate quantitatively for the decreasing demand for low-educated labor? Beyond these questions, it would be interesting to look more critically at the studies reporting the technological bias toward skilled labor in order to refine the analysis of imperfect compensation. First, these studies reveal the current limits of some classical compensation mechanisms with regard to the mechanisms of contemporary labor markets. For example, take the compensation via a decrease in

136

Marie-Claire Villeval

wages mechanism. If competition is free and if capital and labor are substitutes, a decline in wages should compensate the impact of labor-saving technological change in increasing the demand for labor. However, if one considers the heterogeneity of both labor and capital, the analysis of factor substitutions linked to shifting relative prices becomes more complex. Recent studies reveal (1) that the use of ICT can be associated with a wage premium (and not with a wage decline) and (2) that innovation today entails a rising relative labor demand for the input whose relative price has risen the most (skilled labor). Beyond these conclusions, the heterogeneity of factor inputs (ICT vs. other capital inputs; skilled vs. unskilled labor) could be used to reformulate a more current analysis of adjustment processes, as has already been used to update the classical Hicksian analysis of the neutrality of the technological progress. Last, an appraisal of the impact on employment of technological change must account for the speed of the adjustment processes. Compensation, even if it still works, is not necessarily immediate. Further any study of the process of reallocation requires one to examine the interactions of gross (and not only net) job creation and gross job destruction, along the lines of Davis and Haltiwanger. Even if the aggregate result of ICT on the job content of growth were to be negative, this does not exclude a positive influence of innovation on gross job creation. The analysis could be improved by assessing the role of technological change in the joint dynamics of gross job creation and gross job destruction, on job and worker reallocation between and within sectors. To which extent does innovation help explain the dispersion of employment growth across sectors and occupations? Does it contribute to a higher need of job reallocation in the recent decades? Spiezia and Vivarelli give too little a weight to the recent analyses of reallocation, but in their defense, it has to be said that in the literature, empirical estimations of worker and job reallocation seldom examine the specific influence of technological change on reallocation processes. However, it has been established that the speed and the duration of job creation and job destruction are not equivalent. The speed of compensation has to be estimated, especially in a cross-country perspective. The time dimension of job and worker reallocation— and thus the timing of compensation—may differ from country to country according to the structural specificity of labor markets, the


137

efficiency of matching processes, the nature of good market structures and competitive organization. These comments suggest the need to reinvestigate the connections among innovation, the extent and the rhythm of job and worker reallocation, and the structural characteristics of markets (labor regulations, relative importance of internal labor markets, intensity of product market competition, etc.). Such analysis should also check whether organizational innovation induces the same employment dynamics as other forms of innovation. It should examine whether the relationship between innovation and employment is indifferent to the stage of innovation diffusion over the economy. Note 1. The authors qualify the ‘‘second orthodoxy’’ improperly because it cannot be put on the same analytical level as economic theory, presented as constituting the ‘‘first orthodoxy.’’


II

The Inequality Puzzle


4

Technological Bias and Employment Inequality: A Macroeconomic Perspective Henri R. Sneessens

4.1

Introduction

This chapter focuses on the impact of new technologies (among which are information technologies) on the macroeconomy when new technologies stimulate the relative demand for high-skilled labor. To what extent the deterioration of the position of low-skilled workers (either in terms of employment perspectives or of real wages) should be seen as a consequence of such biased technological change? What are, in such a scenario, the consequences of relative wage rigidities, especially in Europe? Can such a phenomenon have contributed to the persistence of high aggregate unemployment rates? These are the main questions addressed in this chapter. My objective is not to propose a definite and complete answer but rather to try and explain the difficulties met in examining these questions, and by so doing to put available empirical results (sometimes contradictory) in better perspective. The contrast between the United States and the EU countries in terms of social insurance coverage, wage dispersion, and unemployment rate changes over the last twenty years suggests that the poor employment performance of European economies may at least in part result from too little adjustment to structural changes. In part, but how much? Many studies downplay the role of structural factors. They stress the fact that both high- and low-skilled unemployment rates have increased, especially over the last ten years, while relative unemployment rates (high- and low-skilled rates compared to the aggregate rate) have decreased. The interpretation of raw data is far from easy though. As an example, table 4.1 reports the 1995 values of the unemployment and nonemployment rates across age

France

Unemployment rate Nonemployment rate Participation rate

United Kingdom


United States


Low educational level (below upper-secondary) age group

Intermediate educational level (upper-secondary) age group

High educational level (tertiary) age group

All educational levels age groups

15–19

30–44

25–64

20–24

30–44

25–64

30–44

25–64

15–19

30–44

25–64

25.3

14.2 21.0 92.0

14.0 48.0 60.0

24.8

7.4 10.0 97.0

8.9 25.0 82.0

4.1 6.0 98.0

5.9 16.0 89.0

24.4 (48) (37)

8.1 11.0 97.0

9.7 31.0 76.0

20.6 34.0 83.0

12.2 46.0 62.0

13.2

7.5 12.0 95.0

7.4 24.0 82.0

2.8 4.0 99.0

3.5 12.0 91.0

17.3 (19) (77)

8.0 14.0 93.0

7.4 27.0 79.0

12.0 31.0 78.0

10.0 46.0 60.0

9.1

5.4 13.0 92.0

5.0 25.0 79.0

1.9 5.0 97.0

2.5 13.0 89.0

16.9 (25) (70)

5.0 13.0 92.0

4.7 24.0 80.0

(37) 28.1 (77) 20.4 (70.0)

142

Table 4.1 Unemployment, nonemployment, and participation rates (%) of males in France, the United Kingdom, and the United States by education and age group, 1995 (or 1992)

Source: OECD, Education at a Glance. Note: 1992 values given in parenthesis.

Henri R. Sneessens

Technological Bias and Employment Inequality

143

and education groups for three countries: France, the United Kingdom, and the United-States. Implicit participation rates are also indicated. A few clear conclusions can be drawn from this table. In all three countries, unemployment is decreasing with education, and middle-aged workers (30–44), especially the low-educated ones, appear better protected against unemployment (this is even more striking in terms of nonemployment rates). The unemployment rate is larger in France than in the UK, and larger in the United Kingdom than in the United States ðUFr > UUK > UUS Þ, and this holds true for all education groups for workers aged 25 to 64. Workers aged 15 to 19 seem particularly vulnerable in France. The unemployment rate for low-educated young workers is 25 percent, a value similar to that of the two other countries; the nonemployment rate however is likely to be much larger, if we compare the participation rates of male workers aged 15 to 24 (given between parentheses) in the three countries. It is not obvious though, on the basis of these simple figures, that there should be more skill mismatch in France than in the United Kingdom or the United States. Table 4.2 reports standard measures of unemployment (and nonemployment) dispersion, either in the form of relative unemployment rates or in the form of unemployment rate differences across educational groups. These measures are based on the data of table 4.1. The first two columns give the ratio of the unemployment rates of low- and high-educated workers, for middle-aged workers (30–44) and for workers aged 25 to 64 respectively. The next two columns give the ratio of the total and the higheducation unemployment rates for the same two age groups. The last four columns give similar informations in terms of differences rather than ratios. Except for the last column, all measures of dispersion give almost always the same ranking, whether in terms of unemployment or of nonemployment. The United Kingdom comes first with the highest unemployment dispersion indicator, the United States comes next, and France has almost always the lowest coefficient. If one looks at the last column, however, France and the United Kingdom seem to share very similar fates, with dispersion indicators substantially larger than that in the United States. This example suggests that one should not infer too much from raw data and simple statistics. The danger of relying too much on simple unemployment dispersion indicators is further illustrated in

144

Table 4.2 Measures of male unemployment and nonemployment rate (%) dispersion by age group Ratio ul =uh

France United Kingdom United States

Ratio u=uh

30–44

25–64

30–44

25–64

Difference ul uh

Difference u uh

30–44

25–64

30–44

25–64

Unemployment rate

3.46

2.37

1.98

1.64

10.0

8.0

4.0

4.0

Nonemployment rate

3.50

3.00

1.83

1.94

15.0

32.0

5.0

15.0

Unemployment rate

7.36

3.49

2.86

2.11

18.0

9.0

5.0

4.0

Nonemployment rate

8.50

3.83

3.50

2.25

30.0

34.0

10.0

15.0

Unemployment rate Nonemployment rate

6.32 6.20

4.00 3.54

2.63 2.60

1.88 1.85

10.0 26.0

8.0 33.0

3.0 8.0

2.0 11.0

Note: u stands for aggregate unemployment rate.

Henri R. Sneessens


145

Figure 4.1 Comparison of unemployment rate level and dispersion.

figure 4.1. As is shown in the figure, a correlation may or may not exist across OECD countries between the aggregate unemployment rate and unemployment dispersion depending on the measure that is used. In the left panel dispersion is measured as the ratio of low- to high-skilled unemployment rates; in the right panel it is measured by the difference between the two unemployment rates. No significant correlation is obtained in the first case; it is clearly positive in the second. Obviously no conclusion about the role and importance of biased technical change and skill mismatch can be reached by such simple correlation exercises. Furthermore comparisons such as those made in tables 4.1 and 4.2 may give different results for different years or different types of disaggregation. They fail to take into account trends observed over the recent past, both in terms of unemployment rates and participation rates. Table 4.3 illustrates how much participation rates may have changed from 1972 to 1992 for different age groups. There are striking cross-country differences, which surely should be accounted for and explained if one wants to understand cross-country similarities and differences. There is, at the OECD level, a clear negative correlation between unemployment

146

Figure 4.1 (cont.)

Henri R. Sneessens


147

Table 4.3 Male participation rates (%) by age 15–24

25–54

55–64

65 and over

Total

1972

1992

1972

1992

1972

1992

1972

1992

1972

1992

France United Kingdom

58.0 78.0

37.0 77.0

97.0 96.0

95.0 94.0

73.0 88.0

44.0 66.0

18.0 18.0

4.0 9.0

87.0 94.0

75.0 85.0

United States

73.0

70.0

94.0

93.0

79.0

67.0

23.0

16.0

89.0

88.0

Source: Barrell et al. (1995); NIESR.

and participation rates for the workforce aged 15 to 24 (see OECD 1996, fig. 3.6). A full understanding of the structural component of the unemployment problem calls for a better understanding of all these aspects simultaneously. The difficulties encountered in evaluating the macroeconomic impact of a biased technological change are at least in part due to the many dimensions of the problem (skills, age, sex, sectors) and their many interactions influenced by the institutional features of the labor market. Macroeconomic and structural phenomena are interrelated. Evaluating the macroeconomic consequences of asymmetric shocks like biased technical progress cannot be done without a correct understanding of such interactions. A correct appraisal of the macroeconomic consequences of a biased technological change therefore should adequately (i.e., in realistic and workable terms) represent these phenomena. My objective in this chapter is to clarify some methodological issues and provide a means of interpreting and comparing results obtained by different researchers from different specifications and databases. To this end, I first propose a simple and in many ways quite standard analytical framework (section 4.2). In section 4.3, I use this framework to distinguish macroeconomic from structural shocks, and to illustrate the interactions between macroeconomic and structural phenomena as well as their implications for the interpretation of simple unemployment dispersion indicators like those mentioned above. I also discuss the limits of such a simple representation of the economy and give some extensions. Section 4.4 then is devoted to a discussion and comparison of empirical evaluations of the contribution of biased technological change to aggregate unemployment. Section 4.5 gives some concluding remarks.

148

4.2

Henri R. Sneessens

Analytical Framework

The model of equilibrium unemployment can be extended to include structural features. Here the usual NAIRU model is used to include two types of labor (low- and high-skilled labor) and the corresponding labor markets.1 In line with the equilibrium unemployment literature, the idea is to model the behavior of price and wage formation and not model explicitly the demand side of the economy. There are various ways to consider asymmetric effects of growth on the welfare (income or employment opportunities) of low- and high-skilled workers. Capital accumulation as such may have asymmetric effects when the capital-labor substitution elasticities are different for the two groups of workers (e.g., see Krusell et al. 2001). Globalization and changes in the structure of the economy provide another explanation. Our concern here is with the third possibility, biased technical progress.2 Firm’s Behavior Let us consider a setup with one type of goods (one sector) and monopolistically competitive firms. All firms have access to the same production technologies using three inputs: low- and high-skilled labor plus capital. For convenience we assume a constant elasticity of substitution s between low- and high-skilled employment, denoted Nl and Nh and neglect capital stock. This simplification avoids technicalities and complications not essential for our purpose, which is to clarify a few key concepts and issues.3 The firm’s optimal price (output) and technological choices can then be summarized in two relationships: (

X

)1=ð1sÞ s

ðai Þ ðe

gi t

wi Þ

ð1sÞ

¼ a0;

i A ðl; hÞ;

ð1Þ

i

log Nl log Nh ¼ ð1 sÞðgl gh Þt sðlog wl log w h Þ;

ð2Þ

where wl and w h stand for the real wage cost of low- and high-skilled labor respectively, coefficients gl and gh represent asymmetric exogenous rates of technical progress, and parameter a 0 reflects the effect of capital accumulation and unbiased technical progress at given markup rate.4 Equation (1) is the factor price frontier determined by


149

cost minimization and price behavior; equation (2) determines the optimal low- to high-skilled labor ratio. Wage Formation To represent the wage formation process, we rely on a standard monopoly-union-right-to-manage argument and use the following two simple relationships: log wl ¼ l 0 þ bl t wl log ul ;

ð3Þ

log w h ¼ h0 þ b h t wh log uh ;

ð4Þ

where the b’s measure the trend in wage demands generated by technical progress. Equilibrium Unemployment Equations (1) to (4) can be solved for the two equilibrium unemployment rates. We proceed to do this in two steps. We first combine the factor price frontier (1) and the wage equations (3) and (4) to obtain an equilibrium unemployment frontier (see Layard et al. 1991, p. 308). Log-linearization then yields:5 awl log ul þ ð1 aÞwh log uh ¼ a 0 þ al 0 þ ð1 aÞh0 þ aðb l gl Þt þ ð1 aÞðbh gh Þt;

ð5Þ

where a 0 1 log a 0 . This equilibrium condition implies a negative link between the two unemployment rates. This link should be understood as follows: ceteris paribus, if one of the two types of labor experiences a lower unemployment rate and exerts pressure to receive a larger real wage, the other’s wage must necessarily decrease (one moves along a given factor price frontier). This change can be obtained only through an increase in the corresponding unemployment rate. A second relationship between the two equilibrium unemployment rates is obtained by recasting the technological equation (2) in terms of unemployment rates. We first use the approximation log Ni log Li ¼ logð1 ui Þ A ui ; where Li stands for the labor force of type i, and next use equations (3) and (4) to eliminate the wage rates. We obtain the following

150

Henri R. Sneessens

equilibrium technological constraint: ½ul uh þ s½wl log ul wh log uh ¼ f0 þ sfðl 0 h0 Þ þ ðb l bh Þtg þ½ðfl fh Þ þ ð1 sÞðgl gh Þt:

ð6Þ

Parameter fi stands for the growth rate of the labor force of type i; the value of the scale parameter f0 depends on the aggregate labour force’s initial composition. This equilibrium condition implies a positive link between the two unemployment rates. This link follows from the firms’ optimal technology choices and can be understood as follows: if one of the two unemployment rates increases, the corresponding relative wage rate will decrease, which induces the firms to change their technology and reduce their relative demand for the other type of labor, whose unemployment rate also increases as a result. The equilibrium unemployment frontier (5) and the equilibrium technological constraint (6) determine together the equilibrium lowand high-skilled unemployment rates. This is illustrated in figure 4.2, in the ul uh space. The downward-sloping curve corresponds to equation (5), and the upward-sloping one to equation (6). The equilibrium combination of unemployment rates is determined by their intersection. Technology, price, and wage shocks may shift these curves up or down, thereby changing the equilibrium unemployment rate values.

Figure 4.2 Equilibrium unemployment rate.


4.3

151

Macroeconomic and Structural Shocks

The analytical framework developed in the previous section allows us to distinguish two types of shocks. We will call macroeconomic shocks all shocks that have identical direct impacts on the two labor markets;6 we call structural shocks all shocks that have different direct impacts. We will look at the consequences of each type of shock in turn, and then draw out their implications about the use of so-called mismatch indicators. We close the section with a few qualifications. Macroeconomic Shocks The macroeconomic shocks include all shocks that affect the price equation (e.g., changes in a 0 reflecting global productivity shocks) plus all shocks that affect symmetrically the two wage equations (i.e., wage shocks such that Dl 0 ¼ Dh0 , or Dbl ¼ Dbh ), the two labor demands (technological shocks satisfying Dgl ¼ Dgh ), or the two labor supplies (labor force changes such that Dfl ¼ Dfh ). By definition, a macroeconomic shock will shift the downwardsloping equilibrium unemployment frontier but leave unchanged the upward-sloping equilibrium technological constraint (see figure 4.3). A macroeconomic shock thus necessarily changes the two unemployment rates (low and high skilled) in the same direction, although the

Figure 4.3 Macroeconomic shock.

152

Henri R. Sneessens

magnitude of the change may differ from one market to the next. In other words, since symmetric shocks can have asymmetric effects, a macroeconomic shock may change the dispersion of unemployment and wage rates. The outcome depends on the slope and position of the technological constraint schedule. To illustrate the consequences of a macroeconomic shock, let us consider the case of a global productivity slowdown ðDa 0 < 0Þ. We start with the particular case where low- and high-skilled wage rates are equally sensitive to their own unemployment rate (i.e., wl ¼ wh ¼ w). This particular case is unlikely to be realistic for countries with official or de facto minimum wage rules, but it may serve as a useful starting point. We simplify the notation using this assumption to get the expressions for the equilibrium unemployment frontier and the equilibrium technological constraint respectively: a log ul þ ð1 aÞ log uh ¼ m0 ;

ð7Þ

½ul uh þ sw½log ul log uh ¼ s0 ;

ð8Þ

where m0 and s0 summarize all the terms on the right-hand side of the original equations. A productivity slowdown corresponds to an increase in m0 at fixed s0 , as illustrated in figure 4.3. Such a shock increases both unemployment rates. The effect will be asymmetric though whenever the high- and low-skilled unemployment rates have different initial values. This can easily be checked by differentiating the second equation. More precisely an increase in m0 implies that Dul b Duh when the initial unemployment rate values satisfy ul b uh ; namely the macroeconomic shock exacerbates the initial differences between the two unemployment rates. The same equation also implies that the ratio of the low- to high-skilled unemployment rates decreases toward one. The decrease is larger, the larger is the initial discrepancy between the two unemployment rates. It implies a smaller dispersion of the relative unemployment rates u i =u.7 Let us briefly consider the other extreme case, where the two wage rates have totally different sensitivities to their own unemployment rate. Let us assume, for instance, that low-skilled wages are completely insensitive to unemployment ðwh g wl ¼ 0Þ. We can then easily check that a productivity slowdown increases the difference between the two unemployment rates, though it may increase or decrease their ratio and the dispersion of the relative unemployment rates u i =u.


153

This example illustrates how a macroeconomic shock can have structural effects, namely how it can change the difference and/or the ratio between the two unemployment rates. Whether the dispersion of unemployment rates (either in levels or in relative terms) is increased or decreased crucially depends on initial conditions and on wage rigidities (sensitivity to own unemployment rate).8 Structural Shocks Structural shocks include asymmetric shocks on wages, technical progress, and labor force and do shift the equilibrium technological constraint schedule (equation 6). It may or may not simultaneously shift the equilibrium unemployment frontier (as does the macroeconomic shock), depending on the type of structural shock considered. Because these shocks can affect both the unemployment frontier and the technological constraint, they may have all sorts of effects on equilibrium unemployment rates. By way of example, let us briefly compare three scenarios: labor force composition change, asymmetric wage shock, and biased technological progress. Labor force composition changes (Df0 0 0 or Dfl 0 Dfh ) shift only the technological constraint and thus have an opposite effect on the two equilibrium unemployment rates (see figure 4.4). All other structural shocks shift both the unemployment frontier and the technological constraint.

Figure 4.4 Structural shock I: Change in workforce composition.

154

Henri R. Sneessens

Figure 4.5 Structural shock II: Imposition of minimum wage.

An exogenous increase in low-skilled wage claims (higher minimum wages, e.g., represented here by Dl 0 g 0) will shift both curves upward (ul is measured along the vertical axis) and can thus increase both the low- and the high-skilled equilibrium unemployment rates (see figure 4.5). A biased technological progress unfavourable to low-skilled workers (either gl b gh when s a 1, or gl a gh when s b 1) moves the technological constraint upward and the unemployment frontier inward; ceteris paribus (in particular, at unchanged minimum wage and unchanged b l and bh ), the high-skilled unemployment rate will certainly decrease, while the low-skilled and the aggregate unemployment rates may increase or decrease. This analysis is of course partial; it fails to take into account the effects that biased technical progress may have on wage demand behaviors (represented by coefficients bl and bh ). When such effects are allowed for, biased technological change may well have no effect at all on unemployment. To illustrate this point, let us first define exogenous mismatch (EMI) as the net exogenous relative labor demand change induced by biased technological progress. The net change is measured by the last bracketed term on the right-hand side of equation (6): EMI 1 ðfl fh Þ þ ð1 sÞðgl gh Þ:

ð9Þ


155

When EMI ¼ 0, the effect of biased technical progress is matched by labor force composition changes (which we also take as exogenous). A net exogenous relative labor demand shift ðEMI > 0Þ leaves all unemployment rates unchanged if and only if two ‘‘neutrality’’ conditions are simultaneously satisfied. The first neutrality condition is that average wage growth be compatible with average labor productivity growth, that is, ab l þ ð1 aÞb h ¼ agl þ ð1 aÞgh :

ð10Þ

When this macroeconomic condition is satisfied, the unemployment frontier (5) keeps the same position (the trade-off between the two unemployment rates remains unaffected). The second neutrality condition is that the imbalance between skill demands and supplies generated by EMI > 0 be compensated by relative wage changes. From equation (6) we have that the wage growth coefficients must then satisfy bl ¼ bh

EMI a bh : s

ð11Þ

When this second neutrality condition is satisfied, biased technical progress leaves the position of the equilibrium technological constraint (6) unaffected. Combining the two neutrality conditions yields bl ¼ agl þ ð1 aÞgh ð1 aÞ bh ¼ agl þ ð1 aÞgh þ a

EMI ; s

EMI : s

In other words, when EMI is different from zero, the stabilization of all unemployment rates is achieved at the cost of a growing wage dispersion. An alternative scenario is the one where, although EMI is positive, the high- and low-skilled wages grow at the same rate, compatible with total exogenous productivity gains, so that only the first neutrality condition (10) remains satisfied. With bl ¼ bh and EMI > 0, the technological constraint (6) shifts upward; the two unemployment rates change and move in opposite directions. Ceteris paribus, the relative wage of low-skilled workers decreases by an amount determined by the unemployment rate sensitivity of individual wages (equations 3 and 4). Compensating this effect by minimum wage

156

Henri R. Sneessens

increases (Dl 0 g 0) would, however, break the first neutrality condition as well and shift out the unemployment frontier, which could eventually yield an increase in both unemployment rates. The fact that we observe a rise in both the low- and the high-skilled unemployment rates thus cannot be interpreted as a proof that biased technological progress has not contributed to the rise and persistence of aggregate unemployment in EU countries. Pursuing an objective of stable relative wages in the face of an exogenous, technology-induced net labor demand shift is thus bound to have macroeconomic effects, whose size will depend on many parameters of the model: sensitivity of wages to unemployment, minimum wage rules, plus other propagation mechanisms that we have so far neglected, like labor supply behavior or the government’s budget constraint and its impact on labor taxes and wage costs in a context of growing unemployment. Mismatch Indicators Can we find a simple aggregate indicator that would measure the contribution of structural shocks to observed aggregate unemployment changes? The answer is likely negative. This comes from the fact that macroeconomic and structural phenomena are intimately interrelated phenomena. Because symmetric shocks can have asymmetric effects, mismatch indexes cannot give information on the nature of the shocks. Observed changes in the difference or the relative value of the low- and the high-skilled unemployment rates may a priori be the result of either macroeconomic or structural shocks. Furthermore the propagation mechanism started by a structural shock may well lead to substantial aggregate unemployment changes with little apparent structural imbalances (at least in terms of unemployment rate dispersion), as illustrated in figure 4.5. Hence unchanged unemployment dispersion does not mean that there has been no structural shock and no structural problem. ‘‘Mismatch indicators’’ may of course be designed for more specific and less ambitious objectives. For instance, Layard et al. (1991) use the variance of relative unemployment rates not to measure the contribution of structural shocks to observed aggregate unemployment changes but rather more simply to ‘‘assess how the structure of unemployment is related to its average level (both of course being


157

endogenous)’’ (Layard et al. 1991, p. 307; my emphasis). In the case where the elasticity of wage claims to unemployment is identical for both groups of workers ðwl ¼ wh ¼ wÞ, the variance of relative unemployment rates measures by how much the aggregate unemployment rate might be decreased in moving along a given unemployment frontier toward the point where all unemployment rates are equal.9 It says, of course, nothing about the nature of the shocks that produced the observed unemployment dispersion; it says also nothing about the nature of the shocks that pushed the unemployment frontier in its current position. A shift of the unemployment frontier and a movement alongside it may be two sides of the same phenomenon. There additionally exists (in our framework) a conceptually simple and natural indicator to measure the importance of biased technological shocks. It is the exogenous mismatch indicator EMI defined in equation (9). This is essentially the mismatch indicator used by Manacorda-Petrongolo (1999) and Manacorda-Manning (1997). This indicator measures the size of the shock, and not its effect on unemployment rates, which depends on induced wage changes (see above). Qualifications and Extensions Our representation of the way structural phenomena interact with macroeconomic phenomena remains extremely stylized. Structural phenomena are multidimensional. We focused here on one dimension (skills) because the available empirical evidence suggests that biased technological progress (or more generally, a biased growth process) can have profound implications for wage and employment opportunities of skill groups in all sectors of the economy. Still one should keep in mind that even if biased technological process were the sole structural shock, the existence of other structural dimensions (sectors, age, sex) would considerably affect the propagation mechanism and the final outcome of a biased technological shock, both at the aggregated and the disaggregated levels. For reasons related to institutions and bargaining power, some groups are better protected than others (typically middle-aged workers), and this asymmetry, like the others, could have macroeconomic effects. Deindustrialization could also interact with biased technological change, especially if there are wage rigidities and if moving from the manufacturing to the service sector implies relatively larger wage losses for low-skilled workers.10

158

Henri R. Sneessens

Another limitation of our simple framework deserves special attention. We have proceeded so far as if neither macroeconomic nor structural shocks had any induced effect on the labor force composition. There are serious reasons to believe that this is not the case. For instance, in the face of a biased technological shock unfavorable to low-skilled workers, public and private reactions can modify the composition of the labor force by changing the participation rates (early retirement programs, longer compulsory education, discouragement effects) or the composition of the working age population (acquisition of skills by low-skilled workers). Macroeconomic shocks can also induce (genuine or apparent) changes in the composition of the labor force by the de-skilling of long-term unemployed workers or by the so-called ladder effect whereby high-skilled workers take the jobs of low-skilled workers. Observed changes in the composition of the labor force would thus be decomposed into population of working age and participation rate changes. This means replacing in equation (6) the trend coefficients ðfl fh Þ by ðfl fh Þ 1 ðcl ch Þ þ ðpl ph Þ; where ci and pi stand for trend changes respectively in the population of working age and in the participation rate of group i A ðl; hÞ, which then allows one to examine separately the determinant of each variable. Yet another problem is the absence of a link between the technical progress parameters and the wage demand parameters. The way workers’ claims adjust in reaction to a (biased or unbiased) technological change is crucial. It reflects the characteristics of the institutional environment as well as the nature of the prevailing ‘‘social consensus’’ about income distribution and wage inequalities. The effect of exogenous productivity growth on wage claims depends on strategic interactions across sectors or segments of the labor market; it also depends on the bargaining power of the various groups of workers (skill, age, sex, sectors) and on efficiency wage considerations (to determine the share accruing to wages rather than profits).11 4.4

Empirical Models

Simple as it is, the model of the previous section provides useful insights, and it shows that evaluating the contribution of structural factors to the rise and persistence of unemployment in EU countries


159

is a difficult task. It also provides a nice reference setup to compare the empirical modeling approaches used by different authors and the results they obtain. Because simple measures of unemployment dispersion cannot provide reliable measures of the size of structural shocks, the empirical evaluation of the macroeconomic consequences of biased technological shocks necessarily goes through a genuine structural analysis. The latter involves at least two types of information: first about changes in net labor demands (the last bracketed term in equation 6); second about the determinants of real wages (especially their trend increase and their sensitivity to unemployment; see equation 5). Some studies focus on one of these two issues; others look at both issues simultaneously and draw conclusions about the factors that may explain unemployment. We briefly review a few of them. Evidence on Biased Technological Progress There exist a number of empirical studies aiming to test the biased technological progress hypothesis via the estimation of production or labor demand functions (see synthesis in table 4.4). These studies may differ by the specification of the production function and/or by the nature of the data and the definition of the skill groups. The type of a priori restrictions used to identify the parameters may also play a substantial role. Shadman-Mehta and Sneessens (1995) estimate the parameters of a four-input production function (low- and high-skilled labor, capital and energy) with exogenous labour-augmenting technical progress. The model allows a structural break in the exogenous rates of technical progress in 1974. The parameters are estimated on French annual aggregate data over the period 1962 to 1989. Skills are defined by occupation. A simplified version of the model is used in Sneessens and Shadman-Mehta (1995). The elasticities of substitution between capital and the two types of skills are not significantly different; the elasticity of substitution between low- and high-skilled workers is estimated at around 0.50. There is strong evidence, though, of asymmetric technical progress unfavorable to the lowskilled (i.e., given s f 1, they find ðgl gh Þ g 0). It is worth stressing that this asymmetry is reduced by 50% after 1974: ð1 sÞðgl gh Þ ¼ 0:0480 before 1974, 0.0245 afterward. The trends in the relative labor supply observed during the same periods are respectively ðfl fh Þ ¼

160

Henri R. Sneessens

Table 4.4 Estimates of technological bias in production models with different skill levels Data description Skill groups

Period

Country

Sector

Criterion

Number of categories

ShadmanSneessens (1995)

1962–1989

France

Macro

Occupation

2

5–2.5

DraperManders (1996)

1972–1993

Netherland

Shelter and nonshelter

Education

2

3–1

Krusell et al. (2001)

1963–1992

United States

Macro

Education

2

—

Card et al. (1999)

1982, 1989

United States, Canada and France

Micro

Education and age/ sex

>5

—

Machin et al. (1996)

1970–1990

United States, United Kingdom Denmark Sweden

16 manufacturing industries

Nonproduction vs. production

>6 2

3–2

Manacorda Petrongolo (1999)

Mid 1970s– early 1990s

Pool 10 OECD countries

Macro

Education

2

3–1

Nl =Nh (beg-end)

0:045 before 1974, and 0.023 afterward. This implies a positive but decreasing exogenous net labor demand shift in favor of high-skilled labor, equal to ð4:80 4:50Þ ¼ 0:30 percentage points per annum before 1974, ð2:45 2:30Þ ¼ 0:15 afterward. Draper and Manders (1996) use a flexible functional form (the symmetric generalized McFadden cost function) to estimate why low-skilled employment has decreased relative to high-skilled employment in the Netherlands over the period 1972 to 1993. Most of the change is explained by biased technological change. Krusell et al. (2001) estimate a four-input nested-CES production function on U.S. data. High- and low-skilled labor inputs are measured in efficiency units. Each of the two labor input indexes is defined as the productivity-weighted sum of all hours worked by individuals in different age–sex–education cells. The high-skilled worker group includes all individuals with education attainment


161

Production model

Specification

Number of inputs

Measure of biased technical progress

Elasticity of substitution slh

Sources of asymmetric employment growth

log-lineariz. dem. f.

sk.cat. þ K; M

Piecewise linear trends

0.5

Biased technical progress

SGM cost f.

sk.cat. þK

Quadratic trends

3 1.5

Biased technical progress þ asymmetric K-N substitution

CES prod.f. (nested)

sk.cat. þ Ks ; Ke

—

1.5

Asymmetric K-N substitution (Embodied technical progress)

CES prod.f.

sk.cat.

Computer use

Close to 0


trslg.cost f.

sk.cat. þK

R&D/Y

>0

Biased technical progress þ asymmetric K-N substitution

CES prod.f. (var.weight)

sk.cat.

Linear trends (country sp.)

1


level at least equal to college completion. The (normalized) relative value of the low-skilled labor input index decreases from 2 to 1 over the period covered by the analysis. The authors find evidence of a biased, unfavorable effect of growth on the demand for low-skilled labor. This effect, however, does not come through technical progress but rather through the process of capital accumulation itself, since low- and high-skilled labor have different elasticities of substitution vis-a`-vis the capital stock once the latter is decomposed into buildings and equipment. Their results suggest that nearly all the increased wage inequality could be the consequence of economic growth driven by the introduction of new technologies embodied in capital equipment. Card et al. (1999) estimate CES employment equations (together with labor supply equations and wage flexibility parameters; see below) on micro data from three different countries: United States,

162

Henri R. Sneessens

Canada, and France. The micro data are aggregated in age–education cells (skill is thus proxied by educational attainment). Two observations are available for each cell: one in the late 1970s–early 1980s, the other in the late 1980s. Measures of exogenous labor supply changes (population of working-age changes) are directly available from the data. The bias in technical progress is proxied by the proportion of workers using computers in each group toward the end of the 1980s, assuming that the progression of this ratio has been similar in all three countries during the period considered. The model performs well at least for men aged between 25 and 54. The coefficient associated to the biased technical progress indicator is correctly signed and statistically different from zero; the estimated elasticity of substitution is, however, extremely low (0.16), implying that relative wages should have little impact on relative employment. The difficulty may come from the representation of wage determination and/ or labor supply behaviors included in the model. Machin et al. (1996) estimate a translog cost function on panel data covering 16 manufacturing industries and four countries ( United States, United Kingdom, Denmark, Sweden). Four productive factors are distinguished: low- and high-skilled labor, physical capital, and R&D (in percentage points of value added). The R&D variable is meant to capture the effects of investment in new technologies. Skills are defined by production or nonproduction status, but the authors checked the robustness of their findings on U.S. and U.K. data by using an education criterion. As in Krusell et al. (2001), the authors find evidence of important skill–capital and skill–technology complementarities. This implies biased technological progress, although not large enough to explain the observed changes in the skill structure in the United States and the United Kingdom. The authors give some evidence that institutional factors (giving better protection to low-skilled workers) rather than international trade can explain these differences. These findings stress again the need to try and take simultaneously into account the several dimensions of macroeconomic and structural interactions. Manacorda-Petrongolo (1999) estimate a CES relative employment function on annual data from ten OECD countries (United States, Canada, Australia plus seven EU countries including France, Italy, Spain, and United Kingdom). Skill is here proxied by educational attainment. Due to a lack of data on relative wages, the number of observations may vary considerably from country to country. By


163

pooling all the countries’ observations and imposing the same elasticity of substitution in all countries, it is possible to estimate a country-specific coefficient of biased exogenous technical progress (corresponding to ð1 sÞðgl gh Þ in equation 6).12 The estimated elasticity of substitution is not significantly different from one. In all countries the technical progress coefficient is positive (technical progress is unfavorable to the low skilled); it varies from 0.051 in Australia to 0.076 in the United Kingdom. These estimates are next compared to the trend change in labor force composition ðfl fh Þ. The proportion of low-skilled (i.e., low educational attainment) workers has decreased everywhere; the estimated net labor demand shift, however, remains positive, hence unfavorable to the lowskilled in all countries but one, the Netherlands. The net shift appears to have been much higher in the United States than in France during the 1980s (4:92 3:21 ¼ 1:71 percentage points per annum in the United States, against 6:54 6:09 ¼ 0:45 in France), despite a lower biased technical progress effect in the United States. In the United States the net shift increases from 0.82 in the 1970s to 1.71 in the 1980s, an increase mainly due to a reduction in the relative rate of growth of high-skilled labor supply. Manacorda-Manning (1997) build on this previous work by taking into account explicitly that educational attainment may be an imperfect measure of skill. To this end, they have to start with a continuum of skill levels and next aggregate in two or more groups. Assuming a unitary elasticity of substitution between skills (CobbDouglas production function), they are able to obtain yearly estimates of exogenous net labor shifts for the United States and five EU countries. There is again in all countries evidence of shifts unfavorable to the low-skilled. In four EU countries (France, Germany, Italy, and United Kingdom), the shift keeps the same value throughout the 1980s; it is decreasing in the Netherlands, increasing in the United States. It seems fair to conclude that there is fairly strong evidence of biased technological change unfavourable to the low-skilled.13 In almost all countries the composition of the labor force did not change rapidly enough to compensate this bias, which means that the relative wage cost of the low-skilled had to decrease to stabilize the difference in unemployment rates. There are only a few studies that examine how this net demand shift may have changed over time. A tentative conclusion is that the net demand shift has remained stable

164

Henri R. Sneessens

or even decreased in France and has increased in the United States. It is worth recalling, though, that taking observed changes in the labor force composition as given, as if they were exogenous, may be misleading, especially when there are early retirement programs and changes in the compulsory schooling age that reduce the participation rates of older and younger workers. In such a case the calculated net demand shifts may seriously underestimate the true magnitude of the shock.14 The relative wage adjustment needed to offset a net demand shift of course depends on the size of the elasticity of substitution. There is not much consensus (to say the least) about the latter. The estimates reported here vary from almost zero to about three.15 Evidence on Wage Rigidities A biased technological change may leave relative employment unchanged if relative wage costs are suitably adjusted. Whether or not this will occur depends on many factors, including the sensitivity of wages to unemployment rate changes, the existence of insideroutsider effects, and wage rivalry, for example. Cohen (1997), comparing France and the United States, obtains that bargaining power on the various segments of the labor market is similar in the two countries, except for workers with no diploma, where the bargaining power is 20 percent higher in France. In Sneessens and Shadman (1995) the French low-skilled wage rate appears to be insensitive to unemployment rate changes, while the skilled wage rate is sensitive to changes in its own unemployment rate. Card et al. (1999) obtain a larger wage rigidity in all age–education cells in France, compared to the United States. There is a single coefficient measuring wage flexibility, which broadly corresponds to the ratio between the realized wage adjustment and the one that would be needed to restore full employment. If the adjustment needed to restore full employment is larger on the low-skilled labor market segment (which seems a most plausible assumption in situations of biased technical progress unfavourable to the low-skilled), their result implies a larger nonadjustment of low-skilled wages. Manacorda-Petrongolo (1999) impose the same elasticity of wages to unemployment across skill groups, and obtain similar values for the United Kingdom and the United States (around 0.03). They allow, however, for trend effects similar to those of equations (3) and


165

(4) (our coefficients bl and b h ) and obtain striking differences between the two countries. The trends are positive and similar for low- and high-skilled workers in the United Kingdom (a case similar to the ‘‘second scenario’’ discussed in the previous section); they are negative in the United States, especially for low-skilled workers (twice larger—in absolute value—for low-skilled workers; this case is similar to the ‘‘first scenario’’ discussed in the previous section). Risager (1992) estimates on Danish data (1951–1987) a model with two types of skills and wage rivalry. For both types of labor, ownunemployment wage elasticities are not significantly different from zero; there are, however, significant insider–outsider effects on the low-skilled segment of the labor market (highly significant negative lagged employment effects). The model implies a fairly constant longrun relative wage. A fair conclusion seems to be that there is more wage rigidity in Europe than in the United States, and more wage rigidity at the low end of the wage spectrum. Simultaneously we should recognize that our understanding of the many dimensions of the wage formation process remains limited. Wage–wage interactions, in particular, remain poorly understood and estimated, which certainly limits our ability to evaluate correctly the link between biased technical progress (or structural shocks in general) and aggregate unemployment. This difficulty may help explain the controversy about the effect of minimum wages. Whether minimum wage laws have significant macroeconomic implications remains a debated issue. Empirical work based on fairly aggregated data (e.g., see Bazen-Martin 1991) has indicated that minimum wages have little effect on employment. More recent studies based on micro data (Di Nardo et al. 1996; Abowd et al. 1999) indicate, however, that an increase in the minimum wage can dramatically reduce employment opportunities for workers paid at the minimum wage. An easy way of reconciling the two results is to recognize that the proportion of workers paid the minimum wage is quite small, so the aggregate effects are almost negligible. This argument of course neglects the effects of wage– wage interactions. Biased Technological Change and Unemployment There is thus good evidence of exogenous net labor demand shifts and of wage rigidities, especially for low wages. Exogenous net rel-

166

Henri R. Sneessens

ative labor demand changes induced by biased technological progress are thus likely to have contributed to observed unemployment changes. As we saw earlier, quantifying this effect is difficult, given the many interactions between macroeconomic and structural phenomena. Not all authors attempt this exercise. Card et al. (1999) obtained that relative wage rigidities have almost no impact on relative employment because the elasticity of substitution s is close to zero. These authors conclude that stronger relative wage rigidities in France can not be held responsible for lowskilled unemployment. Sneessens and Shadman-Mehta (1995) reached a different conclusion. Biased technical progress, along with minimum wage changes were found to increase unemployment. Together these two factors contributed to five out of the eight percentage point unemployment increase observed in France between 1962 and 1989. As for the United Kingdom, Manacorda-Petrongolo (1999) found that exogenous net labor demand shifts account for half the increase observed in aggregate unemployment between 1975 and 1992 (three percentage points out of six), the rest being accounted for by increased wage pressure (the trend terms in the wage equations). In the United States, exogenous net labor demand shifts, although stronger, have been fully offset by relative (and absolute) wage changes. 4.5

Conclusions

The fact that there is biased technical progress (or at least, that growth has asymmetric effects) is little disputed. Evaluating the effect of this bias on unemployment still remains a difficult task. We first stressed the danger of working with simple measures of unemployment dispersion because they cannot provide reliable measures of the size of structural shocks. We emphasized instead the need to have a genuine structural analysis. The latter involves at least two types of information: about net exogenous changes in relative labour demands and about the determinants of real wages (especially their trend increase and their sensitivity to unemployment). To clarify these issues, we developed a simple analytical framework with two types of labor: high- and low-skilled. We saw in this setup that pursuing an objective of stable relative wages in the face of exogenous, technology-induced net relative labor demand shifts is bound to have macroeconomic effects, whose size will depend on


167

many parameters. Among these parameters were the sensitivity of wages to unemployment and labor supply behavior and other propagation mechanisms such as the government budget constraint and its impact on labor taxes and wage costs in a circumstances of growing unemployment. In taking such an enlarged view, we observed that a rise simultaneously in the high- and in the low-skilled unemployment rates does not imply that biased technological progress did not contribute significantly to the rise and persistence of aggregate unemployment in EU countries. This analytical framework was next used to compare the results obtained in a number of empirical studies. We pointed out some of the difficulties met in implementing such empirical evaluations of the impact of biased technical progress and of structural shocks. One big difficulty is the endogeneity of labor force composition changes. The sharp reduction in the participation rates of younger and older workers in many EU countries, for instance, should be taken as an endogenous rather than an exogenous phenomenon. The failure to take it into acount probably leads to the underestimation of the effect of biased technological progress. Biased technical progress has probably been at work for a long time. A most relevant question in this perspective is why it should lead today to structural imbalances, while it was compatible in the past with full employment and reduced inequalities. Existing empirical models with structural features allow too few interactions between macroeconomic and structural phenomena, and between different types of structural shocks (e.g., deindustrialisation vs. technical progress). The former may lead to overestimating the importance of structural shocks (by ignoring the structural implications of macro shocks, especially the so-called ladder effect); the latter may lead to its underestimation (by neglecting the amplifying effects of structural interactions). Ignoring macroeconomic and structural interactions may be a serious shortcoming. Dre`ze (1997) shows that when there are real rigidities similar to those considered here (relative real wage rigidities), pessimistic self-fulfilling demand expectations can make the ‘‘size’’ of the economy arbitrarily small, via capital scrapping and persistent unemployment. Such a macroeconomic evolution can clearly have structural implications, especially if high-skilled workers can reduce their own unemployment risk by moving to the lowskilled labor market and taking the jobs of low-skilled workers

168

Henri R. Sneessens

(ladder effect). Such behaviors may explain our difficulty to disentangle the effects of macroeconomic and of structural shocks. More work on these issues should be most welcome. Notes This chapter was prepared for the XIth Se´minaire franco-ame´ricain CREST-NBER on Information and Communication Technologies, Employment and Earnings, organized by the CSERC at the Universite´ de Nice-Sophia Antipolis, 22–23 June 1998. The author is grateful to Michel Lubrano and an anonymous referee for useful comments on an earlier version. The research is part of a research programme supported by the Belgian government (Poˆles d’attraction inter-universitaires PAI IV) 1. This section is based on the first section of Sneessens (1994). 2. These three ‘‘explanations’’ may of course be strongly interrelated, for example, when there is embodied technical progress that facilitates communication and globalization. 3. Neglecting the capital stock implicitly amounts to assuming that the two types of labor have the same elasticity of substitution with respect to capital, which may not be the case. High-skilled labor is usually considered to be less substitutable by capital. As a result higher real interest rates—as observed in Europe over the last ten or fifteen years—will have a favorable effect on low-skilled employment opportunities, ceteris paribus. 4. More precisely, we write the production function as Y ¼ kfal ½e gl t Nl r þ ah ½e gh t Nh r gb=r ; where b < 1 and al þ ah ¼ 1. The substitution elasticity is equal to s ¼ 1=ð1 þ rÞ. With this specification, it can be shown that ð1bÞ=b bk k a0 1 ; 1þm Y where m is the markup rate. 5. Parameter a is defined as a1

als ðegl t wl Þð1sÞ

Si ais ðegi t wi Þð1sÞ

6. It is worth noting that there is no room for macroeconomic demand shocks in this model. 7. The variance of the relative unemployment rates is proportional to ðul uh Þ=u, where the aggregate unemployment rate u is given by u ¼ aul þ ð1 aÞuh . 8. Allowing different elasticities of substitution between labor and capital for the two types of skill would introduce yet another channel by which macroeconomic shocks— such as a change in the capital usage cost—could have structural implications. 9. The result is obtained by adding and subtracting the log of aggregate unemployment in the unemployment frontier equation, under the assumption that wl ¼ wh ¼ w.


169

The equilibrium unemployment frontier can then be written as ul uh 1 ui log u ¼ a 00 a log þ ð1 aÞ log A a 00 var : 2 u u u 10. The change in aggregate unemployment is positively correlated to industrial employment losses in OECD countries over the period 1983 to 1993. See OECD (1997, p. 36). 11. Ageńor and Aizenman (1997) provide an interesting analysis of such interactions and their implications. 12. The authors estimate ð1 sÞðgl gh Þ as one coefficient, without imposing the condition that it be equal to zero when s is one. This amounts to using a Cobb-Douglas function with changing wage share coefficients. 13. Estimates of biased ‘‘technological change’’ relying on unexplained trend coefficients may well pick up the effects of factors others than technical progress such as international trade and globalization or changes in the structure of the economy. 14. Card et al. (1996) treat participation rates as endogenous and solely as a function of the real wage rate. Part of the difficulty, however, is avoided when focusing on men aged 25 to 54 whose participation rates are more stable. 15. See also, for instance, Bound and Johnson (1992), who report a value larger than 2. Values of s larger than 1 are typically linked to negative values of gl gh , so the relative wage implications remain essentially unchanged.

References Abowd, J. M., F. Kramarz, T. Lemieux, and D. N. Margolis. 1999. Minimum wages and youth unemployment in France and the United States. In D. Blanchflower and R. Freeman, eds., Youth Unemployment and Joblessness in Advanced Countries. Chicago: University of Chicago Press, pp. 427–72. Ageńor P. R., and J. Aizenman. 1997. Technological change, relative wages and unemployment. European Economic Review 41(2): 187–206. Bazen, S., and J. Martin. 1991. L’incidence du salaire minimum sur les gains et l’emploi en France. Revue ećonomique de l’OCDE 16. Bound, J., and G. Johnson. 1996. Changes in the structure of wages in the 1980’s: An evaluation of alternative explanations. American Economic Review 82(2): 371–92. Card, D., F. Kramarz, and T. Lemieux. 1999. Changes in the relative structure of wages and employment: A comparison of the United-States, Canada and France. Canadian Journal of Economics 32(4): 843–77. Cohen, D. 1997. A comparison of inequalities across French and US labour markets. Paper presented at the CEPR Conference on Unemployment Persistence and the Longrun: Reevaluating the natural rate, Vigo, November 1997. Di Nardo, J., N. M. Fortin, and T. Lemieux. 1996. Labor market institutions and the distribution of wages, 1973–1992: A semi-parametric approach. Econometrica 64(5): 1001–44.

170

Henri R. Sneessens

Draper, N., and T. Manders. 1996. Structural changes in the demand for labour. Centraal Planbureau, Research Memo no 128. The Hague. Dre`ze, J. H. 1997. Walras-Keynes equilibria, coordination and macroeconomics. European Economic Review 41(9): 1737–62. Krusell, P., L. Ohanian, J. V. Rios-Rull, and F. Violante. 2001. Capital-skill complementarity and inequality. Econometrica 68(5): 1029–53. Layard, R., S. Nickell, and R. Jackman. 1991. Unemployment: Macroeconomic Performance and the Labour Market. Oxford: Oxford University Press. Machin, S., A. Ryan, and J. Van Reenen. 1996. Technology and changes in skill structure: Evidence from an international panel of industries. Center for Economic Performance. The Labour Market Consequences of Technical and Structural Change Discussion Paper Series, DP no 4. London School of Economics. Manacorda, M., and B. Petrongolo. 1999. Skill mismatch and unemployment in OECD countries. Economica 66: 181–207. Manacorda, M., and A. Manning. 1997. Just can’t get enough: More on skill-biased change and unemployment. Mimeo. Center for Economic Performance. London School of Economics. OECD. 1996. Regards sur l’e´ducation. Analyse, Paris: OECD. OECD. 1997. Implementing the OECD jobs strategy: Member Countries’ Experience. Paris: OECD. Risager, O. 1992. Wage rivalry and insider–outsider relations: Evidence from skilled and unskilled men in Denmark. Scandinavian Journal of Economics 94(4): 543–59. Shadman-Mehta, F., and H. R. Sneessens. 1995. Skill demand and factor substitution. CEPR Discussion Paper 1279. Sneessens, H. R. 1994. Croissance, qualifications et choˆmage. Revue francaise d’ećonomie 9: 1–33. Sneessens, H. R., and F. Shadman-Mehta. 1995. Real wages, skill mismatch and unemployment persistence; France, 1962–89. Annales d’ećonomie et de statistique 37/38: 255–92.

Comments on Chapter 4 Jean-Pierre Laffargue

An often-quoted stylized fact by macroeconomists is that low-skilled workers saw their position deteriorate in the 1980s and at the beginning of the 1990s. In North America and the United Kingdom this deterioration appeared in the real wage rate. In Continental Europe it took the form of a sharp decrease in employment and an increase in the rate of unemployment. Henri Sneessens investigates this problem using the method of labor macroeconomists. He builds a simple theoretical equilibrium model of a closed economy. This model assumes monopolistic competition on the goods market. Firms use two production factors: skilled and unskilled labor and determine their levels to maximize their profits. The wages of each qualification depend on its unemployment rate. Thus the deterioration in the position of unskilled labor (wages and unemployment) can be related to various shocks (technological progress, labor supply, etc.). These deteriorations depend on the substitution elasticity between skills, in the production function of firms, and on the elasticity of wages to unemployment (the real rigidity of wages). A lesson of the model is that these elasticities cannot be ignored by any serious empirical investigation of the problem. There is no simple relation between the properties of the shocks and their effects on the economy. More precisely, a symmetric shock can have nonsymmetric effects, and inversely. Then Sneessens gives a survey of the macroeconometric literature on the subject. The results he gets differ on many points, especially in the evaluation of the elasticity of substitution between skills and in the measurement of the rigidity of real wages. However, the general conclusion emerges that the deterioration in the position of unskilled labor is related to a nonneutral break in technological progress that took place in the 1980s. However, the nature of this break differs across studies. For Shadman-Mehta and Sneessens (1995) the

172

Jean-Pierre Laffargue

elasticity of substitution between skills is lower than one, and the efficiency of labor increases faster for the less skilled. Draper and Manders (1996) think just the opposite. A different position was taken by Krusel et al. (1997), who estimated that although technological progress is nonneutral, no break in its trend occurred in the 1980s. The deterioration in the position of the unskilled would have been the result of a sharp increase in the rate of capital accumulation, especially of investments in high technology. I will focus my comment on the case of France. A first problem concerns the definition of unskilled labor. In the Sneessens model unskilled labor is imperfectly substitutable with skilled labor in the production process. Moreover the determination of wages follows different processes for the two kinds of labor. Thus the definition of unskilled labor must be related to the precise job it fills in the firm, and not to the characteristic of the person who fills this job (the number of years of schooling or experience) nor to his wages. More precisely, unskilled labor may concern people with a good education background, and skilled labor may sometimes receive a very low earning. In France, by this definition, it is difficult to evaluate the average wages and the employment of unskilled labor, and there are differences in the results. The reasons for this fact are that a few years are missing in labor surveys, and there are changes in the classification of jobs. National accounts give a narrow definition of unskilled labor that includes concern manual workers and clerks who fill jobs that are essentially unskilled. Of course, the boundary between a skilled and an unskilled manual worker is fuzzy. Likely in France the workers in mills, as investigated by Kathryn Shaw, would be considered skilled labor. Audric, Givord, and Prost (1999) estimated that the share of unskilled labor among wage earners decreased linearly from 22 percent to 16 percent over 1982 to 1992. Since 1992 the share has remained the same. In my own evaluations, the ratio of unskilled labor cost (which includes social compensation) to skilled labor cost increased in 1970 to 1982 and decreased afterward. One difficulty with the definition of unskilled labor is that wages, are not only related to the kind of job a worker fills but also to his levels of education and experience. In France the number of years of education of all workers has increased enormously since the 1970s. However, Cre´pon and Gianella (1999) show that the return in education has decreased over time for unskilled labor, and not for skilled labor. So we might attempt to view this in Sneessens’ terms


173

by concluding that unskilled positions in firms cannot efficiently use improvements in the human capital of workers whereas skilled positions can. A deterioration in the position of unskilled labor would result from this feature. In a work with Anne Saint Martin (1999) we built a general equilibrium model for France that is more complete than the one used by Sneessens. This model was not econometrically estimated but was calibrated to the 1974 to 1993 period. Our first finding was a low elasticity of substitution between skilled labor and capital, but a high elasticity of substitution with the unskilled labor. This conclusion is about the same as the one reached for the United States by Krusell and his colleagues. A second finding was that the efficiency of skilled labor increased by a higher rate in the 1980s than in the 1970s; the efficiency of unskilled labor remained the same throughout the 1980s. On the other hand, market power and the rate of markup of firms decreased a lot in the 1980s, and this rise in competition canceled the effects of the sharp increase in interest rates in that period. Next we found that the power of the insider workers in bargaining their wages increased in 1974 to 1983, and then decreased. Blanchard (1997, fig. 7) obtained a similar result for the shift of labor supply. Caballero and Hammour (1997) support this last result by noting that with a putty-clay technology it took some time for firms to adjust the increase in power of insiders by substituting capital for labor (over the whole period the average productivity of capital decreased at a steady rate). As the cost of unskilled labor is mostly determined by the legal minimum wage legislation, the decrease in the power of the insiders eased the decrease in the relative cost of skilled labor, which was good for the employment of these workers. Finally, for France we found that the return on education increased among skilled labor, so the efficiency of this factor increased and it was complementary with capital. We got the opposite results for unskilled labor, which experienced a sharp decrease in employment. Therefore the indications are that new technologies are complementary with skilled labor and substitute for unskilled labor. References Audric, S., P. Givord, and C. Prost. 2000. L’Emploi non qualifie´ et son couˆt. Revue Economique, vol. S1. Blanchard, O. J. 1997. The medium run. Brookings Papers on Economic Activity 2: 89– 158.

174

Jean-Pierre Laffargue

Caballero, R. J., and M. L. Hammour. 1998. Jobless growth: Appropriability, factor substitution, and unemployment. Carnegie-Rochester Conference Series on Public Policy 48: 51–94. Draper, N., and T. Manders. 1996. Structural changes in the demand for labour. Research Memo 128. Centraal Planbureau, The Hague. Krusell, P., L. Ohanian, J. V. Rios-Rull, and F. Violante. 1997. Capital-skill complementarity and inequality. Research Department Staff Report 236. Federal Reserve Bank of Minneapolis. Cre´pon, B., and C. Gianella. 1999. Wages inequalities in France 1969–1992: An application of quintile regression techniques. INSEE Working Paper G9905, Paris. Laffargue, J. P., and A. Saint Martin. 1999. Ine´galite´s, biais de progre`s technique et imperfections de marche´ en France, de 1974 a` 1993. Economie et Pre´vision 2–3: 138–39. Shadman-Mehta, F., and H. R. Sneessens. 1995. Skill demand and factor substitution. CEPR Discussion Paper 1279, London.

5

Technical Change and the Structure of Employment and Wages: A Survey of the Microeconometric Evidence Lucy Chennells and John Van Reenen

5.1

Introduction

The effect of the development of tools on human activity has long been a principal concern for students of social behavior. Marx viewed the development of the productive means as the key evolutionary force in human history. The identity of the dominant class was determined by their ability to best muster the development of technology. In neoclassical economics, technological progress is also regarded as the driving force behind economic growth, a notion that is reinforced by endogenous growth theory. Given its role in economic growth, technical progress leads to higher standards of living on average. But how are the benefits of technical progress distributed across society? In the past many commentators worried that technology could lead to a ‘‘de-skilling’’ of workers. The pin factory symbolized the destruction of skilled artisans and their replacement by workers who were required only to perform the most menial repetitive tasks (Braverman 1974; Edwards 1979). More recently, however, debates by economists have focused on whether modern technologies are generally biased toward more skilled workers. The participants are particularly vocal in the debate over the causes of the increasing inequality of wages and employment between the skilled and the unskilled. Although closely related to it, the existence of skill-biased technical change does not provide the explanation for recent changes in the wage and employment structure. To demonstrate that technology is biased toward more skilled labor is not sufficient, and some would argue not even necessary, to establish technical change as the dominant explanation for increases in inequality. We also have to consider the supply of skills, for example.

176

Lucy Chennells and John Van Reenen

In this chapter we seek to survey econometric work that analyses the association of observable measures of technology with skills, wages, and employment. Our focus is mainly at the enterprise level, but we also consider some studies at the industry and individual level. The survey attempts to be comprehensive, but is limited to English language papers up to January 1999 and is heavily biased toward publications in economic journals. Macroeconometric studies and case studies1 are outside the scope of this survey. We seek to identify empirical regularities and also to evaluate the main methodologies critically. This is intended to help point the direction for future work in this rapidly growing area. The plan of the chapter is as follows. Section 5.2 briefly discusses some theory that implicitly or explicitly forms the background of the empirical studies. Section 5.3 discusses empirical problems with implementing the theory. Section 5.4 discusses the results of the papers explicitly, and section 5.5 draws some conclusions. 5.2

Theoretical Guide2

The Skill Bias of Technical Change We start with a general framework based within the context of a neoclassical model of production. For simplicity we consider the case of three variable factors (skilled labor, unskilled labor, and materials) and two quasi-fixed factors (physical capital, denoted by K, and ‘‘technological capital,’’ denoted by R). Consider a quasi-fixed translog cost function: X X X X ln C ¼ a 0 þ ahi Dh ln wi þ bij ln wi ln wj h

i¼B; W; M

þ bq ln Q þ

X

i¼B; W; M j¼B; W; M

biq ln wi ln Q þ bk ln K

i¼B; W; M

þ

X i¼B; W; M

biK ln wi ln K þ bR ln R þ

X

b iR ln wi ln R;

ð1Þ

i¼B; W; M

where C are the variable costs (blue-collar labor—B; white collar labor—W; and materials—M). The a parameters reflect own price effects. We allow these to differ in different ‘‘units,’’ indexed by Dh (D ¼ 1 if in unit h, etc.). For example, we might allow the own price effects to vary in different industries.

Technical Change and the Structure of Employment and Wages

177

The b parameters measure the effect on total cost of the other factor prices (w), the log of plant output (Q), technology (R), and the capital stock (K ). Since cost is homogeneous of degree one in prices, there are a series of restrictions as follows: X X X X b ij ¼ b ij ¼ bij j¼B; W; M

i¼B; W; M

X

¼

i¼B; W; M j¼B; W; M

b iR ¼

i¼B; W; M

X

biK :

ð2Þ

i¼B; W; M

These allow equation (1) to be normalized by one of the factor prices. Taking the materials price (wM ) as the unit of normalization, we obtain a normalized translog cost function where costs (relative to materials price) are a function of the relative prices, output, capital, technology, and their interactions. From Shephard’s lemma, the cost share si for input i is given as: Unskilled workers X

S B ¼ aB þ

biB ln

i¼B; W

wi þ b Bq ln Q þ b BK ln K þ bBR ln R: wm

ð3aÞ

Skilled workers S W ¼ aW þ

X

b iW ln

i¼B; W

wi þ bWq ln Q þ bWK ln K þ bWR ln R: wm

ð3bÞ

Note that the materials equation has been dropped because the cost shares sum to unity. We can test for homotheticity of the structure of production (i.e., that the cost shares depend on the quasi-fixed factor intensity, not the level of output) by imposing the following restrictions: b iq ¼ ðbiR þ b iK Þ;

i ¼ B; W:

If these can be accepted, the cost share equations simplify as follows: Unskilled workers S B ¼ aB þ

X i¼B; W

biB ln

wi K R þ b BK ln þ b BR ln : Q Q wm

ð4aÞ

178


Skilled workers SW ¼ aW þ

X

biW ln

i¼B; W

wi K R þ bWK ln þ bWR ln : Q Q wm

ð4bÞ

The elasticities of substitution and complementarity can now be calculated. In terms of the technology variable, if the coefficients bWR > 0 and b BR > 0, we would generally say that technology is labor biased. If b WR > 0 and bBR < 0, then technology is skill biased. In fact things are more complicated than this as, formally speaking, the Allen elasticity of substitution will also depend on the crosscorrelation between the quasi-fixed factors (see Brown and Christensen 1981 for a full treatment). The formulation is often further simplified using value added (VA) rather than output. In this case the dependent variable is the share of skilled labor in the wage bill, and the factor demand equation is simply: Skilled workers SW ¼ aW þ b W ln

wW K R þ b WR ln : þ b WK ln VA VA wB

ð5Þ

Again, skill-biased technical change would be indicated by a positive coefficient on bWR . Versions of structure (5) are very common in the literature. It seems a natural one given the difficulties in accurately measuring a cost of physical or technological capital (especially one that varies exogenously across microeconomic units). Sometimes the physical capital factor is allowed to be variable and only the technological component is fixed (e.g., Duguet and Greenan 1997). Many researchers have estimated (5) in employment shares rather than cost shares. Although less appropriate from a theoretical point of view, this clearly has the advantage that it allows a statistical decomposition of the effects of technology into a relative wage component and a relative employment component. This is only a framework for organizing our thoughts over the effects of technology in a well-known neoclassical framework. Other models suggest different rationalizations for the correlation of technology with cost shares. For example, the neoclassical model here takes factor prices as exogenous, which is clearly a questionable assumption since wage-setting is not conducted in a competitive spot


179

market. Models of bargaining would suggest that workers may be able to ‘‘capture’’ some of the rents from innovation. If skilled workers are more able to do this than unskilled workers (e.g., because of higher turnover costs associated with more skilled employees), then the technology–cost share correlation could be driven by relative wage movements rather than relative employment movements. This underlines the importance of analyzing movements in factor prices and quantities. The literature on the effects of technology on wages has been primarily motivated by attempts to assess the productivity effects of computers on highly skilled workers. Abstracting from nonpecuniary aspects of jobs, a competitive labor market in equilibrium would only have one wage for each skill type, so the underlying model behind these correlations is not entirely clear. The rent-sharing story is not the only potential reason for finding higher wages in firms with relatively intensive R&D expenditures. Another explanation could be related to employment contracts. If part of a worker’s wage is tied to the performance of the firm (e.g., to reduce shirking), then workers will demand a higher mean compensation in a more risky environment (assuming that R&D intensive firms face, ceteris paribus, a higher variance in their performance over time than non-R&D intensive firms). Garen (1994) has found evidence for this in his examination of managerial remuneration. The impact of technology on labor demand can be derived from the structure outlined above. One problem with this, of course, is that much of the effect of innovation might derive from increased output, which implies estimating the production function directly. In fact most researchers have tended to estimate simpler equations of employment, based on aggregating across all workers and estimating employment growth equations. There are, of course, serious difficulties in extrapolating results from the micro-level to produce macro-level implications. We have focused on the demand side, but the equilibrium effects of technological change will also depend on what is happening in other areas of the economy, and in particular to the supply of more skilled labor. Furthermore reallocations of output and employment will occur within and between sectors, tending to complicate the aggregate effects. The microeconometric evidence is only a small part of the story, and researchers should resist extrapolating too much from these partial equilibrium results.

180


Skill Bias and Unemployment In this section we consider what the implications of our model of skill-biased technical change are for unemployment and jobs. There are a great number of complex interactions between innovation and employment, but we begin with what we think is the most important route. If technology is skill biased, an exogenous increase in the stock of technological capital (a ‘‘technology shock’’) will increase the demand for skilled labor relative to unskilled labor. As the demand curve shifts out, in equilibrium there is both a rise in the relative wages and the relative employment of the more skilled group. Note that there is no unemployment in this model since the labor market clears. Now consider introducing some institutional limits to how far the wages of less-skilled workers can fall. These could arise because of minimum welfare levels, minimum wages, trade unions, or efficiency wage considerations. These institutional limits would produce a smaller increase in wage inequality, but would be accompanied by some unemployment for unskilled workers. This is not a new idea. In the last twenty years, this contrast between price and quantity adjustment has been evident in the labor markets of the industrialized countries. In the flexible labor market of the United States, wage inequality has increased, and unemployment has remained stable. In the relatively inflexible labor markets of Europe (outside the United Kingdom), wage inequality has been stable but unemployment has increased dramatically. Paul Krugman (1996) has aptly described U.S. inequality and European unemployment as ‘‘two sides of the same coin.’’ The debate on these matters is fierce. As noted in the introduction, the existence of skill-biased technical change and the question of whether technology is responsible for recent labor market trends are related, but quite distinct, analytical issues. Explaining recent history is a far harder task than simply understanding skill bias. This is not least because of strong disagreement on the appropriate model of the labor market. There are three key questions to be addressed: 1. Has the demand for skilled workers outstripped the supply of skilled workers? Or more accurately, has the demand/supply gap become greater over time?


181

2. If demand has accelerated relative to supply, is this due to technical change or some other factor, such as increased trade with less developed countries? 3. If the answer to both 1 and 2 is yes, how much of the change in unemployment and inequality can be accounted for by technological factors? Has the Demand for Skilled Workers Outstripped the Supply of Skilled Workers? Katz and Murphy (1992) and Autor, Katz, and Krueger (1998) try to date the timing of the increase in demand for skills in the United States. They use a weighted average of the growth of relative wages and employment, assuming that the labor market is in equilibrium with no unemployment. Given an assumption over the degree of substitutability between the skilled and the unskilled, it is possible to use a CES production function to estimate the relative demand shift. It is very hard, however, to date precisely the timing of the acceleration in demand, although both authors argue it exists (see Mishel and Schmitt 1996). More general methodologies have been proposed to take into account the unemployment in Europe and elsewhere. Nickell and Bell (1995), Jackman et al. (1996), and Manacorda and Manning (1997) argue that there has been relatively little increase in mismatch outside the United Kingdom and United States and that most of the increase in European unemployment has other roots. Has the Demand Change Been due to Technical Change? There is greater agreement that to the extent that demand has shifted toward the skilled, this is due to technology rather than trade. The methodologies used to reach this conclusion are usually based on the fact that most of the change in skills has been a within-industry phenomenon (see Berman, Bound, and Machin 1998 for more discussion of this debate). How Much Can Technology Account For? This question needs a full general equilibrium analysis, which has rarely been attempted. Back of the envelope calculations in Machin and Van Reenen (1998) suggest that technological factors alone can only account for a third or less of the changes in the United States and United Kingdom, but far more outside these two countries.

182


Technology, Homogenous Labor, and Employment The debate of the previous section is a crucial one for policy makers. Yet there is another strand in the literature that asks whether technology is responsible for falls in jobs even when it is not skill biased. Although a great deal has been written on this topic, the literature and the surrounding policy debate are littered with confusions. Information and communication technologies (ICTs) have diffused rapidly in Europe over the last twenty years, and unemployment has also risen. The temptation is strong to suggest that there is a causal link between the two. Yet waves of technology have passed over Europe in the past without creating persistent and structural unemployment. The debate over technological unemployment, on the other hand, has proved persistent. Similar arguments were being made in the 1960s over the introduction of automation, while in the 1930s Lord Kaldor (1932) commented (in a paper relating to the unemployment in the Great Depression): Today there is scarcely any political or journalistic observer of world affairs who does not attribute to the rapid growth of technical improvements one of the major causes of the present trouble.

Yet the fact remains that an examination of unemployment over very long periods shows no upward trend, despite the presence of technical change for several hundred years. It is possible that technology has a temporary destabilizing effect on employment, but it is difficult to believe that it is the major cause of the recent rise in European unemployment levels. Only technology combined with something else—such as wage rigidity—could be part of the cause. What can economic theory tell us about the likely effects of technical change on employment (see Spiezia and Vivarelli, Chapter 3 in this volume)? One form of technological change to consider is laboraugmenting process innovations. This case has been explored thoroughly in the literature. There are essentially two forces at work. For a given level of output, this type of technical change means that employment must fall, since the same output can be produced with a lower level of inputs. To offset this, however, is the fact that output will increase as prices fall because costs have fallen. This is the primary ‘‘compensation mechanism’’ of technical change. It means that examining the impact of technology on output (the production func-


183

tion relationship) is fundamental to understanding the effects of technology on employment. In the appendix we consider a simple model that shows how the effects of technical change work. This model leads us to focus on the following factors affecting the technology-jobs relationship: 1. Price elasticity of product demand. The greater is the sensitivity of consumers to price changes, the more likely it is that an innovation will raise employment. The higher is the price elasticity, the greater is the increase in output generated by an innovation. 2. Substitution of capital for labor. The easier it is to substitute, the more likely it is there will be positive employment effects of labor augmenting technical change, since labor is now relatively cheaper than capital and the firm will substitute into labor. The opposite is true for capital-augmenting technical change. 3. Monopoly power. If the firm has some degree of market power, not all of the reduction in cost will be passed on in the form of lower prices. This will blunt the output expansion effect and make positive employment effects less likely. Generalizations of the model lead to the consideration of further possible effects: 4. Market share effects. If the innovation does not diffuse immediately throughout the industry, the firm will have a cost advantage and so will tend to expand at the expense of its rivals. This will mean larger effects at the firm level in the short run. It also means that researchers should be careful in generalizing from the micro results to the economy level. 5. Union effects. If some of the efficiency gains from innovation are captured by unions in the form of higher wages (or reduced effort, etc.), this will also blunt the output expansion effects. The results are uncertain if the union also bargains over the wage (see Ulph and Ulph 1994). 6. Product innovation. Product innovations will tend to have stronger output expansion effects and are therefore more likely to result in employment increases (see Katsoulacos 1986 for a fuller analysis). 7. Economies of scale. These will tend to magnify the positive employment effects.

184

5.3


Econometric Models

We focus on the econometric problems of fixed effects, endogeneity, and measurement. Consider the basic equation to be estimated as the stochastic form of equation (5): SW ¼ aW þ b W ln

wW K R þ b WR ln þ u; þ b WK ln VA VA wB

ð6Þ

where u represents a stochastic error term. This could be justified by allowing for measurement error or optimization mistakes. It is unlikely, however, that this error term is uncorrelated with the other right-hand side variables. For example, some firms may have dynamic managers who employ both highly qualified workers and high-quality technology. Controlling for such fixed effects is important, so researchers often estimate the equation in differences (or by including dummies if the time series is long enough): DSW ¼ b W D ln

wW K R þ b WR D ln þ t þ e; þ bWK D ln VA VA wB

ð7Þ

where t denotes time dummies, and e the error term. Unfortunately, estimating this type of model usually requires panel data, which is rare in firm-level work. This is one reason why most research has concentrated until recently on the industry level. A second fundamental problem is dealing with the issue of endogeneity. Even when unobserved heteroegeneity is removed, firms might still change their technology in response to a change in the makeup of skills available, and not vice versa. If the ‘‘technological’’ factor were fixed in the short run, this would not be an issue. This may be more plausible for R&D than for other technology proxies (e.g., computer use). The use of longer differences (i.e., to mitigate measurement error, etc.) will actually exacerbate these problems of endogeneity. The only solution is to develop instrumental variables to deal with the fact that the technology and the skills decisions are being taken simultaneously. Unfortunately, such instruments are not easy to find, and researchers have been reluctant to take the standard econometric approach of using lags as instruments because of concerns over the low level of correlation between the change in the endogenous variable and the lagged instruments. A related issue is the interpretation of the coefficients on the relative wage terms. These terms are directly involved in the con-


185

struction of the dependent variable. It is doubtful how much of the inter-firm or inter-industry variation in relative wages is due to changes in the price of labor, rather than due to changes in the quality mix of labor, which is imperfectly captured by observable skill. An intellectually respectable solution would be to use credible instruments for relative wages. A commonly encountered shortcut in the literature is to argue that time dummies will capture the real variation in wages, and to include these instead of the relative wage terms. The third and perhaps the most basic issue, however, is how to measure technology. This is a serious problem, since the technology input is a far more nebulous concept than the input of, say, labor, which itself is difficult enough to measure. The traditional approach is simply to use time trends. The problem here, of course, is that the trends are likely to be picking up a lot more than just technical change, such as unmeasured price movements, changing demand conditions, and cost shocks. These criticisms are well known from the debate on how suitable total factor productivity (TFP) is as a measure of technology. Researchers have turned to a variety of alternatives in seeking observable measures of technology. We can distinguish crudely among three types of measure: inputs into the knowledge production function, outputs from the knowledge production function, and subsequent diffusion of these outputs throughout the economy.3 Inputs are generally measured by R&D activities. R&D expenditure has the advantage that it is measured in many databases over time, across countries, and in a reasonably standard way 4 —at least by comparison with the alternatives. Also R&D is measured in terms of a unit of currency, which provides a natural weighting, whereas other innovative measures are more qualitative. There are several disadvantages in using R&D as the technology measure. First, it is only an input into creating knowledge. A firm might invest in large amounts of R&D without receiving any benefit from it, if the R&D does not produce any outputs. These outputs could be either in the form of innovations, or in acquiring a greater ability to learn from other firms’ innovations (sometimes called ‘‘absorptive capacity’’). Second, there are long and unknown variable lags between the act of investing in R&D and reaping useful output from it.5 Third, the transmission mechanisms for knowledge to spill over from one firm to another are also poorly understood. For ex-

186


ample, the R&D spending of Intel dramatically affected the development of computer technologies used by other firms all over the world, but the process by which this knowledge was absorbed by other firms remains unclear, and is rarely addressed in the literature. Patents are a widely available and standard way to measure the outputs of knowledge. The problem with patents is that a large number of them have very low value, and there is no obvious method of weighting them to take account of this.6 In some countries expert innovation surveys exist, which can be viewed as a method of cutting off the lower tail of low-value patents. The U.K. Science Policy Research Unit (SPRU) Innovation Survey is a good example of this, since industry experts were asked to list the most important innovations in their field in order to weed out the innovations with little value. Output measures such as patents suffer from some of the problems of R&D—such as spillovers and variable time lags—but add some problems—such as the difficulties of dealing with count data. Diffusion measures seem to be closely related to what is usually thought of as ‘‘technology.’’ A common example would be the use of computers in a firm. Researchers are usually faced with the problem of which technologies to include: what sort of computers (word processors, mainframes), whether also to include production-based technologies (lasers, robots, NC, CADCAM), how to weight the usage (the proportion of people using the computer is a common form of weighting). The most satisfactory method seems to be constructing the capital stock of information technology, although separating out this component becomes increasingly difficult as it becomes hardwired into more and more modern organizations. Measuring the diffusion of a particular technology is difficult in any time series context, since the passage of time changes the significance of using a particular type of technology. For example, in 1978 an indicator of whether a computer was extensively used within the firm gave a very different signal to that same indicator in 1998. Diffusion-based measures of technology are more likely to suffer more from simultaneity problems than, say, R&D. Current changes to a firm’s environment will have less of an effect on something like R&D than on the decision whether or not to postpone investing in more computers. This is primarily because of the greater adjustment costs attached to restructuring or canceling a research programme than in purchasing a new piece of hardware.


187

The measurement of skills is a less controversial issue, and the problems associated with it are well known. There are two main methods of measuring skills. Perhaps the most common in the literature is to use an indicator of occupation, often simply by dividing the population into manual (production) and nonmanual (nonproduction) workers. Such categorizations can be criticized, since many nonmanual occupations require very low levels of skill. Education-based measures are more closely tied to ideas of levels of human capital, but face the problem that even highly educated workers may not be employed doing very skillful jobs. Some authors have developed measures based on job content, where an occupation is broken down into different levels of task complexity (see Wolff 1996). In studies that have compared them, these measures all tend to be highly correlated across industries (e.g., Berman et al. 1998). Nevertheless, there are real worries that the categories chosen are not comparable over time and across countries. Another measurement issue relates to double counting. Innovative activities tend to be labor intensive and involve skilled workers. R&D is a good example, since typically about half of all R&D is staff costs and only 10 percent capital costs. This will automatically generate a positive correlation between the level of skilled (i.e., better paid) employees and the level of R&D. Correcting for this ‘‘double counting’’ has been found to be important in the productivity literature. The problem reappears here in many guises. Finally, there are issues to be grouped under ‘‘selectivity.’’ The usual problems of sample response and survivor bias are encountered, but there are particular problems relating to the use of R&D expenditure. In most European countries disclosure in company accounts of the amount of R&D carried out is not compulsory. This means that researchers have to be aware that excluding, or setting to zero, those companies that do not disclose any R&D is likely to introduce a selectivity bias. 5.4

Results

The papers are divided into three topics. Table 5.1 contains papers analyzing the effects of technology on the skill structure. Here we concentrate on attempts to correlate technology with the proportion of skilled workers in cost shares or employment shares. The second table examines the evidence of correlation between wages and tech-

188

Table 5.1 Effect of technology on skill structure Study

Method

Data

Autor, Katz, and Krueger (1998)

Mainly longdifferenced changes in skilled share of wage bill regressions (education and occupation)

1960–70, 1970–80 (Census); 1980–90, 1990–95 (CPS); 149 U.S. industries (manufacturing and nonmanufacturing)

Bartel and Lichtenberg (1987)

Estimation of restricted variable cost function

Bartel and Sicherman (1998)

Effects of industry technology on individual training

Proxy for technology

Controls

Result

Computer use in 1984, 1984–93 change in computer use; R&D; TFP

Initial education level; capital, output, imports, exports, outsourcing

High-tech industries have faster upgrading of skills; effect has grown stronger over time (bigger in 1970s and 1980s)

61 manufacturing industries in 1960, 1970, 1980

Proxies for age of capital stock

Age, education, gender cells

Positive association of younger capital with skill proportion

National Longitudinal Survey of Youth in manufacturing industries 1987–92; males 14–21 in 1979

Cross-sectional industry level measures of computer investment, TFP, R&D intensity; introduction of new production processes, patents used

Schooling dummies, experience, race, MSA, part-time, gender, tenure, firm size, marital status, union, time dummies, industry level union coverage, unemployment, job creation and destruction

Workers more likely to get training as technological progress increases. More educated workers are more likely to get training but this ‘‘educationtraining gap’’ narrows as rate of technological progress increases

A. Industry level


Change in proportion of nonproduction workers in wage bill and employment

ASM: 4 digit manufacturing industries 1979–89

R&D intensity, computer investment

Fixed capital, valueadded, time dummies

Positive association of technology proxies with change in skill share

Berndt, Morrison, and Rosenblum (1992)

Within-group estimates of proportion of total hours of a skill group regressed against various capital intensities; also by industry separately

2 digit U.S. SIC manufacturing 1968– 86 merging BEA, BLS and CPS

High-tech office and information equipment as proportion of total industry capital stock ðOF=KÞ.

Blue collar/white collar; education cells for production and nonproduction workers

Positive effect of high-tech capital on white-collar hours; within blue-collar occupations evidence of educational upgrading, less clear within white-collar occupations (middle education groups lose out most)

Gera, Gu, and Lin (1998)

Within-groups estimates of wage bill and employment shares of (a) higher occupational groups; (b) knowledge workers (WolffBaumol)

29 Canadian industries in manufacturing and nonmanufacturing; 1981–94: 1981 (Survey of Work History), 1986–90 (Labor market activity survey); 1993–94 (Survey of Labor and Income Dynamics);

R&D stock; patent stock; age of capital stock; log of TFP

Capital, output, time trend

All technology variables tend to be positive and significant except for TFP


Berman, Bound, and Griliches (1994)

189

190

Table 5.1 (continued) Controls

Result

1909–19; 1959–69; 1969–79; 1979–89 U.S. Census of Manufactures: 450 industries (256 in 1909–19)

Capital

Output

Positive effect of capital–output ratio in all years

Wage bill share of higher and lower grade professionals, administrators and officials; regress industry fixed effects against technology measures

35 French industries (all sectors) 1982–93

Computer use, industrial technologies (e.g., robots, NC machine tools, telewatching)

Capital, output, time dummies

Industry fixed effects positively correlated with computer use, but negatively correlated with ‘‘industrial technologies’’

Hansson (1997)

Change in proportion of educated workers in employment and wage bill regressions

16 Swedish manufacturing industries; 2 long differences 1970–85; 1985–90

R&D intensity; share of technicians

K, Y, time dummy

Both technology measures have positive and significant effects

Machin and Van Reenen (1998)

Changes in wage bill (and employment) share of skilled employees (occupation and education)

1970–89; 2 digit manufacturing industries in Denmark, France, Germany, Japan, Sweden, U.K., U.S. (15)

R&D intensity

Capital, value added, time dummies, imports (OECD and nonOECD)

Skill upgrading faster in high-R&D industries in all countries, no significant effect of trade variables

Method

Data

Goldin and Katz (1996)

Wage bill share of nonproduction workers

Goux and Maurin (1996)



Study

34 industries in three periods (1973–79, 1979–89, 1989–94); CPS with 1990 PUMS; BEA, NIPA

Computer and capital equipment, share of scientists and engineers

Dummy for computer industry; physical capital

Positive effect but no change over time in coefficients

Osterman (1986)

Change in employment after computer installations

20 industries; 1972, 1978

Total amount of main computer memory in industry

Total employment of clerks, nondata entry clerks, managers and others

Computers associated with falls in employment of clerks and managers (although LR effects of managers much smaller)

Wolff (1996)

Three skill variables: SC (substantive complexity), IS (interactive skills) and MS (complexity skills), rated for 12,000 jobs; follow occupations by industry

43 U.S. manufacturing industries; 1970–80; 1980–90

TFP; computing investment; R&D; % computer workers and engineers; K=L growth

Capital–output ratio; time dummies, union, imports, exports.

Positive relation of R&D and computers to growth of SC and IS

2 factor cost shares (white-collar and bluecollar labor, materials equation dropped); R&D ðt 1Þ exogenous. In cost share equations, SUR; symmetry and homotheticity imposed; no enterprise fixed effects

U.S. chemical firms with positive R&D for 5þ years; combined with plant-level data on production (LRD); 1976–88; 9,500 observations

Firm R&D broken down also by NSF applied product field and state; statewide and industrywide applied product R&D spillover pools also created

Relative factor prices (industrystate); fixed capital, number of plants in firm; industry (in share equation) and time dummies, dummy for plant slow-down, plant birth, spillovers

Firm R&D has a positive effect on both labor shares; only consistent effect of firm R&D in same product field as plant; mixed results on spillovers

B. Enterprise level Adams (1999)

191

Change in share of 5 educational groups


Mishel and Bernstein (1997)

192

Table 5.1 (continued) Method

Data

Aguirregabriria and AlonsoBorrega (2001)

Factor demand equations for 5 labor types, and 3 capital types (total, R&D, imported technology); first differences; GMM; attempts to control for selectivity through propensity matching

CBBE; balanced panel of 1,080 Spanish manufacturing firms, 1986–91

Bresnahan, Brynjolfsson, and Hitt (1998)

(a) Human capital acquisition equations, similar to Autor et al., including IT and physical capital controls (b) IT demand equations, controlling for skills and organization measures

IT information for the Fortune 1000, 1987– 94; survey of workplace organization on c. 380 firms, 1995–96; compustat data


Controls

Result

R&D, expenditure on technological capital— ‘‘successful innovations generated externally to the firm’’

Output, capacity utilization, whitecollar blue-collar wage differential, time dummies

No effect of R&D and stock of technological capital has an unskilled bias; but dummy for introduction of ‘‘technological capital’’ has strong negative effects on blue-collar workers; most change in downturns

IT capital stock, MIPs (measure of computing capacity), number of PCs

(a) Decentralization, industry dummies, changes in IT and capital per worker, in capital as a share of output, and in output (b) Skill, college proportions, workforce organization, sector, year, and occupation dummies

IT combined with organizational change increases relative demand for skilled workers, more so than IT alone. Output increases greater when increased IT occurs in firms with highly skilled workers and/or decentralized organization


Study

OLS regressions of change in share of skilled workers

(a) 402 British establishments from WIRS panel 1984–90; 6 occupational groups (b) 992 French establishments Enquete Reponse 1992–96; 5 occupational groups

(a) Percentage of workers affected by microelectronics, computer introduction, industry-level computer use (b) Computer use

(a) Organizational change, total employment growth, 1984 plant characteristics (b) Organizational change (delayering), total employment growth, 1992 plant characteristics

(a) Technology significant effects (negative effect on least skilled and positive effect on most skilled); organizational change has negative significant effect on least skilled (b) No effect of technology in panel (but some in cross section); OC has significant negative effects on least skilled

Doms, Dunne, and Troske (1997)

OLS regressions in cross section and time series; dependent variables include % skills (education and occupation); wages by skill group; growth regressed on 1993 characteristics and changes 1988–93

U.S. plants in SIC 34–38; 1988 and 1993 Survey of Manufacturing Technology (SMT); Worker-Establishment Characteristics Database in 1988 and 1993 (WECD ¼ LRD þ Census); 353 plants in 1988; c. 3,260 in changes; also ASM

5 dummies for numbers of different types of manufacturing technologies used in plant (e.g., CAD, NC, robots, lasers, networks, automatic systems); computer investment

Capital–output ratio, total employment, age, 2 digit industry and regional dummies; MSA dummy

In the crosssectional technology measures associated with higher proportion of skilled workers and higher wages; there is a positive effect of computer investment but no effect of manufacturing technologies


Caroli and Van Reenen (1999)

193

194


Data

Duguet and Greenan (1997)

(a) Probits of longdifferenced innovations data (for 5 types of innovation) (b) Trans-log cost share equations in long differences

Panel of 4,954 French manufacturing firms, 1986 and 1991

Greenan, Mairesse, and Topiol-Bensaid (2001)

Essentially crosssectional and 4 year differences of the occupational structure (aggregated into 4 groups)

c. 11,000 observations on French firm-years in three time periods (1986, 1990, 1996)— whole economy; from combining ESE, BIC; correct for double counting using measure


Controls

Result

Five types of innovation: product improvement; new product; product imitation; process breakthrough; and process improvement

(a) Firm size, market share, diversification, industry dummies, cost shares of ‘‘conception’’ and ‘‘execution’’ workers (b) Both types of workers, capital, production volume

Skill bias in favor of ‘‘conception’’ labor. ‘‘Execution’’ labor a stronger substitute for capital. Reduction in demand for execution labor largely due to new product innovation

(a) IT capital from firms balance sheet: Basically office and computing (including photocopying equipment, etc.) (b) Use numbers of IT workers— computer staff/ electronics, specialists/ research, staff/ analysis staff

Capital, valueadded, sector dummies

Strong correlations in cross section, but only the negative effect of IT on lowest-skilled group robust in time series


Study

Change in nonproduction workers wage bill share and employment share; short and long differences; OLS and GMM; pooled and by 2 digit industry

U.S. manufacturing plants—LRD 1972– 88; about 1,820 plants in SMT/LRD merged data; 11,000þ in larger data set

Change in R&D stock; dummy for adoption of (a) IT (b) production technologies (1988 SMT—17 types of new technology)

Equipment, structures, ownership, industry, dummies for regions, time, region time

R&D significant and positive effect on skill share quantitatively important. In accounting for secular change in skill share; IT also positive; production technologies negatively correlated (latter results not robust)

Kaiser (1998)

Ordered probit for expected net employment change (3 categories) for 4 groups of skills 1995– 97

German firms in ‘‘business-related service industries’’ 1995; 1,059 firms

IT investment as a share of total investment

Total investment per capita number of employees in each skill group, sales, credit rationing, export status

Positive and significant effect on most-skilled group; negative and significant effect on least-skilled group

Lynch and Osterman (1989)

Labor demand regressions for 10 occupational groups

1 company (Bell telephone company) 1980–85; year–state– occupation cells

Change in technology of office switching equipment

Machin (1996)

Employment change for 6 occupational groups

U.K. WIRS panel data 1984–90; 402 plants; all industries

Introduction of any computers, 1984–90

Positive for technical and professional employees Dummy for fall in total employment, any manuals 1984

195

Positive for most skilled groups (managers and technicians) and negative for least skilled group (unskilled manuals)


Dunne, Haltiwanger, and Troske (1997)

196


Data

Siegel (1998)

Employment share and wage bill share regressions

79 Long Island (NY) manufacturing firms; 1987–90; 6 skill groups

Vainioma¨ki (1998)

Wage bill and employment share regressions of nonproduction workers and educational groups; long differences 1985– 90; 1990–94

Finnish manufacturing establishments (sample size varies from c. 500–c. 1300 plants;) 1985–94; Census of Manufacturing; linked employee data; SMCT; R&D surveys


Controls

Result

Introduction of various kinds of manufacturing technologies, R&D intensity

Age, size

Positive effects of technology on skill composition

(R&D/sales) in levels and changes; introduction of advanced manufacturing technology in 1990; computer investment 1990

Export share, capital, output, industrial outsourcing, ownership, industry, and regional dummies

Change in R&D intensity has positive effects on wage bill share; other measures have unstable effects across different specifications


Study


197

nology. These could be driven by skill-biased technical change: the average wage in a plant could reflect movement in the distribution of employment of different types of workers, for example. To complete the survey, we examine correlations of employment with technology. The structure of each of the three tables is the same. Studies are divided according the level of aggregation used: first the industry level, then the level of the enterprise, and finally the individual. Skills There is a preponderance of studies from the United States in all of the tables, and in particular, studies at the industry level. A key paper in this area is Berman, Bound, and Griliches (1994), who estimate a version of equation (7) on four digit U.S. manufacturing data in long differences. They use R&D expenditures and computer investment as their measures of technical change. These technological proxies are found to have positive and significant effects on the growth in the wage bill share of nonproduction workers, the computer variable accounting for about a third of the increase in the share. Autor, Katz, and Krueger extend this study over a longer time period (from the 1940s to early 1990s) and to nonmanufacturing. They corroborated the importance of technical change (especially computer use) in accounting for the increase in skilled workers as a proportion of the wage bill. Machin and Van Reenen (1998) extend the U.S. results to the manufacturing sectors of six other countries (Denmark, France, Germany, Japan, Sweden, and United Kingdom). They find results that broadly support the importance of skill bias across all countries using their measure of R&D intensity. Other papers with country-specific analyses have also tended to find evidence of skill-biased technical change (e.g., Gera et al. 1998 for Canada; Hannson 1997 for Sweden), but Goux and Maurin (1997) are more sceptical about its importance in France. Aggregation may be a serious problem for these industry studies, so panel B considers the analyses based at the level of the enterprise. There are a greater range of countries represented in these studies, as well as a larger number of alternative proxies for technical change. The overall results still suggest the presence of skill-biased technological change. The Longitudinal Research Dataset (LRD), a manu-

198


facturing panel dataset for the population of larger plants, has been a prime resource in the United States.7 Doms et al. (1997) and Dunne et al. (1997) both find evidence of skill bias, but Doms et al. (1997) stress that they cannot find evidence for significant effects in the time series dimension of their data. This is a worrying result, for it suggests that some other unmeasured factor may be driving both skills and technology. On the other hand, measurement error issues and the fact that they use counts of production technologies (rather than computer usage) might account for their results. Indeed, when Doms et al. (1997) use computer investment as an alternative measure, they find that this is associated with the growth of skilled workers, even in the time series dimension. Haskel and Heden (1997) use the equivalent large dataset of British plants and also find evidence of a positive impact of computers on the growth of skill intensity in the two years where computer investment data is available. Adams (1997) focuses on firms mainly operating in the chemical industry. In his careful study he finds that firm R&D in the same product field as that produced by the plant is associated with skill bias. He could not find consistent evidence for skill bias from total firm, state, or industry R&D. Although Adams was able to control for detailed industry effects, he did not include plant fixed effects. Duguet and Greenan (1997) use an innovations survey to estimate cost share equations for a panel of French manufacturing firms, 1986 to 1991, in long differences. They find evidence for skill bias and argue that it comes primarily from the introduction of new products, although their results here are mixed. One problem with subjective innovations surveys is the comparability of the notion of innovation across different firms. An interesting extension, given the increasing availability of this type of innovation survey, would be to use the longitudinal aspect of the panel when the question is asked to the same firms in future. Machin (1996) uses the British Workplace Industrial Relations Survey (WIRS), panel 1984 to 1990, that contains information on the introduction of computers and also finds evidence for skill bias. Aguirregabriria and Alonso-Borrega (1997) use Spanish panel data and find effects of their measure of ‘‘technological capital,’’ but they find no robust effects of R&D. We end this subsection with three general comments. First, there does appear to be considerable support for the notion of skill-biased technical change across a range of studies, and these studies are usually (but not always) robust to controlling for fixed effects. Sec-


199

ond, there have been few attempts to find instrumental variables to deal with the potential endogeneity of technology. Candidates could include government-induced schemes to alter the incentives to accumulate technological capital (R&D tax credits, government grants, etc.8). Third, there are surprisingly few studies that try to analyze the mechanisms by which technological change translates into higher demand for skills (see Shaw, chapter 6, this volume). One mechanism is through organizational changes such as delayering, decentralization, and giving greater autonomy to workers. These organizational factors have been found to be important in the case study evidence and in the literature on the productivity paradox (investigating why computers have not raised measured productivity by as much as might have been expected). Some preliminary work suggests that this organizational restructuring could be the link between technology and labor demand (Bresnahan et al. 1998; Caroli and Van Reenen 2001). Wages We have included studies that examine both the level and structure of wages in table 5.2. Interestingly the majority of the studies here are carried out at the micro level (mainly on individuals). The authors of these studies have paid much more attention to the need to control for fixed effects and endogeneity. Dickens and Katz (1987) is an early piece of work that focused on factors correlated with interindustry wage premiums. They found that there were many observable industry-level factors associated with these premiums, including R&D intensity. The rise in wage inequality prompted some additional interest in this type of work in the United States Inspired by the well-known analysis by Mincer (1991) of time series data, Allen (1996) calculated that the increase in the rate of return to schooling between 1979 and 1989 was most dramatic in industries with greater R&D intensities. This is similar to Bartel and Lichtenberg’s (1991) finding that the more recent vintages of the capital stock are positively related to educational premiums. At the enterprise level, most studies find a positive correlation of technology with wages (e.g., Dunne and Schmitz 1995 for United States; Martinez-Ros 1998 for Spain; Casavola et al. 1996 for Italy; Machin et al. 1998 for United Kingdom; Tan and Batra 1997 for Columbia, Mexico, and Taiwan). There is a less clear pattern that

200

Table 5.2 Effect of technology on wages Study

Method

Data

Allen (1996)

(a) Wage equation by industry; regress levels and changes of coefficients against industry technology (b) 32 education– experience–gender cells regressed against industry dummies; dummies regressed against technology

Individual-level data from U.S. CPS, 1979 and 1989, combined with industry-level data on technology (about 39 manufacturing and nonmanufacturing industries)

Bartel and Lichtenberg (1991)

Pooled crossindustry wage equations (sometimes control for industry fixed effects)

1960, 1970, 1980 U.S. CPS; manufacturing; 70 age–education– gender cells by industry (35)


Controls

Result

R&D intensity, growth in capital– labor ratio; age of capital; TFP; % of scientists and engineers

Education, experience, parttime, gender, race, SMSA

Levels and changes in returns to schooling and education significantly related to R&D, high-tech capital, K=L acceleration.

Age of equipment; computing investment/output; R&D to sales

Union, plant age, year dummies, K=L, output growth, profits

Younger technologies have higher wages; larger educational premium in industries using younger technology

A. Industry level


National Longitudinal Survey of Youth (NLSY) in manufacturing industries 1979 and 1993 14–21 in 1979; approx. 50 industries

Cross-sectional industry-level measures of computer investment, TFP; % scientists and engineers; R&D intensity; introduction of new production processes, patents used

ATQT score, Schooling dummies, experience, race, MSA, part-time, gender, tenure, firm size, marital status, union, time dummies, industry union coverage, unemployment, job creation and destruction

Positive effect on earning and earnings premium of technical change; severely reduced in fixed effects estimation; argue that premia is due to sorting of high-ability workers into hightech industries

Dickens and Katz (1987)

Individual wage equations and industry dummies; regress dummies against industry variables

1983 U.S. CPS and industry-level measures of technology

R&D intensity

Education, experience, parttime, gender, race, SMSA

Positive impact of R&D, especially for non-union workers

Simultaneous system of wage and technology equations; earnings for skilled, semiskilled, and unskilled manual labor

c. 900 British plants in 1984 and 1990 Workplace Industrial Relations Surveys (WIRS); 100 plant panel.

Introduction of microelectronic technologies affecting manual workers (ATC); computer presence

Lagged size, union, single site, parttime, female, % manual, foreign, industry dummies. Gender and industry wages, instrumental variable for plant wages; industry R&D and patents, instrumental variable for ATC

OLS gives significant technology effects on wages; but disappears in IV results; in IV results get an effect of higher wages on technology adoption

B. Enterprise level Chennells and Van Reenen (1997)

201

Effects of industry technology on individual earnings and educational premium


Bartel and Sicherman (1999)

202


Data

Casavola, Gavosto, and Sestito (1996)

Average wages of (a) blue-collar workers and (b) white-collar workers; crosssectional regressions for each year

Private sector Italian firms (over 20,000 per year) 1986–90; INPS

Doms, Dunne, and Troske (1997)

See table 5.1B

See table 5.1B


Controls

Result

The share of intangible capital in total capital relative to industry average (includes patents, software, and advertising)

Age, size, share of white-collar workers, sales/L, value added/L, K=L, returns on investment, profits, severance fund/L, inventories, interest payments, growth of K, depreciation; industry (3 digit) and regional dummies

About 2–6% increase in wages for each group associated with technology measure

See table 5.1B

See table 5.1B

In the cross section, positive effects on wages but zero effect in the panel, regardless of measure of technology used


Study

Average wages of nonproduction, production workers and share of nonproduction workers, as a function of plant characteristics

Cross section of 6,909 U.S. plants in 1988, from the Survey of Manufacturing Technology, matched to Census of Manufactures data from 1987

Use of advanced computer-based machinery, nature of manufacturing at plant, average product price

Firm size, industry and regional dummies, multiplant indicator, age of plant, number of products produced

Plants using advanced technologies pay highest wages and employ largest proportion of skilled workers. Technology use reduces size-wage premiums up to 60%

Garen (1994)

Salary equation; interactions of payperformance sensitivity with industry R&D (as proxy for risk)

Cross section of 415 U.S. corporations, detailed CEO remuneration info

Industry average R&D

Total firm assets, beta, age, industry dummies

Positive and weakly significant effect of R&D on salary

Holthausen, Larcker, and Sloan (1995)

Regress ratio of ‘‘long-term compensation’’ (e.g., stock options) to total compensation against future industry patents

1982–84 confidential survey of divisional CEO compensation in U.S.

1987–90 industry patents from CHI

Market share, sales, industry union, Herfindahl index, industry dummies

Weak positive effect of patents on proportional longterm and total compensation


Dunne and Schmitz (1995)

203

204

Table 5.2 (continued) Proxy for technology

Result

Lagged R&D per worker

Capital, employment, time dummies, fixed effects

R&D-earnings elasticity significant for workers and directors; directors’ wage– R&D elasticity twice that of workers

Global Vantage firms from Italy, France, Britain, and Germany, 1982–90

Lagged R&D per worker

Capital–labor, time dummies, fixed effects, average industry wage

Positive effects, even after controlling for selectivity and fixed effects; strongest in U.K. and Germany.

Arellano-Bover (1995) IV procedure

Spanish panel data; 1,306 manufacturing firms, 1990–94

Firms surveyed on process and product innovation

Lagged wage, industry wage, market share, initial skills

Significant positive effect when firms do both product and process

Size–wage differentials for investing and noninvesting firms

500 firms in Columbia (1992), 5,070 in Mexico (1992), and 8,408 in Taiwan (1986)

Investment in R&D and know-how, exports and/or formal workplace training

Foreign ownership, firm age, single- or multiple-plant, industry dummies

Large positive effects of R&D and training for skilled workers, smaller or zero effects for unskilled

Method

Data

Machin, MenezesFilho, and Van Reenen (1998)

Four equations: average wages, directors pay, productivity and R&D disclosure; quasi-differenced GMM

U.K. Datastream companies 1983– 94; 660 firms, unbalanced panel

Machin and Van Reenen (1996)

Average wage as function of lagged R&D per worker; controls for R&D disclosure selectivity; fixed effects

Martinez-Ros (1998)

Tan and Batra (1997)


Controls

Study

Wage regressions by skill group (production vs. nonproduction; educational group)

Unbalanced panel of Finnish establishments; see table 5.1B

R&D, AMT, computers; see table 5.1B

Export share, capital, output, industrial outsourcing, ownership, industry and regional dummies

Technology–wage correlation unrobust: driven out by conditioning on worker characteristics or looking at changes

Van Reenen (1996)

Average firm wage regressions (first differenced GMM)— distributed lag of innovations

Unbalanced panel of 598 U.K. quoted firms, 1976–82

Count of major innovations (SPRU) at firm and industry level; firm-level patents (taken in U.S.)

Lagged wage, market share, capital–labor ratio, industry wages, industry unemployment, industry R&D, time dummies

Innovations have positive significant effect on average wages; no additional effect of patents; interpreted as a rent-sharing

Arabsheibani, Enami, and Marin (1996)

OLS wage regressions; selectivity adjusted (but no exclusion restrictions)

British Social Attitudes Survey

Computer use

Bell (1998)

Wage growth regressions between ages 23–33

British National Child Development Survey (all individuals born in March 1958), final sample about 1,000

Use of computer age 33 (1991)—no information on 1981 computer use

C. Individual level Positive effect but no different for skilled than unskilled workers Ability as measured by reading and maths scores

Positive computer effect robust to controls for ability


Vainioma¨ki (1998)

205

206


Data

Card, Kramarz, and Lemieux (1996)

Changes in (a) employment– population ratio; (b) average wages in a cell regressed against computer use % in cell; separately for men and women; WLS

Age–education cells; adult whites; in U.S. 225 cells (CPS: 1979 and 1989); in France 70 cells (EE: 1982 and 1989), in Canada 29 cells—1981 (SWH), 1988 (LMAS)

DiNardo and Pishcke (1997)

OLS earnings functions with dummy for computer use interaction with education

German Qualification and Career Survey 1979, 1985–86, 1991–92; c. 60,000 individuals


Controls

Result

Computer usage by cell; U.S. 1989 (CPS); France 1991 (EE); Canada 1989 (GSS)

Change in population share; initial wage instead of computer use as index of demand shock

Wages: significant positive in U.S.; insignificant in Canada; significant negative in France. Jobs: Strong positive in U.S.; insignificant in other countries

Use of computers, pencils, hand tools, telephones, and other tools

Years of schooling, experience, parttime, city, gender, married, civil servants; detailed occupation (up to 1,000)

12–18% premium in Germany; increasing over time; but also effects for other tools like pencils (which are stable over time). Authors conclude no causal effects


Study

OLS earnings equations with firm and individual characteristics, controls for individual dummies in some specifications

French TOTTO (1987 survey of new technology use) matched to individual and firm panel (3,694 individuals in the panel, with 8,192 observations), 1985–87

Time at which an individual started to use a new technology (e.g., computer); experience with computers

Education, tenure, firm skill shares, assets, exports, profits, size, occupation, industry, time dummies, parttime, experience

No linear effect of technology after controlling for fixed effects; evidence of an interaction of NT use with experience (quadratic)

Hildreth (1998)

OLS earnings functions with detailed plant and individual controls; imports and technology as IVs for profits

Matched establishmentworker data survey; 1994 U.K. manufacturing; 685 plants

Introduction of process and product innovation

Age, gender, union, education, occupation, workplace conditions, profits

Process innovations have bigger effects on estimates of rentsharing

Krueger (1993)

OLS earnings functions with dummy for computer use interaction with education

1984 and 1989 U.S. October CPS (c. 13,000 individuals); High School & Beyond survey (4,684)

Computer use (broken down by purpose—e.g., programming, email)

Education, experience, race, MSA, part-time, veteran, gender, marital status, union, occupation (8), region (3)

19–21% premium for computer use; higher for educated; effect higher in 1989 than 1984


Entorf and Kramarz (1997)

207

208

Table 5.2 (continued)

Industry-level R&D and equipment age

Age, gender, race, union membership, employment, education, experience

Positive effect of innovation on hourly wage, particularly in union sample. Training more likely for workers in innovative industries

607 Canadian employees in 60 firms, 1979

Access to a computer

Education, experience and tenure, supervisors per employee, capital stock per employee, age of capital stock

Positive effect of technology and wages accounts for the employer sizewage effect

118,320 individuals from the WECD (LRD and 1990 Census) and the 1988 SMT

Use of manufacturing technologies and computer investment in plant (not individual specific)

Size of plants, capital in plant, individual characteristics

No effects of plant computer presence conditional on plant characteristics

Data

Loh (1992)

OLS wage regressions, and training probit

U.S. Current Population survey (1983), matched to some industrylevel data

Reilly (1995)

OLS wage regressions investigating employer size-wage effect

Troske (1999)

OLS earnings equations with plant and worker characteristics


Result

Method

Note: IV ¼ instrumental variables.


Controls

Study


209

skilled workers receive a higher premium than unskilled workers, however. More worrying, these results are sensitive to various econometric tests. The study by Doms et al. (1997) finds that their positive effects of diffusion on wages disappear in their differenced models. Chennells and Van Reenen (1997) find that instrumenting the adoption of new technologies at the plant level with industrylevel measures of technological opportunity reduced the effect of technology to zero. There is a similar pattern in the individual data. Krueger (1993) found strong effects of computer use on wages. The computer– earnings correlations were significantly greater for educated workers. Similar observations have been made in other countries (e.g., Bell 1998 for United Kingdom; Reilly 1995 for Canada), but there is much evidence that this is due to the fact that workers with higher ability are given the best technologies to use. Entorf and Kramarz (1998) emphasise that the cross-sectional association in France disappears once fixed effects have been controlled for. More cynically, DiNardo and Pischke (1997) show that the cross-sectional correlation of wages with computer use exists in the German data, but so does an equally robust correlation with pencil use! Van Reenen (1996) finds a positive effect of technology on average wages in British companies, even after controlling for endogeneity and fixed effects. This result, however, could be due to the different type of data used. Instead of computer use, he analyzes a count of major technological innovations. These generate substantial economic rents, and the paper interprets the correlation as a form of sharing in the rents from new technologies. The purchase of a computer by a firm is unlikely to generate substantial rents. Overall, there is evidence that the computer–wage correlation cannot be interpreted as simply the causal effect of technical change on individual or enterprise wages. More likely it reflects the fact that the best technologies tend to be used by the most able workers who were already earning higher wages. Employment Finally we come to the relationship between employment and technology (table 5.3). There have been relatively few cross-industry econometric studies of the impact of technology on total employment. Those that do exist tend to be mainly descriptive in character and focused on specific industries. The analysis in Blechinger et al.

210

Table 5.3 Effect of technology on employment Study

Method

Data


Controls

Result

Berndt, Morrison, and Rosenblum (1992)

OLS regression of labor-output ratio on capital intensity (equipment and high tech)

See table 5.1A.

See table 5.1A.

Time

Positive effect of high-tech capital on employment intensity

Nickell and Kong (1987)

4 equation system (pricing, production, wages, demand)

55 U.K. manufacturing industries (3 digit), 1974–85 panel

Residual based approach

Various–capital, average wages, etc.

In 7 out of 9 sectors a positive effect of labor-augmenting technical change

Employment growth regression (managers asked about employment in previous years)

U.K. 1984 WIRS (cross section) 948 establishments

Any introduction of new technology involving microelectronics in last 3 years

Age, unions, demand, ownership

Positive and significant technology effect

A. Industry level

Blanchflower, Millward, and Oswald (1991)


B. Enterprise level

(a) Manufacturing firms in Germany (1,821), Denmark (528), France (3,600), Norway (743), Spain (1,998), Luxembourg (241), Belgium (557), Italy (16, 374) in 1992 (b) 772 mnfg and 836 service firms in Netherlands

(a) Community Innovation Survey (CIS)—subjective question; whether firm performs any R&D; whether R&D directed at product or process (b) Product and process R&D personnel %; indicators for office and production automation

(a) Sales, sales squared, labor costs (industry level), qualitative indicators of barriers to innovation, exports, subsidiary status (b) Dummies for size class

(a) Innovation indicator insignificant in every country except Italy (more small firms); R&D has a positive correlation (probably due to fixed effect) (b) R&D has positive effect in both sectors (process stronger than product); office automation positive effect in services; production automation positive effect in manufacturing

Blanchflower and Burgess (1999)

Employment growth regression

1990 U.K. WIRS (831 plants), 1992 Australian AWIRS (888 plants)

Any introduction of new technology involving microelectronics in last 3 years

Employment 4 years earlier; unions, financial performance, ownership

Positive and significant in Britain; positive and weakly significant in Australia

Brouwer, Kleinknecht, and Reijnen (1993)

1983–88 employment growth regressions, Heckman two-step selectivity correction (no real identifying instruments)

1983 and 1988 859 Dutch manufacturing firms (survey)

R&D intensity, type of R&D

Firm size, industry sales growth, single plant, lagged firm sales growth, industry dummy

No effect of R&D intensity level, growth of R&D intensity has significant negative effect; mitigated by product R&D and R&D toward IT

211

(a) OLS static conditional labour demand equations separately for each country (b) Employment growth 1988–92 on 1988 characteristics; separate estimation for manufacturing and services; attempt to control for survival bias using Heckman method


Blechinger, Kleinknecht, Licht, and Pfeiffer (1998)

212


Data


Controls

Result

Doms, Dunne, and Roberts (1994)

Employment growth (and survival)

U.S. plants from LRD and Survey of Manufacturing, 1987–91

Dummy variables for numbers of advanced manufacturing technologies in the workplace

Age, capital, size, productivity

Positive effects

Entorf, Gollac, and Kramarz (1999)

Multinomial logit of individual employment paths, with individual fixed effects, controlling for economic conditions in some specifications

EE, the French household-based labor force survey; TOTTO, the 1993 technology supplement to the labor force survey; EET, the quarterly follow-up to the EE; and DMMO, an establishment based survey of labor turnover

Computer use, computer experience, use of other types of new technology (e.g., robot, video, fax)

(a) Gender, education, region, part-time indicator, occupation, size and status of employer, experience, firm age (b) Establishment turnover rates, experience, firm age, part-time indicator, retirement rate

Computer use protects workers from unemployment in the very short run, but not in the long run

Entorf and Pohlmeir (1991)

3 system equations (exports, innovation, employment); GLS

2,276 German firms in 1984 (IFO)

Responses to a survey of innovation

Export/sales, labor costs, concentration

Product innovations have positive effect; process innovations no effect


Study

3 equation system with value-added, labor and capital as endogenous, growth regressions

Balanced panel of up to 5,919 firms, 1985–91 in France

Indicator of intensity of process and product innovations in 1991

Labor costs, capital costs, size, industry

Innovating firms create more jobs; product innovations create more jobs at sector level; process innovations create more jobs at firm level (zero at sector level)

Klette and Førre (1998)

OLS estimation of job creation rates, weighted by employment shares

Over 4,000 Norwegian manufacturing firms, 1982–92

R&D intensity

Industry and time dummies, size, foreign competition

Slower growth of Norwegian high R&D firms compared to low R&D firms

Konig, Buscher, and Licht (1995)

OLS and probit estimation of factor demand models

c. 3,000 German firms from Mannheim Innovation Panel and Mannheim Enterprise Panel in 1993

Broad range of subjective indicators

Cost of capital, wages, demand expectations

Positive effect for product innovation; none for process innovations; expected demand most important

Leo and Steiner (1994)

OLS and multinomial logit models for changes in employment; lags of innovation

400 Austrian firms from WIFO (Institute of Economic Research), 1990–92

Technology and Innovation survey

Regev (1998)

Employment growth regressions

3,260 Israeli firm observations, 1982, 1985, 1988, 1992

Technology index based on R&D, skilled labor, and capital vintage

Positive effect from lagged product innovations; no effect from process innovations Positive effect

213

Industry and time dummies, size of firm, export and imports, concentration


Greenan and Guellec (2000)

Table 5.3 (continued) Data


Controls

Result

Ross and Zimmerman (1993)

Probits of planned stock of labor

5,011 German firms (manufacturing) from Munich IFO Institute in 1980

Subjective survey

Demand, labor costs

Negative effect of process innovations

Smolny (1998)

OLS estimate of employment changes

1980–92 unbalanced panel of 15,992 observations (c. 2,405 firms) in West German manufacturing (IFO)

Subjective survey (IFO), lagged variables

Capacity utilization, investment/sales ratio, size dummies, time dummies

Positive and significant impact of firm product innovations; industry product innovations have a significant negative effect on employment growth (rivalry); no significant effect of firm process innovations but significant positive effect of industry process innovations; higher volatility of employment for product innovations

Van Reenen (1997)

Dynamic employment growth model; OLS and GMM;

Unbalanced panel of 598 quoted U.K. manufacturing firms, 1976–82

Major innovations (SPRU) counted at firm and industry level (expert survey); firm level patents (taken in U.S.)

2 lags of employment, wages, capital, industry innovation, time dummies, long lags of innovation and patents

Innovations (esp. product) have large effects on employment; patents effects not robust to fixed effects

Zimmerman (1991)

Probit model for planned change in labor stock

3,374 German firms in 16 industries from IFO

Subjective survey

Demand, labor costs

Negative effect of process innovations


Method

214

Study


215

(1998) captures some of the salient points. An examination of the OECD STAN/ANBERD database (which covers manufacturing) reveals that the high-technology industries (those with higher R&D intensity) expanded more quickly (or contracted less slowly) than the medium- or low-technology industries. Focusing on the firm-level studies, there are a wide variety of results from different countries (see Spiezia and Vivarelli, chapter 3, this volume). Overall, there appear to be consistently positive effects of proxies for product innovations on the growth of employment (e.g., Konig et al. 1995 and Entorf and Pohlmeier 1995 for German firms; Leo and Steiner 1994 for Austrian firms; Van Reenen 1997 for British firms). The results for process innovations are very mixed— although usually insignificant, several examples of positive effects exist (e.g., Blanchflower and Burgess 1999 for U.K. and Australian plants; Blechinger et al. 1998 for Dutch firms; Regev 1998 for Israeli firms). In an interesting study of French data, Greenan and Guellec (2000) find that process innovations have a strong positive effect at the firm level, but this washes out at the industry level. The story is reversed for product innovations. When measures such as R&D are used, negative correlations frequently arise (see Klette and Førre 1998 for Norwegian plants; Brouwer et al. 1993 for Dutch firms). The most plausible explanation for these results is that the effects of innovation depend critically on the type of innovations being produced. Also there is a serious concern that firms are introducing new technologies when they expect demand conditions to improve, thus biasing the coefficients on the technology proxies upward. In general, existing employment studies have rarely been conducted with as detailed an eye to the econometric problems involved as those studies investigating wages and skills. This perhaps reflects the greater theoretical ambiguity involved in estimating the relationship (and policy interest in the microeconomic results). The econometric problems are particularly difficult in these studies, however, and future work needs to address these more seriously. 5.5

Conclusions

In any survey it is difficult to reach definitive conclusions, aside from methodological ones. Nevertheless, we hazard the following stylized description of our brief survey. First, there is considerable evidence of a positive correlation of various measures of technology with the

216


skill structure suggesting that technology is on average biased toward skilled labor. Second, there is also strong evidence of a positive correlation between wages and innovation. Third, the evidence on total employment is more mixed, with some measures (diffusion based) suggesting a positive association and others (R&D based) being more negative. The three main problems with these results is the presence of unobserved heterogeneity, endogeneity, and measurement problems. For the (still relatively few) studies attempting to deal with fixed effects and/or endogeneity, the skills–technology correlation appears to be stronger than the wages–technology relationship. Indeed, we would go as far to say that most studies appear to find no causal effect of diffusion-based measures (e.g., computer use) on wages. Finally there do seem to be important differences in the results stemming from different notions of innovation. The diffusionbased measures of innovation (computer use) might have no effect on wages, but large technological innovations might have a more substantial impact. Certainly the enterprise level correlations with employment growth differed enormously, depending on whether R&D or diffusion was the variable of interest. In terms of future work, two immediate points are obvious. First, we need more studies attempting to deal with the problems of the endogeneity of the technology choice by searching for better instruments that exogenously shift firms’ incentives to introduce new technologies. Work on this is only beginning. Second, the theoretical framework for analyzing these issues is still very crude. The basic neoclassical model needs to be supplemented by a richer understanding of technological adoption in a tractable manner. There are a plethora of theoretical models; the task is to translate them into an empirically coherent form for implementation and testing. In particular, examining the role of organizational change in translating the effects of technology into labor demand should be a key area of future research. Finally, although we welcome the constant quest for new indicators, one of the key concerns is to achieve stability and comparability in the time series measures of technology. Appendix: The Microeconomics of Technology and Employment: A Simple Example A special case of the translog cost function arises where there is a constant elasticity of substitution between the factors (the translog


217

allows for more general patterns of substitution and complementarity). To simplify the discussion we will work with this form. Write the production relationship as VA ¼ T½ðANÞðs1Þ=s þ ðBKÞðs1Þ=s s=ðs1Þ ;

ðA1Þ

where K ¼ capital, N ¼ labor, VA ¼ value added. T represents a neutral technology parameter, A is labor-augmenting technology, and B is capital-augmenting technology. If a firm maximizes profit, then the labor demand equation is log N ¼ log VA s log

W þ ðs 1Þ log A: P

ðA2Þ

The elasticity of labor demand with respect to a change in labor augmenting technical progress is given by q log N q log VA q log P q log MC ¼ þ ðs 1Þ; q log A q log P q log MC q log A

ðA3Þ

or more succinctly, q log N ¼ hP my þ ðs 1Þ; q log A

ðA4Þ

where the effect of technical change on labor demand is now written as a function of four factors: the price elasticity of product demand9 ðhP Þ, the mark up elasticity (a measure of market power, m), the ‘‘size’’ of the innovation as measured by its effect on marginal cost ðyÞ, and the elasticity of substitution between capital and labor ðsÞ. The interpretation of all of these results is quite intuitive and discussed in the text. Some points to note are as follows:

. When there is perfect competition m ¼ 1, and no substitution between labor and capital (e.g., if labor is the only factor of production s ¼ 0), then for a normalized innovation ðy ¼ 1Þ the effect on labor demand will hinge on whether demand is elastic. If product demand is elastic ðhP > 1Þ, then employment will rise; if it is inelastic ðhP < 1Þ, then employment will fall.

. Since it is difficult to know the effect of any given measure of in-

novation on marginal cost, it is very difficult to compare different studies to determine the quantitative effect of an innovation—there is no natural scale of normalisation. For further discussion of these points, see Van Reenen (1997).

218


Notes 1. A good survey and an interpretation of the case study evidence are given by Attewell (1987, 1990). As expected the evidence is highly mixed, with the exact effect of technology on skills being highly dependent on the particular context. 2. This section owes much to the exposition in Adams 1999. 3. This roughly corresponds to the Schumpeterian triptych of invention, innovation, and diffusion. 4. In OECD statistics most countries follow the guidelines of the Frascati manual (1993). Within countries accounting regulations often define how R&D is to be reported (e.g., in the United States under FAS and in the United Kingdom under SSAP13 [revised]). 5. Of course, the same is true of the standard way in which the physical capital stock is measured. The main difference here is that the degree of uncertainty involved with R&D investments is much greater, and there is usually a method of benchmarking the physical capital stock in a particular year. 6. Some current ideas include renewal fees, number of countries where the patent is registered, surveys of inventors, and citations. 7. Note that similar datasets are also available in European countries, but confidentiality clauses restrict general access to them. 8. The Machin and Van Reenen (1998) study investigates the sensitivity of their results to instrumenting total R&D with government-financed R&D. 9. We are assuming the elasticity between value added and output is unity.

References Adams, J. 1999. The structure of firm R&D and the factor intensity of production. Review of Economics and Statistics 81: 499–510. Aguirregabriria, V., and C. Alonso-Borrega. 2001. Employment occupational structure, technological capital and reorganisation of production. Labour Economics 8(1): 43–73. Allen, S. 1996. Technology and the wage structure. NBER Working Paper 5534 (forthcoming in Journal of Labor Economics). Arabsheibani, G., J. Enami, and A. Marin. 1996. The impact of computer use on earnings in the UK. Mimeo. Royal Holloway College, University of London. Attewell, P. 1987. The de-skilling controversy. Work and Occupations 14(3): 323–46. Attewell, P. 1990. What is skill? Work and Occupations 17(4): 422–48. Autor, D., L. Katz, and A. Krueger. 1998. Computing inequality: Have computers changed the labor market? Quarterly Journal of Economics 113(4): 1169–1213. Bartel, A., and F. Lichtenberg. 1987. The comparative advantage of educated workers in implementing new technology. Review of Economics and Statistics 69(1): 1–11.


219

Bartel, A., and F. Lichtenberg. 1991. The age of technology and its impact on employee wages. Economics of Innovation and New Technology 1: 215–31. Bartel, A., and N. Sicherman. 1998. Technological change and the skill acquisition of young workers. Journal of Labor Economics 16(4): 718–55. Bartel, A., and N. Sicherman. 1999. Technological change and wages: An interindustry analysis. Journal of Political Economy 107(2): 285–325. Bell, B. 1998. Skill-biased technical change and wages: Evidence from a longitudinal data set. Industrial and Labour Relations Review, forthcoming. Berman, E., J. Bound, and Z. Griliches. 1994. Changes in the demand for skilled labor within US manufacturing industries. Quarterly Journal of Economics 109(2): 367–97. Berman, E., J. Bound, and S. Machin. 1998. Implication of skill biased technological change: International evidence. Quarterly Journal of Economics 113: 1245–79. Berndt, E., C. Morrison, and L. Rosenblum. 1992. High-tech capital formation and labor composition in US manufacturing industries: An exploratory analysis. NBER Working Paper 4010. Cambridge, MA. Bernstein, J., and L. Mishel. 1996. Did technology’s impact accelerate in the 1980s? Mimeo. Economic Policy Institute, Washington, DC. Betts, J. 1997. The skill bias of technological change in Canadian manufacturing industries. Review of Economics and Statistics 79(1): 146–50. Blanchflower, D., N. Millward, and A. Oswald. 1991. Unionisation and employment behaviour. Economic Journal 101: 815–34. Blanchflower, D., and S. Burgess. 1999. New technology and jobs: Comparative evidence from a two country study. Economics of Innovation and New Technology 6(1/2), forthcoming. Blechinger, D., A. Kleinknecht, G. Licht, and F. Pfeiffer. 1998. The impact of innovation on employment in Europe: An analysis using CIS data. ZEW Dokumentation 98-02, Mannheim, Germany. Braverman, H. 1974. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. New York: Monthly Review. Bresnahan, T. 1997. Computerisation and wage dispersion: An analytical reinterpretation. Working Paper 97-031. Stanford University. Bresnahan, T., E. Brynjolfsson, and L. Hitt. 1998. Information technology, work organisation and the demand for skilled labor: Firm-level evidence. Industrial Performance Center Working Paper 98-010. Massachusetts Institute of Technology. Quarterly Journal of Economics, forthcoming. Brown, R., and L. Christensen. 1981. Estimating elasticities of substitution in a model of partial static equilibrium: An application to U.S. agriculture 1947–74. In C. Firled and E. Berndt, eds., Modelling and Measuring Natural Resource Substitution. Cambridge: MIT Press. Brouwer, E., A. Kleinknecht, and J. Reijnen. 1993. Employment growth and innovation at the firm level: An empirical study. Journal of Evolutionary Economics 3: 153–59.

220


Buckberg, E., and A. Thomas. 1996. Wage dispersion in the 1980s: Resurrecting the role of trade through the effects of durable employment changes. IMF Staff Papers 43(2): 336–54. Card, D., F. Kramarz, and T. Lemieux. 1996. Changes in the relative structure of wages and employment: A comparison of the U.S., Canada and France. NBER Working Paper 5487. Cambridge, MA. Caroli, E., and J. Van Reenen. 2001. Skills and organisational change: Evidence from British and French establishments in the 1980s and 1990s. 116(4): 1449–92. Casavola, P., P. Gavosto, and P. Sestito. 1996. Technical progress and wage dispersion in Italy: Evidence from firms’ data. Annales d’Economie et de Statistique (Jan–June): 387– 412. Chennells, L., and J. Van Reenen. 1997. Technical change and earnings in British establishments. Economica 64(256): 587–604. Davis, S., and J. Haltiwanger. 1991. Wage dispersion between and within US manufacturing plants, 1963–86. Brookings Papers, Microeconomics. DiNardo, J., and J. S. Pischke. 1997. The returns to computer use revisited: Have pencils changed the wage structure too? Quarterly Journal of Economics 112(1): 291–303. Dickens, W., and L. Katz. 1987. Inter-industry wage differences and industry characteristics. In K. Lang and J. Leonard, eds., Unemployment and the Structure of Labour Markets. New York: Basic Books. Doms, M., T. Dunne, and M. Roberts. 1994. The role of technology use in the survival and growth of manufacturing plants. Mimeo. Center for Economic Studies. U.S. Bureau of the Census. Washington, DC. Doms, M., T. Dunne, and K. Troske. 1997. Workers, wages and technology. Quarterly Journal of Economics 112(1): 253–89. Duguet, E., and N. Greenan. 1997. Skill biased technological change: An econometric study at the firm level. INSEE working paper. Paris. Dunne, T., J. Haltiwanger, and K. Troske. 1997. Technology and jobs: Secular changes and cyclical dynamics. Carnegie-Rochester Conference Series on Public Policy 46: 107–78. Dunne, T., and J. Schmitz. 1995. Wages, employment structure and the employer sizewage premia: Their relationship to advanced technology usage at US manufacturing establishments. Economica 62: 89–108. Edwards, R. 1979. Contested Terrain: The Transformation of the Workplace in the Twentieth Century. New York: Basic Books. Entorf, H., M. Gollac, and F. Kramarz. 1999. New technologies, wages and worker selection. Journal of Labor Economics 17(3): 464–91. Entorf, H., and F. Kramarz. 1997. Does unmeasured ability explain the higher wages of new technology workers? European Economic Review 41(8): 1489–1509. Entorf, H., and F. Kramarz. 1998. The impact of new technologies on wages and skills: Lessons from matching data on employees and their firms. Economics of Innovation and New Technology 5: 165–97.


221

Entorf, H., and W. Pohlmeir. 1991. Employment, innovation and export activity. In J. Florens, M. Ivaldi, J.-J. Laffont, and F. Laisney, eds., Microeconometrics: Surveys and Applications. Oxford: Basil Blackwell. Garen, J. 1994. Executive compensation and principal-agent theory. Journal of Political Economy 102(6): 1175–99. Gera, S., W. Gu, and Z. Lin. 1998. Technology and the demand for skills: An industrylevel analysis. Mimeo. Statistics Canada, Ottawa. Goldin, C., and L. Katz. 1996. Technology, skill and the wage structure: Insights from the past. American Economic Review 86 (May): 252–57. Goux, N., and E. Maurin. 1997. The decline in demand for unskilled labour: An empirical analysis method and its application to France. CREST mimeo. Paris. Greenan, N., and D. Guellec. 2000. Technological innovation and employment reallocation. Labour 14(4): 547–90. Greenan, N., J. Mairesse, and A. Topiol-Bensaid. 1998. Information technology and research and development impacts on productivity and skills: Looking for correlations on French firm level data. In M. Pohjola, ed., Information Technology Productivity and Economic Growth. Oxford: Oxford University Press, pp. 119–48. Hansson, P. 1997. Trade, technology and changes in employment of skilled labour in Swedish manufacturing. In J. Fagerberg, P. Hansson, L. Lundberg, and A. Melchior, eds., Technology and International Trade. Cheltenham: Edward Elgar. Haskel, J., and Y. Heden. 1997. Computers and the demand for skilled labour: Panel evidence for the UK. Economic Journal Conference Proceedings 109: C68–C79. Hildreth, A. 1998. Rent sharing and wages: Product demand or technology driven wage premia? Economics of Innovation and New Technology 5: 199–226. Holthausen, R., D. Larcker, and R. Sloan. 1995. Business unit innovation and the structure of executive compensation. Journal of Accounting and Economics 19(2–3): 279– 313. Jackman, R., R. Layard, M. Manacorda, and B. Petrongolo. 1996. European vs. US unemployment: Different responses to the same shock? Centre for Economic Performance Discussion Paper 349. London School of Economics. Kaiser, U. 1998. The impact of new technologies on the demand for heterogeneous labour. Working Paper 98-26. Centre for European Economic Research (ZEW). Mannheim, Germany. Kaldor, N. 1932. A case against technical progress. Economica (May): 180–96. Katsoulacos, Y. 1986. The Employment Effect of Technical Change. Oxford: Oxford University Press. Klette, T., and S. Førre. 1998. Innovation and job creation in a small open economy: Evidence from Norwegian manufacturing plants 1982–92. Economics of Innovation and New Technology 5: 247–72. Konig, H., H. Buscher, and G. Licht. 1995. Employment, investment and innovation at the firm level. In The OECD Jobs Study: Evidence and Explanations. Paris: Organisation for Economic Co-operation and Development.

222


Krueger, A. 1993. How have computers changed the wage structure: Evidence from micro data, 1984–89. Quarterly Journal of Economics 108(1): 33–61. Krugman, P. 1994. Past and prospective causes of high unemployment. Federal Reserve Bank of Kansas City Economic Review (3rd quarter): 23–43. Leo, H., and V. Steiner. 1994. Innovation and Employment at the Firm Level. Vienna: Osterreichisches Institut fu¨r Wirsschaftsforschung. Loh, E. 1992. Technological changes, training and the inter-industry wage structure. Quarterly Review of Economics and Finance 32(4): 26–44. Lynch, L., and P. Osterman. 1989. Technological innovation and employment in telecommunnications. Industrial Relations 28(2): 188–205. Machin, S. 1996. Changes in the relative demand for skills in the UK labour market. In A. Booth and D. Snower, eds., Acquiring Skills. Cambridge: Cambridge University Press. Machin, S., N. Menezes-Filho, and J. Van Reenen. 1998. R&D, pay and performance: Evidence from UK firms. IFS mimeo. London. Machin, S., and J. Van Reenen. 1996. The impact of R&D knowledge accumulation on wages: Evidence from European corporations. In A. Belcher ed., R&D Decisions: Strategy, Policy and Disclosure. London: Routledge. Machin, S., and J. Van Reenen. 1998. Technology and changes in skill structure: Evidence from seven OECD countries. Quarterly Journal of Economics 19(4): 347–74. Manacorda, M., and A. Manning. 1997. Just can’t get enough: More on skill biased change and unemployment. Centre for Economic Performance Discussion Paper 412. London School of Economics. Martinez-Ros, E. 1998. Real wages and innovations in Spanish manufacturing firms. Mimeo. Universitat Autonoma de Barcelona. Mincer, J. 1991. Human capital, technology, and the wage structure: What do time series show? NBER Working Paper 3581. Cambridge, MA. Nickell, S., and B. Bell. 1995. The collapse in the demand for the unskilled and unemployment across the OECD. Oxford Review of Economic Policy 11(1): 40–62. Nickell, S., and P. Kong. 1987. Technical progress and jobs. Centre for Labour Economics Discussion Paper 336. London School of Economics. OECD. 1994. The Jobs Study. Paris: OECD. Osterman, P. 1986. The effect of computers on the employment of clerks and managers. Industrial and Labour Relations Review 39(2): 175–85. Regev, H. 1998. Innovation, skilled labour, technology and performance in Israeli industrial firms. Economics of Innovation and New Technology 6(1/2), forthcoming. Reilly, K. 1995. Human capital and information: The employer size-wage effect. Journal of Human Resources 30 (winter): 1–18. Ross, D. R., and K. Zimmerman. 1993. Evaluating reported determinants of labour demand. Labour Economics 1: 71–84.


223

Smolny, W. 1998. Innovations, prices and employment: A theoretical model and empirical application for West German manufacturing firms. Journal of Industrial Economics 46(3): 359–82. Solow, R. 1966. Wiskell Lectures. Oxford: Basil Blackwell. Siegel, D. 1998. The impact of technological change on employment and wages: Evidence from a firm-level survey of Long Island manufacturers. Economics of Innovation and New Technology 5(2): 227–46. Soete, L., and G. Dosi. 1983. Technology and Employment in the Electronics Industry. London: Francis Pinter. Tan, H., and G. Batra. 1997. Technology and firm size-wage differentials in Colombia, Mexico, and Taiwan (China). The World Bank Economic Review 11(1): 59–83. Troske, K. 1999. Evidence on the employer size-wage premium from workerestablishment matched data. Review of Economics and Statistics 81(1): 15–26. Ulph, D. and A. Ulph. 1994. Labour markets and innovation: Ex post bargaining. European Economic Review 38: 195–210. Vainioma¨ki, J. 1998. Technology, skills and wages: Results from linked worker-plant data for Finnish manufacturing. Mimeo. University of Tampere. Van Reenen, J. 1996. The creation and capture of economic rents: Wages and innovation in a panel of UK companies. Quarterly Journal of Economics 111: 195–226. Van Reenen, J. 1997. Technological innovation and employment in a panel of British manufacturing firms. Journal of Labor Economics 15(2): 255–84. Wolff, E. 1996. Technology and the demand for skills. Science, Technology Industry Review. Paris: OECD. Zimmerman, K. 1991. The employment consequences of technological advance: Demand and labour costs in 16 German Industries. Empirical Economics 16: 253–66.


Comments on Chapter 5 Eric Maurin

What is the impact of new technologies on wages, jobs, and skills? This question lies at the heart of one of the most active areas of economic research today. In their contribution Lucy Chennels and John Van Reenen offer us a lucid, stimulating overview of the topic very much in line with their own work on the subject. I especially welcome their invitation to separate the issue of the role of technologies from the issue of rising inequality. It is not because a technology ‘‘requires’’ a fairly skilled labor force that it necessarily generates inequality between skilled and unskilled workers. On the plausible assumption (in my view) that blue-collar workers complement engineers in the same way as arms complement a head, all factors beneficial to engineers will benefit blue-collar workers too. Chennels and Van Reenen also chart some very interesting paths for future inquiry. In particular, it is clear that tomorrow’s research will have everything to gain from a better conceptualization of work organization, and from a better understanding of the way in which the effects of technology are mediated by work organization. Some recent studies suggest that work organization should be treated as a special form of capital, that is, as a special input that has so far been neglected in productivity analysis (e.g., see Caroli and Van Reenen 1999). A slightly different approach consists in expanding the modeling of firms to include the diversity of functions and tasks assembled in the firm (Maurin and Thesmar 1999). Using skilled and unskilled labor, the vast majority of firms produce a typically abundant set of intermediate outputs that reflect their specific organizations: they simultaneously produce R&D, sales and marketing, transportation, handling, and so on. In most cases the total output and the value added of firms are not produced by a direct combination of skilled

226

Eric Maurin

and unskilled labor; rather, they are generated from a combination of complex intermediate products that have manifestly required specific production functions and technologies. If we subscribe to this analytical framework, it is not just one input that has been omitted, but the firm’s entire intermediate output. Accordingly the increase in labor skills in firms may be a mere ‘‘composition’’ effect, meaning the result of factors that do not promote skilled labor in itself but promote the basic functions (e.g., sales and marketing) that generally rely on labor with above-average skills. I have two general reactions to Chennels and Van Reenen’s study. First, perhaps the most striking feature for me is the ‘‘static’’ character of the econometric models typically used to identify the impact of technology diffusion on employment and wage dynamics. None of the models reviewed by the two authors takes into account the employment-protection rules and the adjustment costs that these rules entail. The impact of technology diffusion is analyzed in a framework where adjustments for all types of labor are instantaneous. Such an assumption is highly unrealistic and, no doubt, responsible for significant errors in the empirical assessment of the role of technical progress. In fact adjustment costs and the degree of job protection are not only crucial to understanding the dynamics of employment and wages in a given country at a given time; they are also variable over time, and differ from one type of workforce to another (see the survey by Hamermesh and Pfann 1996). In most western countries today, for example, it is far less costly to lay off low-seniority workers than high-seniority ones (e.g., see Goux, Maurin, and Pauchet 2001), and this asymmetry has probably increased in recent decades. In many countries workers must have several months’ tenure in a firm before being entitled to severance pay and advance notice of dismissal; the regulations on temporary-employment contracts have been considerably liberalized since the early 1980s. In other words, it is probable that the discounted cost of employing low-seniority workers has fallen significantly in many western countries. In periods when the education system is expanding, new entrants into the labor market have a higher educational attainment, at each successive date, than the average employee. In theory, institutional changes have therefore promoted the substitution of younger, better-educated workers of older, less-educated


227

workers, and the development of frictional unemployment at the start of working careers and unemployment of a more structural kind at the end of working careers. Actual statistics do confirm the substitutions and types of unemployment listed above, at least in continental European (OECD 1999). The second salient feature of the Chennels and Van Reenen study is that the literature they review focuses almost exclusively on the link between technologies and skills. There seem to be no studies about the effect of new technologies on the other major forms of human capital: the age and job tenure of payroll workers. As I have suggested, if we were to extend the analysis to adjustment costs and institutional change, we would need to examine the length of job tenure, and to treat it as a full-fledged input along with skills. For the moment we have two literature that virtually ignore each other: the first speculates on the existence of a technological bias and its link with the increasing returns to education; the second speculates on the reasons for the fall in returns to job tenure (or rather on the increase in their dispersion) and on the links between this fall and the advent of a ‘‘post-Fordist’’ society dominated by the service sector (e.g., see Osterman 1994). Clearly, the two literature would gain from an encounter. I do not believe one can credibly identify a technological bias in labor demand without making an allowance for the institutions that protect senior workers; nor is it likely that one can properly understand the shifts in job duration in recent decades without taking account of technological change. On the age issue there are no studies that seek to determine whether new technologies exclude older workers or not. This omission is all the more surprising as—according to OECD figures—age is one of the most crucial factors of inequality in many countries. In France, for example, inequality in hiring and in long-term unemployment is far greater by age than by educational attainment. As often, one of the likely reasons for this seeming lack of interest on the part of researchers is the absence of available data. Today there are many statistical sources for analyzing the dynamics of labor skills in firms, but to my knowledge, there are far fewer sources describing the dynamics of job distribution by age or tenure. In this field perhaps more than the others, tomorrow’s pioneers will be the researchers with the tenacity to construct new data bases relevant to the crucial issues that remain unsettled.

228

Eric Maurin

References Caroli, E., and J. Van Reenen. 1999. Skills and organisational change: Evidence from British and French establishments in the 1980s and 1990s. Quarterly Journal of Economics, forthcoming. Goux, D., E. Maurin, and M. Pauchet. 2001. Fixed-term contracts and the dynamics of labour demand. European Economic Review 45: 533–52. Hamermesh , D., and G. A. Pfann. 1996. Adjustment costs in factor demand. Journal of Economic Literature 34: 1264–92. Maurin, E., and D. Thesmar. 1999. Changes in the functional structure of firms and the demand for skill. CREST Working Paper. Paris. Osterman, P. 1994. Internal labor markets: Theory and change. In C. Kerr and P. D. Staudohar, eds., Labor Economics and Industrial Relations. Cambridge: Harvard University Press. OECD. 1999. Job Studies. Paris: OECD.

6

By What Means Does Information Technology Affect Employment and Wages? Kathryn Shaw

Information technology (IT) has been frequently mentioned as a possible cause for rising wage inequality and declining employment levels for less-skilled labor.1 IT is a prime suspect in the quest for a culprit for rising income inequality because the circumstantial evidence is strong. The circumstantial evidence lies in the obvious correlations: as income inequality increased worldwide, the price of computing power has declined and the prevalence of computers has risen. Strong circumstantial evidence does not imply guilt: to determine that IT is guilty of contributing to the rising inequality, many have searched for a causal relationship between rising IT and rising inequality. There are several reasons why it is so difficult to find IT guilty of causing rising inequality using the available evidence. First, IT is very heterogeneous, and thus hard to define in theory (Is it computing power, or communications, or network access to information?). Second, the measures of IT are often poor and also heterogeneous (reflecting in part the heterogeneity in the meaning of IT); thus it is very difficult to provide a quantitative and hence generalizable assessment of the effects. Third, the use of IT is likely to be correlated with (or embodied in) other inputs, like R&D or physical capital, and hence it is can be difficult to assign a value to the effects of IT. Fourth, the econometric identification of IT effects faces the usual problems of endogeneity and selection issues (see Chennells and Van Reenen, chapter 5 in this volume). Since it is difficult to find causal evidence that the greater use of IT has contributed to rising inequality, we are trying an alternative tactic to assigning guilt—a search for the motives of why IT might be partially responsible. That is, if it is difficult to econometrically assign an impact of IT on inequality, then search for the underlying motives or

230

Kathryn Shaw

reasons why IT might be guilty. Thus I am addressing the question, by what means do information and communications technologies affect employment and wages? This question—‘‘by what means?’’—implies that we must look within firms to understand how the use of IT within firms might be altering the demand for labor. That is my purpose here, and because we have so little hard data on what occurs within firms, I will be citing case study evidence more than econometric evidence. In addition, I will be focusing more on the manufacturing sector rather than services, because to the extent that we have firm-level data, it is largely for manufacturing. Last, since my knowledge of the steel industry is the most extensive, I will begin with that example. For a brief summary of the conclusions please turn to the final section of this chapter. In the remainder of the chapter, I begin with the steel industry example, then broaden the perspective to other manufacturing industries in section 6.2. In section 6.3, I discuss nonIT reasons for changes in employment demand, and in section 6.4, I describe the ways in which IT and HRM practices contribute to knowledge capital. Finally, in section 6.5, I discuss the effects of IT on wage rates. 6.1

Evidence from the Steel Industry

I start my analysis with an example of the steel industry because there is so little ‘‘hard’’ data on the means by which IT is changing jobs. Following this case study evidence, I will ask, is there much evidence that the steel industry example is representative of other industry experiences? In the last eight years, I and my co-authors have visited 84 steel mills owned by 40 companies in the United States, Japan, and France (covering over half of the mills in the United States).2 Our primary purpose has been to learn what key factors contribute to the variation in productivity across plants. At our plant visits we combined an interest in technology and in organizational factors, such as human resource management (HRM) practices. At each visit, we toured the mill to obtain information on the technology in use, we interviewed operations managers, engineers, and human resource managers using a prepared interview survey guide, and we obtained data on the production performance of the mill as well as on the technology and the workers (age, tenure). The timing of the data

By What Means Does IT Affect Employment and Wages?

231

varies across mills, but largely covers the years 1988 to 1997. In order to make our analysis the most comparable across mills, we focused on one type of finishing process in integrated steel mills and one type of process in minimills. On the basis of this very extensive knowledge of the steel industry, I have formed a number of conclusions regarding the effects of IT on employment and wages. How Does IT Affect the Jobs of Steelworkers? Our extensive plant visits gave us exceptional evidence of how IT changed steelworkers’ jobs. To help the reader, I will describe a particular steel process and the use of IT in that process. One common steel process is tin-plating—taking thin sheets of steel and putting a coating of tin on the sheet so that the steel can be used to make tin cans. This process would be considered a ‘‘finishing’’ process in a large integrated steel mill. In the process, a thin sheet of steel about a mile long begins as a coil of steel. This coil is then loaded on the production line, it is uncoiled and coated with tin, and finally it is recoiled at the end of the line. This same process has existed for decades, but it is now monitored differently by production workers. In 1980, a typical tin mill would have workers positioned along the line to do mechanical activities: a worker at the beginning would load the coil and weld the steel to the previous coil; several workers in the middle of the line would constantly monitor the tin-coating process, changing the mix of the coating as needed, or changing the position of the steel, or the speed of the line; finally, at the end, a worker would coil the steel and cut the strip to remove the coil. Each job is very mechanical, and the worker would receive little feedback as to how the process is going, short of obvious failures in the line. At the end of the line, the steel would be inspected to check the quality of the coating and the steel strip (e.g., avoiding uneven coating or wrinkled edges). Today, there would be very few workers actually working on the line—production workers would work from a computerized pulpit above the line. In that air-conditioned room, there would be a vast array of computer monitors that would show two things: pictures of the steel process (from cameras on the line) and diagrams of the process with numbers from gauges along the line. The mechanical part of the line would be controlled by machinery—a crane operator

232

Kathryn Shaw

would load the coil on the line and the machinery would take over from there (setting it up on the line and beginning the process). The operators in the pulpit would then monitor all aspects of the line using computerized information from gauges along the line: the quality of the weld between the coil that is loaded and the strip already in the line, the quality of the coating that is placed on the steel strip, the position of the steel strip in the line, and the position of the steel as it is recoiled off the line and cut from the strip. In some ways the steelworker’s job has changed dramatically from the introduction of computers or information technology, and in some ways it has not changed. How has it not changed? The basic steel-coating process is unchanged, and therefore the worker’s basic understanding and control of steel coating process is also unchanged. Instead of working on the line to manually change the coating process, the operator changes it from the pulpit. On the other hand, the operator is now running the line from a series of computers, and therefore the work process is clearly very different. Does this change in the work process of coating steel require a more highly skilled or more highly educated operator? To some extent it does not. As in the past the operator must be capable of understanding the basic mechanical process, and this requires a high school education and some mechanical aptitude. The beauty of the introduction of computers in the workplace is that the software that integrates computers is so good that production workers do not require extensive computer skills. The computers merely take the information from the line and make it more accessible to the operator, in a visual and mechanically logical fashion. It is this last point that is crucial in understanding whether IT use now requires a more skilled operator—the operators now have far more information than they did in the past. In the past they had very limited information—visual inspection of the coating process, of the mechanics of the line, and of the final product that is coated. Now they have second-by-second information on each element of the process. So the key question is, How do operators work differently given this vast increase in information? To answer, I turn to the introduction of HRM practices. The Use of Innovative HRM Practices in the U.S. Steel Industry In my studies of the steel industry, my focus has been on the use of innovative HRM practices as a means of raising the productivity of


233

the line. My co-authors and I conclude in these studies that innovative HRM practices have raised performance. In this subsection I would like to provide the evidence on the use of HRM practices, before turning in the next subsection to the question, How do innovative HRM practices interact with IT use to change worker productivity and to change the demand for labor? What is meant by the phrase ‘‘innovative HRM’’ practices? The use of innovative human resource management (HRM) practices refers to the goal of increasing employee involvement (EI) through the use of management practices such as problem-solving teams, enhanced training, job rotation, better information sharing, careful screening of employees, and the use of incentive pay. Plants using so-called traditional HRM practices would have little teamwork or job rotation for their employees, little plant-level data shared with employees, very limited screening of production-level employees, and weaker pay incentives for performance. One of the key differences between traditional HRM practices and innovative HRM practices is the degree of decision making that the production employee has in the two different environments. In the traditional environment, if the production employee encounters a problem in the performance of the line, he calls over the foreman and stops work while the foreman solves the problem. In the innovative environment, a common problem is likely to be solved by the production employee, and a more complicated problem would require that he gather together a group of operators and managers. Moreover, in the innovative environment, the production employee is encouraged to recognize small problems and act on them before they become bigger problems. For example, if the innovative production worker notices that the tin coating on the steel is uneven, he would respond to the problem immediately, but in contrast, the traditional production worker might wait a longer interval until the inspector at the end of the line detects the problem and acts on it. To measure the state of HRM practices on each production line, we use clustering methods to group individual HR practices (training, teams, screening, etc.) to form ‘‘systems’’ of HR practices. These systems are largely scales of the extent of innovative HRM use. We have four systems: 1 through 4, where system 1 is the most innovative system of HRM practices, and system 4 is the most traditional system (having no teams, little training, no screening, no job security, no information sharing, no job rotation, and no innovative incentive pay). The middle systems are the communications system of

234

Kathryn Shaw

Table 6.1 Adoption of innovative HRM systems on steel finishing lines (in %) System 4 Newly built lines

System 3

System 2

System 1

0

0

57

43

Reconstituted lines

20

20

40

20

Old lines

31

57

12

0

Source: Ichniowski and Shaw (1995). Note: System 1 has the most innovative HRM practices and system 4 the least (see text). Newly built lines are those built after 1975, and old lines are those built before that. Reconstituted lines are old lines that have new owners and employees.

information sharing (system 3) and the high-teamwork system (system 2). Thus these systems measure the degree of employee involvement and the use of incentive pay to elicit involvement. For the 45 production lines that are in the integrated segment of the steel industry, the distribution of systems varies by plant age, as shown in table 6.1. This table shows that all ‘‘newly built lines’’ adopt more innovative HRM practices that introduce extensive employee involvement in the form of either self-managed work teams or problem-solving teams, and they have a full set of complementary HR practices (e.g., enhanced training, information sharing, rotation, and careful screening). The ‘‘old lines’’ are much further behind: either managers do not believe in the new practices, or they are unable to transition readily to the new systems given the older workforce (see Ichniowski and Shaw 1995 for further explanation). The lines in the middle systems are the ‘‘reconstituted lines’’ that have new owners and employees, but old capital. Given new employees, many have made the transition to more participatory HRM practices. Finally, though not shown in the data, almost 20 percent of the lines made changes in HR practices over the five years of observation, and all changes are up the scale: lines moved from systems 4 and 3 to systems 3 or 2 (so that by the end of the period, no lines remained in system 4). For the 34 production lines that are in the minimill segment of the steel industry, the key factor measuring EI is the use of problemsolving teams (mini-mills do not fit as readily into the same systems as integrated mills). Teams were often introduced in the last five years: from approximately 1991 to 1997, the percent of lines having problem-solving teams rose from 12 to 41 percent. Thus, as with the integrated mills, the movement is toward greater participation. At


235

the same time, as team use rose, the use of incentive pay rose from 73 to 92 percent of all lines. Finally, the majority of the mills with teams are newer mills: more than half of those with teams were built after 1975, but only 20 percent of the non-team mills were built after 1975.3 Thus our steel industry data on the introduction of innovative HRM practices shows that (1) the use of innovative HRM practices is a recent phenomenon, and (2) new mills are much more likely than old mills to adopt EI practices. This is true for integrated mills and minimills. How Do IT and HRM Practices Interact for Steelworkers? In the two sections above, I have made the following arguments: (1) steel mill operators now work with more computers and thus have much more information than they did in the past, and (2) newer steel mills are more likely to adopt innovative HRM practices than are the older steel mills, giving mill operators more decision-making power. These two changes in mills—more information for production workers and more decision-making power for production workers— now act as complements in the production process. First, the newer mills that have more innovative HRM practices are also more heavily computerized. Of course, there are no company data to make this point. Companies do not collect data on ‘‘computerization’’ or IT investments for the lines we are studying. On our plant visits, we created a survey measure of the degree of computerization of the integrated steel lines we visited. Using this measure, the simple correlation between the degree of computerization and the index of the degree of innovative HRM practices is 0.23. This correlation implies that older mills do some upgrading of their computerization, but that more upgrading occurs when new mills are built that also adopt innovative HRM practices. Second, given a higher degree of computerization on the line, the mills with more innovative HRM practices are better able to produce productivity gains from computerization. The computerized pulpits provide far more information to the operator than did the older more mechanical processes. If the operator must call the foreman before he can act on that information, as is true in mills with traditional HRM practices, the value of the immediate information is diminished. Either a foreman must stay in the pulpit (and thus be an

236

Kathryn Shaw

added cost) or the production worker must act on that information. For example, in the tin mill, if the operator finds that the steel strip in the line is tracking poorly so that the edges are crumbling, he should quickly act to correct the problem; he therefore would make less bad steel than if he had waited for a foreman to arrive and act. Thus innovative HRM and IT would appear to be complements— to be adopted in unison and to jointly raise productivity more than they would individually. In our empirical studies of the effects of innovative HRM practices on productivity, we found that lines that adopted the full set of HRM practices, rather than only a minimal set of practices, had considerably higher performance levels (Ichniowski, Shaw, and Prennushi 1997). We have no extensive data on IT use, but the plant visits provided interview evidence that there should be performance gains to the combined use of IT and innovative HRM. How Do IT and Innovative HRM Practices Alter the Demand for Labor? In the course of obtaining this data through plant visits, we inquired extensively about changes over time in hiring standards. In the past, when the steel industry was growing, they did virtually no screening of applicants. In the steel communities that surrounded integrated steel mills, it was simply standard practice for a young man to grow up and follow his father or uncle into the mill (Hoerr 1988; Biesen 1996). Biesen describes a representative situation: ‘‘Dennis Adams had his pick of the mills in 1964. He went to work for Youngstown Sheet and Tube . . . because the first bus at his stop was going there’’ (Biesen 1996, p. 29). Thus manufacturing jobs that paid well were plentiful and easily obtained for a worker with very limited education. This was true of integrated and minimills. The steel mills that are hiring today are looking for an entirely different type of employee—one who can think for himself, solve mechanical problems, participate in teams, communicate well, is responsible and reliable, and has a ‘‘positive attitude’’ toward hard work and rewards. The educational standards have not changed perceptibly over time, but it is now more likely that the production worker is expected to have a high school degree and some experience (either educational or on-the-job) with mechanical processes. And as Robert Garvey, CEO of Birmingham Steel stated in 1996, ‘‘Most outsiders do not appreciate the high-level technical and com-


237

puter skills required of workers to perform their jobs properly.’’4 The primary change in hiring is that firms seek a different type of person, not a different set of educational skills. The average high school educated young man simply does not have access to these high-paying jobs as he did in the past, but the better-skilled high school educated young man does. In sum, in the steel industry labor demand has changed over time to introduce a preference for more skilled production workers. The use of IT and innovative HRM practices that encourage higher levels of decision making by production workers encourages firms to demand workers who have better technical and team skills than they did in the past, though not necessarily higher educational levels.5 6.2

IT and Labor Demand: General Evidence

The evidence from the steel industry is valuable for the level of convincing detail that is offered, but the degree to which it is generalizable outside steel is, of course, unknown. However, the steel industry conclusions have intuitive analytical appeal: as IT increases the amount of information going to production workers, changes occur in nature of their duties, and innovative HRM practices may facilitate the changes in job duties. In this section I gather limited evidence that this logic does apply to manufacturing jobs outside steel. The ‘‘IT Shock’’ Changed the Products Demanded and the Production Processes As a starting point, we ask the question, Has IT changed the products we consume so that these changes have repercussions in the labor market? There are many examples of ways in which IT has changed the products that we consume in terms of quality and other features. Some typical changes in product demand are the following:

. IT enables firms to meet customers’ demands for higher-quality products (from steel, to autos, to banking services, to grocery store diversity), because firms have the IT and organizational tools to produce quality more cheaply. The total quality management (TQM) movement was begun before IT prices had fallen substantially, but as IT prices fell, firms used IT tools to improve quality (e.g., more

238

Kathryn Shaw

careful monitoring of production lines, SPC, the provision of realtime data in managers’ offices). For example, steel companies now produce much higher quality steel, and that has been driven in part by the customers’ demand (especially auto companies) and in part by the availability of cheaper monitoring processes. Firms upgrade their lines either by putting in place new physical capital that is more IT intensive or by upgrading the computerization of their lines.

. IT enables firms to produce products that meet a specific cus-

tomer’s demand specifications (steel products, automobiles, printing, clothing, or management consulting, etc.). When GM spent billions of dollars to add robotics to their production lines in the 1980s, they neglected one key factor—the payoff to robotics comes from having product diversity. Thus, when they put robots on the line, they needed to market automobile variety to receive a higher payoff from the robotics (Milgrom and Roberts 1992). In many other production and service industries, IT gives the firms the ability to alter the line and to customize the work for the particular customer.

. IT enables firms to offer products on a much more timely basis,

with a much shorter lag between order and receipt. Retail clothing outlets now order new stock for constant replenishment as clothing sells, rather than ordering prior to each season (and Levi jeans can now be ordered to a person’s exact measurements).6 Manufacturers are using JIT (just in time) for their inputs rather than ordering in large quantities and running inventories.

. IT enables retail firms to operate on slimmer markups and offer

some products more cheaply, thus raising demand for these products relative to others. IT enables Walmart to connect directly to their suppliers, so that as product sales rise, products are constantly replenished. Walmart also operates on a volume basis, so they are able to operate on slimmer markups than do traditional department stores (putting such stores out of business).

. IT changes relationships with suppliers so that relationships are

now much closer (as suppliers meet very specific needs of their customers). This can change market structure, as firms have less need for vertical integration.

. IT elevates the amount of trade worldwide by facilitating commu-

nications, and thus changing the mix of domestic product demand. U.S.-produced products should be those that use less labor and are more skill intense (because labor is expensive in developed coun-


239

tries), or are more costly to ship (e.g., high-quality steel), or require rapid shipping times (e.g., specific apparel lines that are replenished rapidly, or JIT inventories). There are also examples of ways in which IT changes the production processes. Though some of these effects were described above for the steel industry, we list several categories of effects here.

. IT changes the production process directly, by substituting robots or computer-controlled machines for workers. This would be particularly true in hazardous jobs (e.g., automobile paint shops).

. IT changes the production process directly, by introducing gauges

and controls on the line that monitor and control production more closely, often to produce higher-quality products.

. IT changes the production process indirectly, by introducing ex-

tensive software, such as software that forecasts product demand and arranges production schedules, that controls inventories, that links information between suppliers and producers, that monitors products produced and shipping, and that controls many of these simultaneously (e.g., Enterprise Resource Planning programs). An important feature of these changes in product demand and processes is that they suggest that IT is likely to have noticeable changes in product demand within industries. That is, though we know that there has been a shift over time in employment from manufacturing toward services, the examples above pertain to changes within manufacturing or services—these are changes in the quality of cars that are consumed, or in the types of banking services or clothing purchased. These within-industry changes are important as we turn later to labor demand. The HRM Technology Shock Changed the Production Process The primary purpose of this chapter is to examine the effects of the so-called IT technology shock—the abrupt introduction of computers and information technology into the workplace as IT prices fell. Nowadays IT investment is thought to be as much as one-third of all private U.S. investment and private investment in IT grew at a 28 percent real annual rate from 1995 to 1999 (Council of Economic Advisers, 2001). Thus the shift from traditional capital to IT capital would represent a shock to the production function.

240

Kathryn Shaw

However, the introduction of innovative human resource management (HRM) practices over time also represents a technology shock that affects labor demand. Robert Cole (2000) does a very persuasive job of describing participatory HRM innovations as a shock to managers. Twenty-five years ago the Japanese system of HRM practices focused heavily on problem-solving teams and the importance of the production worker, whereas the traditional U.S. system of HRM practices gave production workers very little problem-solving involvement (as described in the steel industry example above). Cole posits that U.S. manufacturers had a huge gap in product quality relative to the Japanese in the 1970s and early 1980s, and yet the U.S. managers were slow to react to this gap and to adopt practices that could close the gap. As he states, ‘‘the quality challenge appeared to U.S. companies, for all practical purposes, to be a one-time ‘never having been seen before’ event’’ and thus U.S. managers were slow to react, questioning the evidence of the value of the Japanese example and reluctant to change beliefs’’ (Cole, p. 72). Some innovative firms took up the challenge, others moved more slowly, and some of the early changers were among Japanese transplants to the U.S. (CutcherGershenfeld et al. 1994). Given this description of the abrupt change in the technical knowledge regarding managerial practices, the mid-1980s is a period in which this HRM technology shock struck U.S. firms, and firms were faced with the decision as to whether or not to adopt the new technology. Of course, as with any new technology, the value of the technology is uncertain to all, and the value of the technology is likely to vary considerably across work environments. Several surveys provide data on the extent of the adoption of HRM, and on the changes over time in HR use.7 In summarizing the results of these surveys, I focus on measures that capture changes in the overall HRM environment rather than on the detailed policies that changed.8 In tables 6.2 and 6.3, I show that (1) at least half of all establishments have now adopted some form of employee involvement (EI), (2) manufacturing uses a slightly different set of practices than services as they seek employee involvement (focusing more on quality and job rotation), and (3) extensive use of the practices within firms remains low—varying by practice, but by 1997, TQM, quality circles, and job rotation were widely in use. Many firms are trying EI in some areas, but it remains difficult to use in all areas of the firm.


241

Table 6.2 Adoption of employee involvement (in %) 1992 Any use

50% or more All

Manufacturing

1993 Any use

1994 Any use

1997 50% or more

50

41

32

47

54

38

45

25

32

76

73

41

46

27

30

66

43

56

27

37

All

Manufacturing

Teams

55

TQM

34

Quality circles Job rotation

57 58 56

Sources: 1992 data: Osterman (1994); 1993 data: Lawler, Mohrman, Ledford (1995); 1994 data: Black and Lynch (1996); 1997 data: Osterman (2000). Note: Numbers are percent of establishments that have adopted ‘‘any use’’ of the HRM practice or that have 50 percent or more of their employees participating.

Table 6.3 Firms having 50 percent or more employees covered by EI, by occupation (in %) Manufacturing Customer service

62 56

Distribution

48

Marketing

47

Technical

49

Staff

53

First-level managers

55

Mid-level managers

58

Top managers

58

Source: Lawler, Mohrman, and Ledford (1995, p. 98).

Several surveys provide time series information on the evolution of innovative HRM practices. Lawler, Mohrman, and Ledford (1995, p. 47) report the year in which a TQM program was instituted in the firms in their survey, as follows: Prior to 1986

19%

1986–1988

25%

1989–1990

32%

1991–1993

24%

Because TQM tended to be the first type of Japanese EI practice that was introduced in the United States and other innovative HRM

242

Kathryn Shaw

Table 6.4 Evolution of EI (in %) 1987

1990

1993

Employee participation groups Zero participation 1–20% of employees participating 20% or more employees participating

30 33 37

14 35 51

9 26 65

Self-managed work teams Zero participation 1–20% of employees participating 20% or more employees participating

72 20 7

53 37 10

32 49 20

Movement of decisions to lower levels little change moderate change great change

19 63 19

16 60 24

10 63 28

Source: Lahler, Mohrman, and Ledford (1995, pp. 27, 28, 31). Note: The percentages show the percent of firms that fall into that category. For example, in 1987 30 percent of the firms had zero employees in employee participation groups.

practices followed (Cole 2000; Ichniowski and Shaw 2000), it is fairly safe to conclude that prior to 1980 a very small minority of firms had formal HRM systems encouraging employee involvement. Osterman (1994) reports that the vast majority of EI programs were introduced in the five years from 1987 to 1992: 49 percent of teams, 71 percent of problem-solving groups, and 68 percent of TQM programs were introduced during that period. There is also evidence on the evolution of EI in recent years. Comparing columns 3 and 7 of table 6.2, Osterman finds that from 1992 to 1997 there was a significant increase in the use of TQM, quality circles, and job rotation. Using survey data for very large firms from 1987 to 1993, Lawler, Mohrman, and Ledford show clear evidence of change on several of the elements that underlie EI (see table 6.4). Overall, participation groups have risen, and one very dramatic use of EI, the rearrangement of the work force into selfmanaged teams on the job remains relatively rare but has grown considerably. Finally large firms report that they are moving decision making down toward the production worker level. In sum, we could say that before the 1980s, almost no firms had HRM practices emphasizing employee involvement. Limited introduction occurred in the early 1980s; by 1990 approximately half of all


243

firms had some innovations. Since 1990 there has been limited but continued progress. Thus, to the extent that HRM represents a technology shock, this shock occurred and adjustments were made beginning in the mid-1980s and into the 1990s. This HRM shock changed the production process by elevating the involvement and incentives of production workers. Though a complete discussion of these effects is beyond the scope of this chapter, the goals of innovative HRM are evident as they interact with IT use, as described next.9 IT Use and Innovative HRM Practices as Complements Other researchers are now emphasizing the importance of redesigning jobs to integrate IT capital into the production and managerial processes of the firm (Breshnahan, Brynjolfsson, and Hitt, forthcoming; Brynjolfsson and Hitt 1998, 2000, 2001; Caroli and Van Reenen 1999). For example, if a firm puts in place a system of networked computers that provides real-time production data to the desks of all managers, then the managers in that firm must have their jobs redesigned so that they can make use of that data. As described for the steel industry, production workers should also have their jobs redesigned to make use of the information technology that now runs the production line. For example, if computers on that production line can automatically adjust the mechanics of the line as the product dimensions vary, then the production worker’s job changes to match these changes in job duties. Innovative HRM practices that raise employee involvement are often aimed at changing job design so that the worker has the ability, authority, and desire to make the changes on the line as needed. To give production workers the ability and authority, they should be better trained and selected, should sit on problem-solving teams, should rotate across jobs on the line for cross-training purposes, should have information on the goals and financial position of the firm, and should have the incentive pay that encourages appropriate decision-making action. This describes, in essence, a complementarity between IT and innovative HRM practices that was emphasized in my steel industry example above. Other researchers, who are known for their expertise on IT use, have reached the same conclusion regarding complementarity of IT and HRM practices:

244

Kathryn Shaw

. In a survey of computer use and organizational change for Fortune 1000 companies, Breshnahan, Brynjolfsson, and Hitt (forthcoming) and Brynjolfsson and Hitt (1998) find strong evidence that computer use and innovative HRM practices are correlated. (See also Greenan and Guellec 1998 for similar results using French data.) . In a survey of 135 retail banking establishments, Hunter and Laf-

kas (forthcoming) also find a correlation between computer use and employee involvement.10 Thus further evidence suggests that HRM practices and IT use tend to be complements on the job.11 How Do IT and Innovative HRM Practices Alter the Demand for Labor? At a general theoretical level, IT or HRM can alter labor demand through either of two mechanisms: they can alter the ways goods are produced (i.e., change the production functions for existing products), or they can alter the types of products that are produced (i.e., introduce new products with new production functions). The economic model of labor demand emphasizes both: labor demand is ‘‘derived demand’’ from the demand for the product, thus combining both the quantity demanded and the method of producing the products. If labor demand is derived demand, does labor demand shift within firms or across firms due to the product and process shocks from IT and innovative HRM? For example, do current manufacturing firms retool their production lines so that they can produce new products or can use a different production process? Or, alternatively, are new production lines built that embody the new production process and produce the new products? Surely both occur, but I will make the argument that extensive U.S. data show that new lines play a big role in introducing valuable IT innovations. Thus labor demand shifts across firms, and these new firms demand a different type of worker. Labor Demand Shifting Employment from Old to New Plants Establishment-level data suggest that many product demand changes have occurred within industries over time, and not across industries. Because the manufacturing sector has shrunk as a percent of total


245

employment in the United States, the demand for products has shifted across industries and services have risen relative to manufacturing. However, while there have been shifts across sectors, the IT examples given above suggest that within industry changes in product demand cause shifts from declining establishments to growing establishments. For example, Walmart department stores has displaced traditional department stores in the United States (where Walmart is a huge volume warehouse operation), and minimill steel plants have displaced older integrated steel mills (described further below). Consistent with these within-industry changes in product demand, the data show that there are dramatic within-industry changes in employment. The U.S. has very detailed data on manufacturing establishments from surveys conducted over the last thirty years, surveys asking for information on employment, capital, value added, and so on. Davis and Haltiwanger (1992) use these data, called the Longitudinal Research Database (LRD), to show that the variance of employment growth rates across establishments within industries is much greater than the variance of sectoral mean employment growth rates across industries. Or, looking within very narrowly defined industries, they find that a great deal of employment growth at some establishments co-exists with falling employment at others, even in industries that are declining in total employment. This vast amount of employment change across plants within industries is consistent with the story that innovations are very likely to be introduced in new or growing establishments. Alternatively, if firms were introducing changes in products produced or in production processes by retooling existing plants, there would be much less employment change across plants. A second piece of evidence suggests that most changes are implemented through the creation of new firms and new establishments, not from the retooling of older establishments within industries. Baily, Hulten, and Campbell (1995) use the LRD data to show that productivity gains within industries result from shifting production from plants that are less productive to plants that are more productive (not from the reinvestment in older plants to raise their productivity levels). For the most part, new plants replace old plants when productivity-enhancing activities, such as IT or innovative HRM, are introduced.

246

Kathryn Shaw

This evidence suggests that the biggest effects of IT on employment demand are occurring in growing establishments and new establishments, and to a lesser degree in existing establishments. Thus IT will affect employment demand as new employees at new establishments, replace employees who lose jobs at declining establishments. Labor Demand—Changes in Skills in New and Old Plants As described above, in U.S. firms the IT shocks and the HRM shocks occurred at about the same time. When these coincident shocks are coupled with the complementarities between IT and HRM practices, the demand for labor might shift toward a preference for more skilled labor. Employees who are using IT and have the HRM practices that provide the incentive and ability to do problem solving need different skills: they require, on average, more advanced cognitive skills, communications skills, negotiations skills, and interpersonal skills. It is difficult for firms to take existing employees and impart them with these new skills, because it requires both a change in innate ability as well as considerable changes in the way work is conducted. Therefore firms wishing to fully integrate IT in their workplace will often hire new employees. Evidence from the establishment-level data suggests that employee turnover is considerable in the United States, and therefore sufficient to accommodate a change in demand for more skilled labor. In the process of establishment births and deaths, employment constantly changes—old jobs are lost and new jobs are created—and the evidence from the establishment-level data suggests that the new jobs that are created will tend to be more skill intensive than the old jobs that are lost. First, the new jobs are associated with higher levels of general technology usage (Domes, Dunne, and Troske 1997) and with computer usage (Dunne, Foster, Haltiwanger, and Troske 2001). Second, new establishments are more productive than older establishments (Baily, Hulten, and Campbell 1995). Third, new plants are also associated with greater use of nonproduction workers and less use of production workers (same references, plus Berman, Bound, and Griliches 1994), implying a bias toward more skilled workers. Finally, wages are, on average, higher on the new jobs than on the old jobs.12 The establishment-level data also suggest that much of the skillupgrading takes place as new firms demand more skilled labor


247

relative to the demands of older plants, and that there is less skill upgrading within existing plants over time. Domes, Dunne, and Troske (1997) find little correlation between skill upgrading and the adoption of new technologies using longitudinal data following particular manufacturing plants as they age. Together, these results suggest that there is more upgrading of both technology and worker skills in new plants than in old, and in part, this may explain some of the ‘‘productivity puzzle’’ of IT. The productivity puzzle is that vast amounts of IT investments have been made, but measured productivity gains are small. Our work on the steel industry would concur with the conclusion of Breshnahan, Bryjolffson, and Hitt (forthcoming) that the IT productivity payoff has, in part, been smaller in older firms who have not managed to redesign jobs and introduce innovative HRM practices that can best utilize the higher IT levels.13 New firms that hire more skilled workers with different job designs see higher performance levels. Autor, Levy, and Murnane (2001) provide evidence that the demand for skills has changed over time due to the use of computer capital. They match data on job tasks by occupation from the Dictionary of Occupational Titles to data on individuals’ occupations from 1960 to 1998. They find that the demand for routine manual or routine cognitive tasks has diminished over time, but demand for nonroutine cognitive tasks has risen, even within occupation and education groups. Howell and Wolff (1992) did a somewhat different earlier study of the requirements of jobs using individual-level data matched to DOT data and also found evidence of rising demand for cognitive skills. Finally, using data on job activities for individuals in French establishments, Maurin and Thesmar (forthcoming) find that same outcome of rising demand for cognitive skills in French manufacturing. To determine whether new hires have different skills than existing employees, another set of new surveys asks employers about the skills they seek in new hires. Black and Lynch conducted an extensive establishment survey in 1994, asking establishments to rank on a scale of 1 to 5 the hiring criteria that is most important to them. As summarized in table 6.5, these firms place the most emphasis on the applicant’s attitude toward work, communication skills, and work experience, with very little emphasis placed on education and school grades.

248

Kathryn Shaw

Table 6.5 Criteria used in hiring new employees Applicant’s attitude

4.6

Communication skills

4.2

Work experience

4.0

Outside recommendations

3.4

Industry credentials

3.2

Years of education Employer tests

2.9 2.5

Applicant school grades

2.5

Source: Black and Lynch (1996, p. 266). Note: The numbers are the mean values from a survey of establishments, where the establishments ranked the criteria on a scale of 1 to 5 (5 being most valuable).

Table 6.6 Job activities for new hires (in %) Occupation Task

Profession/ Manager

Clerk/ Sales

Service

Craft/ Operator

Laborer

Customers

80

84

78

41

31

Read

88

67

58

54

34

Math

76

70

52

62

53

Computers

66

75

23

23

17

Source: Holzer (1998, table 1). Note: Percentages are of all new jobs where the new employee will have to undertake the listed activity: customers (job dealing with customers), Read and Math (uses of reading or writing and arithmetic), and computers (use of computers).

Holzer (1998) conducted a survey of establishments in four major U.S. cities and asked about the types of workers they seek in new hires. Holzer was asking about the daily tasks on jobs, to ascertain the skills required by employers of their new hires. His results are in table 6.6, with breakdowns by occupation. He finds that for all craft/ operative job openings, about 41 percent of new employees will need to work with customers, 54 percent will need to read on the job, 53 percent will need to do math, and 23 percent will work with computers. Overall, the skill demands are fairly extensive for most jobs; only 3 percent of all jobs have virtually no requirements. Furthermore he finds that across all jobs, about 50 percent require some computer use, and this is similar to that reported in Kreuger (1993).


249

The key question is, How have hiring demands changed over time as IT use has grown? There are no detailed time series data on task use, but Holzer does ask some questions about changes over time and estimates that, on average, all the tasks above experienced about a 23 to 25 percentage point increase from 1980 to 1994. These numbers suggest large increases in computer use as well as much more emphasis placed on the basic skills that are necessary in firms that require more employee involvement or more employee autonomy.14 The evidence suggests therefore a rising skill bias in new hires over time. However, is the skill bias associated with new IT and HRM practices? In Osterman’s (1994) survey of private sector establishments, he finds a very consistent positive correlation between all measures of innovative HRM practices and the skill content of the typical ‘‘core’’ occupation at the plant. That is, plants that tend to have more skilled jobs are more likely to have innovative HRM practices. At some plants, skill content is higher for production workers. Other establishments comprised of professional and technical employees also have higher skill levels. Innovative HRM practices also require different social and communication skills of professionals and technical workers who must interact in teams. Teamwork has clearly entered the realm of the well educated. For example, when Clark and Fujimoto (1991) wrote their book on product development teams in Japanese auto companies, the Japanese used teamwork to develop new automobiles twice as quickly as the U.S. automakers. Today U.S. automakers use comparable teams for product development. In sum, the establishment data show that firms have elevated the hiring criteria for new jobs, to place greater emphasis on communications, attitude toward work, reading and math, computer work, and the ability to deal with customers. As in the steel industry example of section 6.1, firms now screen applicants more carefully. The establishment data are not sufficient to show that these changes were caused by elevated IT or HRM innovations. However, the skills that firms are seeking are very consistent with the needs of the new IT/ HRM environments that induce employees to do more problemsolving in a team environment, to work smarter and to work harder, and to take greater responsibility in their day-to-day decision making. Firms are less interested in the employee’s education or expe-

250

Kathryn Shaw

rience, though greater education and experience could certainly improve workers’ analytical or interpersonal skills and the use of IT. 6.3

Other Shocks to Labor Demand

Though I have emphasized the effects of IT and innovative HRM practices and their impact on the changing nature of labor demand, many other changes have affected labor demand over this time. It is now time to put IT in context with other shocks to labor demand that may have affected wages and employment. We want to see how large a role IT plays in altering the demand for labor. I start again with the steel industry example before summarizing the general evidence briefly. Shocks to Employment in the U.S. Steel Industry The steel industry example is again instructive for the level of detail and depth of understanding that it provides. When looking at data across industries, researchers must be content to measure shocks to employment with very general measures of shocks: import penetration, unionization, R&D, and computer use on the job. When looking at steel industry data, the shocks are more specific and thus more identifiable as causal factors (see Beeson, Shaw, and Shore-Sheppard 1999 for further evidence). In the U.S. steel industry, changes in IT and HRM practices would be considered only one set of contributors to the declining demand for labor in steel, and not the most important set. Employment in the U.S. steel industry dropped from 590,000 in 1965 to 143,000 in July 2001, but most of the decline occurred in the 1980s (falling from 450,000 in 1979 to 170,000 by 1987). The employment losses in the industry came from a combination of widespread plant closures (Deily 1991; Beeson and Giarrantani 1993) and sizable employment reductions at remaining mills, where employment reductions arose from shutting down areas of the mill, or from outsourcing some jobs, or from simply using retirements to cut the labor-to-capital ratio. All of these job losses are most likely due to a combination of the following factors.

. Technological change that improved the performance of the electric arc furnace caused the rise of minimills in the United States. These minimills use steel scrap (e.g., from cars) to make steel rather


251

Table 6.7 Steel production U.S. production (million tons) Integrated

Minimill

Total

Percent minimill

Import Sharea Share

Percent

1961

88.5

8.7

98.0

8.9

1960–64

2.5

1965

117.1

22.9

131.5

17.4

1964–69

10.1

1970

111.3

20.2

131.5

15.3

1970–74

10.6

1975

94.0

22.7

116.6

19.4

1975–79

13.3

1980

80.6

31.2

111.8

27.9

1980–84

18.4

1985

58.3

29.9

88.3

33.9

1985–89

19.6

1991 1995

52.7 62.5

33.8 42.4

87.9 104.9

38.5 40.4

1990–94 1995–97

16.6 21.2

1997

60.6

46.9

107.6

43.6

Sources: For U.S. production, AISI reports. For imports: Deily (1994, p. 5) and recent AISI monthly reports. a. Import Share is (imports-exports)/(U.S. production).

than using raw materials (e.g., iron ore), so minimills produce lowergrade steel that is used for bar products, like re-bar or structural steel used in construction. As these minimills grew, integrated mills that produced these same products could not compete, and were shut down. Production shifted dramatically from integrated to minimill producers in the 1970s and 1980s, when minimill production rose from 15 percent of production in 1970 to 44 percent by 1997 (see table 6.7). The integrated mills were much more labor intensive, so employment fell. Technological change that improved the performance of the capital stock in integrated mills, and minimills lowered the demand for labor (see table 6.8). The most significant technological change was the introduction of continuous casters in place of ingot production, thus shortening the steelmaking process by one full stage. The percent of mills using continuous casters rose from 22 percent in 1960 to 100 percent today, and a large number of these casters were installed in older mills in the 1970s and 1980s (Beesen and Giarratani 1998).

. Imports made inroads into U.S. markets, peaking first in 1985 be-

fore declining and then growing again in the last few years as steel demand has surged or as the dollar rose in value. However, the causal relationship is unclear—the largest increases in imports were in periods of rising employment, and not declining employment.

252

Kathryn Shaw

Table 6.8 Technological change in steel production 1965

1975

1972

1985

10.0

5.0

1.5

2.9

1.75

0.9

Electric furnace Tap-to-tap (hours per heat) Man-hours per ton Hot-rolling mill Speed (feet per minute)

10,000

20,000

Man-hours per ton

1.65

0.85

Electricity use (k hours per ton)

180

160

Source: Barnett and Crandall (1986, pp. 57, 59).

Nevertheless, the increase in imports may have served as a wake-up call to U.S. steel producers to cut labor costs and increase efficiency.

. Integrated mills underinvested in new capital for many years, and thus they had a older capital stock that could not compete well with imports and minimills, but also required extensive investments to meet pollution reduction standards.

. Integrated steel producers experienced very significant wage in-

creases in the early 1980s, at the same time that the U.S. recession lowered the demand for workers. The vast majority of plant closures occurred in the early 1980s, when the recession tipped poorly performing plants to closure. I list these sources of changes in labor demand to make three points: (1) many factors simultaneously altered labor demand, and it would be nearly impossible to disentangle their effects in steel; (2) significant changes in labor demand took place within the industry (while many integrated mills shut down and laid-off employees, new minimills were opening); and (3) cross-industry analysis would not have uncovered these true sources of change, because the first two factors listed above are ‘‘technological’’ change that are difficult to measure using available data across industries. Overall, within the steel industry, there was a dramatic drop in employment as well as a shift in employment from old integrated mills to new minimills. Since the integrated mills tended to lay off the less-educated older workers, and the new mills tend to hire the more skilled workers due to IT and HRM factors, there is a positive skill bias in employment in the industry.


253

Shocks to Employment: The Colinearity of Causal Factors When using aggregate, cross-industry data to discern causes of rising inequality, IT shocks that affect labor demand are often not separable from other shocks that may also be skill biased. Income inequality researchers have attempted to find the causal factors that have contributed to this ‘‘skill bias’’ in labor demand and thus income inequality, by introducing measures of technology shocks or sectoral shifts in models of wage or income determination. In addition to computer technology shocks, the typical list of factors that have contributed to rising inequality are:

. Technology shocks to firm-level production functions, arising from new R&D.

. Rising levels of exports (where exports are produced by firms that

use more skilled labor).

. Changes in product demand such that some manufacturing indus-

tries are declining while others are growing.

. Rising levels of imports or of off-shore production that lowers the demand for less-skilled workers. . Institutional changes, like the decline of union power and the rise

of contracting-out of jobs. Though surely these effects do vary across industries, and across firms within industries, it is likely to be very hard to separate their effects. Consider the following linkages: computer use is often intertwined with new technology or new product development (arising often from R&D); firms that export often are larger, more technology intense, and utilize more R&D and computerization; firms that face the threat of imports in their domestic markets are likely to use new technology, R&D, and HRM practices to retain domestic market share; and the decline of union power tends to arise from the above shifts in demand. Much evidence reinforces these linkages. Dunne et al. (2001) show that new plants in manufacturing tend to be those plants with high-skill, high-wage workers who work with more capital and more computer technology. Similarly Jensen and Troske (1997) emphasize that the use of skilled workers, greater technology use, and exporting output are all highly correlated across manufacturing plants, and thus that it will be difficult to disentangle causal relationships.

254

Kathryn Shaw

The best way of disentangling the different causes of changes in employment demand would be to have longitudinal data on firms or plants and on the shocks to these firms or plants over time. The LRD manufacturing database does provide longitudinal data on the U.S. manufacturing plants. However, the data have limited measures of skill—containing only nonproduction versus production workers. And there are limited time series data on demand shocks— containing only data on investments (including computers) and exporting output, but omitting detailed computer data, R&D, and organizational innovations. Using these LRD data, researchers have shown that most changes in labor demand have occurred across plants rather than within plants over time (as described above), calling into question the causal relationship between IT use and the increase in the demand for skilled labor within plants. Across plants, most of the increase in the demand for skilled labor comes from the greater demand for nonproduction workers at new plants. Since new plants tend to pay higher wages, export more, and use all forms of new technology (capital, IT, and organizational), it is difficult to find causal relationships. There is far less evidence that the increasing use of IT within a plant causes an increase in the demand for skilled labor (Dom, Dunne, and Troske 1997). 6.4 Lessons Learned: IT Capital and Knowledge Capital as Complements To explain the rising income inequality worldwide, or the changes in employment and wages, we are looking for the ‘‘technology shocks’’ that may have altered labor demand and thus incomes. The evidence presented above suggests that there have been two types of complementary technology shocks: IT shocks and innovative HRM shocks. In this section I want to suggest that the production function has changed over time to admit two new types of capital. Information technology is undisputedly associated with a new form of capital—computers which are very tangible capital investments. However, as one puts this capital in the production function, the definition of computers as capital is less clear. To turn computing power into final output, the firm must adequately manage the use of the information that comes from the computing power. Therefore in our production function we will insert IT to imply that the firm


255

has capital resources that are tangible (i.e., physical) computing power and that the firm has invested in the use of information from computers—this, of course, is a capital resource as well. Firms that have purchased computers but have done an inadequate job of integrating their information in the production and management processes will have lower returns to their IT capital. The second form of capital that innovative firms invest in is knowledge capital. Knowledge capital is the knowledge base that improves firm performance after controlling for the other standard inputs of physical capital and labor, and thus is a very broad term.15 It can refer to the research and development knowledge that the firm possesses and will develop, or it can refer to the ‘‘intellectual capital’’ of the firm that represents the knowledge and skills of the scientists, engineers, or marketing researchers.16 Increasingly firms are thinking about knowledge capital as the set of fixed labor resources that offers the firm a strategic advantage, and for some firms, talented production workers offer that advantage. When innovative HRM practices are combined with IT to give production workers the opportunity and incentive to acquire knowledge and make use of it on the production line, the firm is developing new knowledge capital. Thus the means by which IT and HRM change the demand for workers is that they introduce production-worker knowledge capital into the production function and thus require somewhat more skilled labor. The effects on employment and wages follow from this change. 6.5

Lessons Learned: Employment and Wage Effects

In this chapter I have used case study evidence and references to establishment-level data to describe the ways in which IT may have increased the demand for skilled labor, particularly in manufacturing. The increased demand for skilled production workers has resulted in part from the direct effect of rising levels of computerization—as production workers utilize computers more, they need higher skill levels, such as the ability to read and do simple math. However, the primary change in demand may arise from indirect effects—as production workers are given more problemsolving authority, due to the introduction of innovative HRM practices that encourage employee involvement and that complement higher IT use, the demand for skills rises.

256

Kathryn Shaw

What are the effects on wage rates? In the steel industry there has been virtually no change over time in the wages of production workers—both new hires and older workers continue to be highly paid for their educational level. What has changed for the steel industry is the access to these high-paying jobs. Twenty years ago, any man could obtain employment, but today to obtain the high-paying jobs, the high school educated applicant must also have good team performance and communications skills. Therefore nowadays the low-skilled high school educated person does not have access to manufacturing jobs, and must turn to lowerpaying jobs outside manufacturing. Moreover, when workers hold manufacturing jobs and are permanently laid off from these jobs, their pay on new jobs is typically 20 to 40 percent lower than it was in manufacturing (Kletzer, forthcoming, table 6.6). Thus wage inequality rises for two reasons. First, the decline of the manufacturing sector has contributed to wage inequality for lessskilled men, as they must seek employment elsewhere at lower wages (see Bernard and Jensen 1998 for a focus on sectoral effects). Second, within manufacturing sectors, inequality has risen. On jobs with more IT and innovative HRM practices, there is greater demand for higher-skilled production and nonproduction workers who are paid well. The data for manufacturing shows that within establishments, the pay of nonproduction workers has risen relative to the pay of production workers (Bernard and Jensen 1997). But on other jobs that have less IT capital, skills are diminished and pay is falling. These results are consistent with the economics literature that shows widening wage inequality across sectors and within sectors in the United States. IT and innovative HRM also contribute to another aspect of wage inequality, the growing variance of wages within the group of high school educated men in the United States. The case study evidence suggests that when the demand for skilled production workers rises, employers are often raising the demand for skills that are difficult to observe and quantify. In manufacturing, the average educational level has risen over time (Katz and Murphy 1992). But equally important, even when establishments continue to demand high school educated employees, they are looking for an additional set of skills not previously demanded. They are looking for problem-solving skills, communications skills, and teamwork skills. For high school educated people having these skills, wages will be high. For high-


257

school educated people with poor skills, wages will be low as they work with less IT and less innovative HRM practices. Thus, the variance of wages rises over time among the high school educated due, in part, to IT use. I want to conclude that within manufacturing, IT probably has increased inequality but not played a big role.17 The biggest contributor to the rising wage inequality in manufacturing is probably the loss of good jobs for high school educated production workers. As is described in section 6.3, most job losses appear to be from industryspecific shocks to technology that may or may not be correlated with IT use. 6.6

Conclusion

Using evidence from the steel industry and from establishment data, I have described ways in which IT use can contribute to an increased demand for more skilled labor in manufacturing. The reason for the increasing skill bias is that IT tends to place more decision-making authority in the hands of production workers, and thus to require more skilled production workers. However, the skills required are different than those in the past. Rather than mechanical skills (or physical dexterity), the new skills that are needed are communications skills and problem-solving skills. This is particularly true in establishments that have introduced innovative HRM practices as they introduce IT. Innovative HRM practices—like problem-solving teams, job rotation, enhanced training, and more information sharing—serve to complement IT use and to require these nonmechanical skills. It is also particularly true for new establishments, because older establishments find it much more difficult to make the organizational changes that accompany the higher skill levels for production workers. Finally, as firms require more cognitive and communications skills of production workers, the firms are attempting to build their ‘‘knowledge capital’’ among production workers. Knowledgable production workers who have greater access to IT can raise the productivity of the firm beyond what is feasible for the more traditional, less-innovative, firm. Though the evidence suggests that IT and innovative HRM contribute to a rising demand for more skilled production workers, there are certainly areas of manufacturing where this would not be the case. When IT is coupled with a greater mechanism and control

258

Kathryn Shaw

of the production line, the production workers require fewer skills (see the semiconductor ‘‘fab’’ example of Brown and Campbell 2001). Furthermore not all manufacturing plants ought to adopt innovative IT or HRM practices. Our minimill study finds that those who gain the most from innovative practices are those that have the most complex products or complex production lines (Boning, Ichniowski, and Shaw 2001), whereas lines that produce low-cost commodity products that change little over time have much less need for innovative practices. In this chapter my goal has been to discuss the means by which IT contributes to changing labor demand, but this discussion also has implications for estimating the size of the IT impact. The establishment data and steel-industry data suggest that innovative practices are most likely to be introduced through the growth of new plants, and less through changes within plants over time. As a result it is more difficult to use longitudinal establishment-level data to estimate the effects of shocks on labor demand. If the shocks cause shifts in demand across firms, and not within firms over time, the crosssectional results are the more relevant results for estimating the impact of shocks on labor demand (though they admittedly suffer from other biases as well). This should be kept in mind when reading the Chennels and Van Reenen chapter (this volume) evaluating the size of the IT impact on labor demand. I have also emphasized that the shocks to IT and innovative HRM practices may be highly correlated with other factors that also increase the variance of wages. In the steel industry, the decline of ‘‘good jobs’’ in integrated steel mills was caused primarily by other technology shocks (not IT), and when workers lose these highpaying jobs, they typically find work at much lower wage levels. Across the industries in manufacturing, it is difficult to assign a value to the impact of IT on wage inequality, because across and within establishments IT is commonly correlated with firm size, R&D activity, and greater exporting. Among high school educated workers, IT may be a contributing factor to rising inequality among new hires, because those with the personal skills to obtain jobs in manufacturing will receive considerably higher wages than the less skilled. Finally, I have focused on manufacturing because there are so little data or evidence outside manufacturing on the means by which


259

IT could have changed labor demand. Outside manufacturing, the data shows that IT levels and innovative HRM practices have risen (Breshnahan, Brynjolffson, and Hitt, forthcoming; Osterman 2000). However, outside manufacturing there is less case study evidence, less establishment-level data on which to rely, and our own research on the steel industry does not provide insight into nonmanufacturing environments. Anecdotal evidence provides examples of instances in which greater IT use would raise the demand for less-skilled labor: the demand for telemarketing firms has clearly risen, and these employees tend to be less skilled; in retail stores, sophisticated cash registers require less skills to operate than did older registers and retailing has grown over time. At the same time other anecdotal evidence provides examples in which IT raises the demand for skilled labor: business and health services have grown and utilize employees who work with growing levels of information and computerization. These two streams of anecdotal evidence may provide the backdrop for the rising variance of incomes outside of manufacturing: IT has contributed to an increase in both the demand for lessskilled low-wage labor and the demand for higher-skilled high-wage labor, thus increasing the observed variance of wages. In two cases, car dealerships and bank-check processing, there is evidence of greater skill demand due to IT use in some jobs, and no change in others (Levy et al. 1999; Autor, Levy, and Murnane 2000). Notes The chapter was prepared for the conference on Information and Communications Technologies, Employment, and Earnings, sponsored by CSERC, June 22–23, 1998, Nice, France. I would like to acknowledge the superb support by the Alfred P. Sloan Foundation for my ongoing research on the steel industry and income inequality. Also my co-authors on this steel work have contributed very significantly to the research that I refer to herein and to my understanding of these issues, though they are in no way responsible for the views I express below. I thank Patty Beeson, Brent Boning, Casey Ichniowski, Giovanna Prennushi, and Lara Shore-Sheppard for their assistance and ongoing collaboration. 1. For evidence on rising income inequality, see Adams (1997), DiNardo, Fortin, Lemieux (1996), DiNardo and Pischke (1997), Freeman (1996), Juhn, Murphy, and Pierce (1993), Krueger (1993), and the reviews by Katz and Autor (1999), Gottschalk (1997), and Levy and Murphy (1992). For evidence on the effects of technology on income inequality, see the reviews by Autor, Katz, and Krueger (1997) and Chennells and Van Reenen (this volume) and Spiezia and Vivarelli (this volume), and the evidence on services by Kaiser (this volume), and the international evidence by Berman, Bound, and Machin (1997).

260

Kathryn Shaw

2. See Ichniowski, Shaw, and Prennushi (1997), Ichniowski and Shaw (1995), Boning, Ichniowski, and Shaw (2001), and Gant, Ichniowski, and Shaw (forthcoming). 3. The minimill data described above are from Boning, Ichniowski, and Shaw (2001). 4. During our plant visits, others have argued that production jobs in mills today require fewer skills. In the past, when employees manually adjusted the machines, they had to have a great deal of plant-specific knowledge about the proper adjustment as conditions varied. Today the computer has that knowledge. However, today the levels of required computer and mathematical skills are greater than in the past, as production workers have more responsibility for fundamental decisions. 5. In the steel industry educational levels have risen modestly. Based on CPS data, the percent of workers with a high school education or lower fell from 74 percent in 1980 to 67 percent in 1990, though the production/nonproduction worker ratio was unchanged (based on LRD data). 6. See Berg, Applebaum, Bailey, and Kalleberg (1996) and Dunlop and Weil (1996). 7. Osterman (1994, 2000) surveys establishments in 1992 and 1997, obtaining responses from 875 and 683 establishments with 50 or more employees in the nonagricultural private sector. To obtain data that accurately depicts the state of the HRM practices, Osterman needed to identify a set of workers to whom the survey addressed. He asked that the survey respondent be in the operations side of the business and that the respondent identify the largest group of nonsupervisory, nonmanagerial workers at the establishment’s location who are directly involved in making the product or providing the service. Lawler, Mohrman, and Ledford (1995) conducted a series of surveys in 1987, 1990, and 1993, but they focused only on the top firms of Fortune 1000 and directed their survey at the firm level, not the establishment level. However, their results are relatively consistent with those of Osterman, as identified by Osterman. Black and Lynch (1996) describe the results of a survey of HRM practices that was conducted under their guidance by the U.S. Bureau of the Census in 1994 using a nationally representative sample of establishments (of 3,000 private establishments with a response rate of 75 percent for manufacturing). They report somewhat higher numbers than those of the above-mentioned surveys, but their survey is two years later and, as usual, has different questions. Applebaum and Batt (1994) provide further evidence. 8. The overall measures are the use of teams, job rotation, total quality management, and quality circles. Quality circles were the first attempts at offline problem solving efforts (but tended to focus on employee concerns and not firm performance improvement). TQM has had a more significant impact, with its focus on improving quality, and it is included because TQM programs often signaled a need to change other HRM practices (though in the steel industry, we find that the changes in other practices followed the introduction of TQM). The Lawler, Mohrman, and Ledford (1995) survey results document that employee involvement is an important part of TQM (pp. 78–85), particularly in successful TQM programs. 9. For further discussion of the effects of HRM practices on performance, see, for example, Appelbaum et al. (2000), Batt (1995, 1999), Black and Lynch (forthcoming), Huselid and Becker (1997), Dunlop and Weil (1997), Ichniowski et al. (1995), Ichniowski, Shaw, and Prennushi (1997), Levine (1995), and MacDuffie (1995). 10. For other evidence on adoption rates, see Hwang and Weil (1996) and Nickell, Nicolitsas, and Paterson (1997). Hunter and Lafkas (forthcoming) show that when


261

employee involvement is interacted with computer use in automating or ‘‘infomating,’’ wages are likely to be higher (than do to the independent effects) in retail banking outlets. 11. These two factors—innovative HRM and greater computerization—can have either complementary or independent effects on the demand for more skilled workers. For the actual discussion of complementarity, see Acemoglu (1998), Athey and Stern (1998), and Milgrom and Roberts (1992). In some companies innovative HRM will be combined with innovative information technology as complements: information technology provides the information that supports greater employee participation and greater participation by employees at all levels makes better use of the available technology. On the other hand, when innovative HRM practices were first introduced in the United States by the Japanese, the gap in HR practices was greater than the gap in technology between the two countries: innovative practices can be valuable in lesscomputerized environments. Furthermore information technology can reduce the value of employee involvement and lower the demand for skilled workers. Consider the case of telemarketing, where telecommunications (and rapid, technology-driven air freight) have created sizable numbers of low-skilled (and low-paying) jobs that are less likely to introduce innovative HRM practices. One possible measure of the greater involvement of production workers comes from the data on ‘‘de-layering.’’ Among the Fortune 1000 firms in the Lawler, Mohrman, and Ledford survey, 72 percent stated that they had reduced the number of managerial ‘‘layers’’ in the previous ten years, and 43 percent stated that they had removed two or more layers of management. Reducing the number of layers of management can often imply that jobs are redesigned so that production workers are given more responsibility or decision-making power. In both the minimill and the integrated segments, our interviewees repeatedly stated that they are running much leaner today than in the past—the middle-level managers are simply gone. 12. There has been a large increase in the dispersion of wages across plants over time, but little increase in wage dispersion within plants (Davis and Haltiwanger 1991; Dunne, Haltiwanger, and Troske 1997; Domes, Dunne, and Troske 1997). 13. Other capital investments must also be made. See the paper on services by Licht and Moch (1997). 14. See also the work of Entorf and Kramarz (1997), Entorf, Gollac, and Kramarz (1997), and Greenan, Mairesse, and Topiot Bensaid (1998). Of course, computer use alone could have caused the increases in the demand for other skills if computer users must undertake more skilled tasks. 15. Knowledge capital will also enter the firm’s maximizing framework in a variety of ways. For example, it can be used to lower costs or to increase the value of the firm’s strategic behavior. 16. The terms knowledge capital or intellectual capital have been developed in both the academic literature (e.g., Griliches 1979) and in the more popular business literature (e.g., Kelley 1998). 17. I have focused on production workers in manufacturing, and the numbers of production workers certainly dominate the numbers of nonproduction workers in manufacturing. But the evidence of rising skill demand is likely to apply to nonproduction workers in manufacturing as well. The employment of nonproduction workers relative to production workers has certainly risen over time in manufacturing, thus raising

262

Kathryn Shaw

the variance of wages in manufacturing (Bernard and Jensen 1998). Nonproduction workers are also more likely to be expected to work in autonomous work teams, and the rising levels of information in firms is likely to increase the demand for the quality of these workers as well as the quantity, thus possibly raising the variance of wages for nonproduction workers.

References Acemoglu, D. 1998. Why do new technologies complement skills? Directed technical change and wage inequality. Quarterly Journal of Economics 113 (November): 1055–89. Adams, J. D. 1997. Technology, trade, and wages. NBER Working Paper 5940. Cambridge, MA. Allen, S. G. 2001. Technology and the wage structure. Journal of Labor Economics 19(2): 440–83. Appelbaum, E., and R. Batt. 1994. The New American Workplace: Transforming Work Systems in the United States. Ithaca, NY: Cornell University Press. Appelbaum, E., T. Bailey, P. Berg, and A. Kalleberg. 2000. Manufacturing Advantage: Why High-Performance Work Systems Pay Off. Ithaca: Cornell University Press. Arthur, J. 1994. Effects of human resource systems on manufacturing performance and turnover. Academy of Management Journal 37: 670–87. Athey, S., and S. Stern. 1998. An empirical framework for testing theories about complementarity in organizational design. NBER Working Paper 6600. Autor, D., L. Katz, and A. Krueger. 1998. Computing inequality: Have computers changed the labor market. Quarterly Journal of Economics 113(4): 1169–1214. Autor, D., F. Levy, and R. Murnane. 2000. Upstairs, downstairs: Computer-skill complementarity and computer-labor substitution on two floors of a large bank. NBER Working Paper. Cambridge, MA. Autor, D., F. Levy, and R. Murnane. 2001. The skill content of recent technological change: An empirical exploration. Working Paper. June. Bailey, T. 1993. Organizational innovation in the apparel industry. Industrial Relations 32: 30–48. Barnett, D. F., and R. W. Crandall. 1986. Up from the Ashes: The Rise of the Steel Minimill in the United States. Washington, DC: Brookings Institute. Bartel, A., and F. Lichtenberg. 1987. The comparative advantage of educated workers in implementing new technology. Review of Economics and Statistics 49: 1–11. Batt, R. 1995. Performance and welfare effects of work restructuring: Evidence from telecommunications services. Ph.D. dissertation. Sloan School of Management. MIT. Batt, R. 1999. Work organization, technology and performance in customer service and sales. Industrial and Labor Relations Review 52(4): 539–64. Beeson, P., and F. Giarratani. 1998. Spatial aspects of capacity change by U.S. integrated steel producers. Journal of Regional Science 38: 425–44.


263

Beeson, P., K. Shaw, and L. Shore-Sheppard. 1999. Changes in the distribution of wages: Evidence from the steel industry. Working paper. Berg, P., E. Appelbaum, T. Bailey, and A. Kalleberg. 1996. The performance effects of modular production in the apparel industry. Industrial Relations 35: 356–73. Berman, E., J. Bound, and Z. Griliches. 1994. Changes in the demand for skilled labor: Evidence from the annual survey of manufacturing. Quarterly Journal of Economics 109(2): 367–97. Berman, E., J. Bound, and S. Machin. 1998. Implications of skill-biased technological change: International evidence. Quarterly Journal of Economics 113: 1245–79. Bernard, A. B., and J. B. Jensen. 1997. Exporters, skill upgrading, and the wage gap. Journal of International Economics 42: 3–31. Bernard, A. B., and J. B. Jensen. 1998. Understanding increasing and decreasing wage inequality. (January). Biesen, R. 1996. Steel town America. NewSteel 12 (September): 24–34. Black, S. E., and L. M. Lynch. 1996. Human capital investments and productivity. American Economic Review (May): 263–67. Black, S. E., and L. M. Lynch. Forthcoming. How to compete: The impact of workplace practices and information technology on productivity. Review of Economics and Statistics. Boning, B., C. Ichniowski, and K. Shaw. 2001. Opportunity counts: Teams and the effectiveness of production incentives. NBER Working paper 8306. Cambridge, MA. Bound, J., and G. Johnson. 1992. Changes in the structure of wages during the 1980s: An evaluation of alternative explanations. American Economic Review 82: 371–92. Bresnahan, T. F. 1999. Computerization and wage dispersion: An analytical reinterpretation. Economic Journal 109: 390–415. Breshnahan, T., E. Brynjolfsson, and L. Hitt. 2000. Technology, organization, and the demand for skilled labor. In M. Blair and T. Kochan, eds., The New Relationship: Human Capital in the American Corporation. Washington, DC: Brookings Institution Press, pp. 145–184. Breshnahan, T., Brynjolfsson, E., and L. M. Hitt. Forthcoming. Information technology, work place organization and the demand for skilled labor: Firm-level evidence. Quarterly Journal of Economics. Brown, C., and B. Campbell. 2001. Technical change, wages, and employment in semiconductor manufacturing. Industrial and Labor Relations Review 54(2A): 450–65. Brynjolfsson, E., and L. Hitt. 1998. Information technology and organization design: Evidence from micro data. Working paper. Sloan School of Management. MIT. Brynjolfsson, E., and L. Hitt. 2000. New evidence on the returns to information systems. Working paper. Sloan School of Management. MIT. Brynjolfsson, E., and L. Hitt. 2001. Beyond computation: Information technology, organizational transformation and business performance. Journal of Economic Perspectives 14(4): 23–48.

264

Kathryn Shaw

Clark, K., and T. Fujimoto. 1991. Product development performance: strategy, organization, and management in the world auto industry. Boston: Harvard Business School Press. Cole, R. E. 2000. Market pressures and institutional forces: The early years of the quality movement. In R. Cole and R. Scott, eds., The Quality Movement in America: Lessons from Theory and Research. Thousand Oaks, CA: Sage Publications, pp. 67–88. Council of Economic Advisers. 2001. Economic report of the President. Washington, DC: United States Government Printing Office. Cutcher-Gershenfeld, J., M. Nitta, B. Barrett, N. Belhedi, J. Bullard, C. Coutchie, T. Inaba, I. Ishino, S. Lee, W.-J. Lin, W. Mothersell, S. Rabine, S. Ramanad, M. Strolle, and A. Wheaton. 1994. Japanese team-based work systems in North America: Explaining the diversity. California Management Review 37(1): 42–64. Davis, S., and J. Haltiwanger. 1991. Wage dispersion between and within U.S. manufacturing plants: 1963–1986. Brookings Papers on Economic Activity; Microeconomics: 115–200. Deily, M. B. 1994. Wage changes in the steel industry 1970–1990. Working paper. Deily, M. B. 1991. Exit strategies and plant-closing decision: The case of steel. Rand Journal of Economics 22(2): 250–63. Deily, M. B., and W. B. Gray. 1989. Enforcement of pollution regulations in a declining industry. Working Paper 8912. Federal Reserve Bank of Cleveland. DiNardo, J., N. M. Fortin, and T. Lemieux. 1996. Labor market institutions and the distribution of wages, 1973–1992: A semiparametric approach. Econometrica 65: 1001– 44. DiNardo, J., and J.-S. Pischke. 1997. The returns to computer use revisited: Have pencils changed the wage structure too? Quarterly Journal of Economics 112: 291–303. Domes, M., T. Dunne, and K. R. Troske. 1997. Workers wages, and technology. Quarterly Journal of Economics 112: 253–90. Dunlop, J., and D. Weil. 1996. Diffusion and performance of modular production in the U.S. apparel industry. Industrial Relations 35 (July): 334–54. Dunne, T., L. Foster, J. Haltiwanger, and K. Troske. 2000. Wages and productivity dispersion in U.S. manufacturing: The role of computer investment. Working paper. Entorf, H., and F. Kramarz. 1997. Does unmeasured ability explain the higher wages of new technology workers? European Economic Review 41: 1489–1509. Entorf, H., M. Gollac, and F. Kramarz. 1997. New technologies, wages and worker selection. Discussion Paper 1761. Center for Economic Policy Research. Farber, H. S. 1996. Job creation in the United States: Good jobs or bad? Working paper. Freeman, R. B. 1996. Labor market institutions and earnings inequality. New England Economic Review (May–June): 157–68. Fruehan, R. J. 1993. Effects of emerging technologies on competitiveness in the steel industry. Global Study of the Steel Industry Working Paper. Gant, J., Ichniowski, C., and K. Shaw. Forthcoming. Social capital and organizational change in high-involvement and traditional work organizations. Journal of Economics and Management Science.


265

Garvey, R. 1996. Steel forum: Who’s to blame for steel’s bad image. NewSteel 12 (September): 100. Gibbons, R., and L. Katz. 1992. Does unmeasured ability explain inter-industry wage differentials? Review of Economic Studies 59: 515–35. Goldin, C., and L. Katz. 1998. The origins of technology skill complementarity. Quarterly Journal of Economics 113: 693–732. Gottschalk, P. 1997. Inequality, income growth, and mobility: The basic facts. Journal of Economic Perspectives 11(2): 21–40. Greenan, N., and D. Guellec. 1998. Firm organization, technology and performance: An empirical study. Economic Innovation, New Technology 6: 313–47. Greenan, N., J. Mairesse, and A. Topiol-Bensaid. 1998. Computerization and R&D; links connecting skills, average salaries and productivity: A study of French firm level data. International conference on. Information and Communications Technology, Nice, June. Griliches, Z. 1969. Capital-skill complementarily. Review of Economics and Statistics 51: 465–68. Griliches, Z. 1979. Issues in assessing the contribution of research and development to productivity growth. Bell Journal of Economics 10 (Spring): 92–116. Hoerr, J. P. 1988. And the Wolf Finally Came: The Decline of the American Steel Industry. Pittsburgh, PA: University of Pittsburgh Press. Hogan, W. T. 1991. Global Steel in the 1990s: Growth or Decline. Lexington, MA: Lexington Books. Holzer, H. 1996. What Employers Want: Job Prospects for Less-Educated Workers. New York: Russell Sage Foundation. Holzer, H. J. 1998. Employer skill needs and labor market outcomes by race and gender. Industrial and Labor Relations Review 52(1): 82–98. Howell, D., and E. Wolff. 1992. Technical change and the demand for skills by U.S. industries. Cambridge Journal of Economics 16: 127–46. Howell, T. R., W. A. Noellert, J. G. Kreier, and A. W. Wolff. 1988. Steel and the State: Government Intervention and Steel’s Structural Crisis. Boulder, CO: Westview Press. Hunter, L. W., and J. J. Lafkas. Forthcoming. Opening the box: Information technology, work practices, and wages. Industrial and Labor Relations Review. Huselid, M. A. 1995. The impact of human resource management practices on turnover, productivity, and corporate financial performance. Academy of Management Journal 38: 635–72. Huselid, M. A., and B. E. Becker. 1996. High performance work systems and firm performance: Cross-sectional versus panel results. Industrial Relations 35: 400–22. Hwang, M., and D. Weil. 1996. Production complementaries and the difference of modern manufacturing practices: Evidence from the U.S. apparel industry. Working paper. Ichniowski, C., T. A. Kochan, D. Levine, C. Olson, and G. Strauss. 1997. What works at work: Overview and assessment. Industrial Relations 35 (July): 299–333.

266

Kathryn Shaw

Ichniowski, C., and K. Shaw. 1995. Old dogs and new tricks: Determinants of the adoption of productivity-enhancing work practices. Brookings Papers: Microeconomics: 1–65. Ichniowski, C., and K. Shaw. 2000. TQM practices and innovative HRM practices: New evidence on adoption and effectiveness. In R. Cole and R. Scott, eds., The Quality Movement in America: Lessons from Theory and Research. Thousand Oaks, CA: Sage Publications, pp. 347–66. Ichniowski, C., K. Shaw, and G. Prennushi. 1997. The effects of human resource management practices on productivity. American Economic Review 86 (June): 291–313. Jensen, J. B., and K. R. Troske. 1997. Increasing wage dispersion in U.S. manufacturing: Plant-level evidence on the roles of trade and technology. Juhn, C., K. M. Murphy, and B. Pierce. 1993. Wage inequality and the rise in returns to skill. Journal of Political Economy 101(3): 410–42. Kaiser, U. 1998. Shifting employment patterns in business related services. Center for European Economic Research. Working paper. Katz, L., and D. Autor. 1999. Changes in the wage structure and earnings inequality. In O. Ashenfelter and D. Card, eds., Handbook of Labor Economics. Amsterdam: North Holland. Katz, L., and K. Murphy. 1992. Changes in relative wages, 1963–1987: Supply and demand factors. Quarterly Journal of Economics 108: 33–60. Kelley, R. 1998. How to Be a Star Performer. New York: Free Press. Kletzer, L. G. Forthcoming. Trade and job loss in U.S. manufacturing, 1979–94. In R. C. Feenstra, ed., The Impact of International Trade on Wages. Chicago: University of Chicago Press. Krueger, A. 1993. How computers have changed the wage structure: Evidence from micro data, 1984–1989. Quarterly Journal of Economics 109 (February): 33–60. Lawler, E. E., S. A. Mohrman, and G. E. Ledford Jr. 1995. Creating High Performance Organizations: Practices and Results of Employee Involvement and Total Quality Management in Fortune 1000 Companies. San Francisco: Jossey-Bass. Levine, D. 1995. Reinventing the Workplace: How Business and Employees Can Both Win. Washington, DC: Brookings Institution. Levy, F., A. Beamish, R. Murnane, and D. Autor. 1999. Computerization and skills: Examples from a car dealership. Working paper. November. Licht, G., and D. Moch. 1997. Innovation and information technology in services. Center for European Economic Research. Working paper. MacDuffie, J. P. 1995. Human resource bundles and manufacturing performance: Organizational logic and flexible production systems in the world auto industry. Industrial and Labor Relations Review 48: 197–221. Machin, S., and J. Van Reenen. 1998. Technology and changes in the skill structure: Evidence from seven OELD countries. Quarterly Journal of Economics 113: 1215–44. Maurin, E., and D. Thesmar. Forthcoming. Changes in the functional structure of firms and the demand for skills. Journal of Labor Economics.


267

Milgrom, P., and J. Roberts. 1992. Economics, Organization, and Management. Englewood Cliffs, NJ: Prentice-Hall. Neumark, D., D. Polsky, and D. Hansen. 1999. Has job stability declined yet? New evidence for the 1990’s. Journal of Labor Economics 17 (October): S29–S65. Nickell, S., D. Nicolitsas, and M. Patterson. 1995. Does doing badly encourage management innovation? Prepared for the R&D Innovation and Productivity conference at IFS. Osterman, P. 1994. How common is workplace transformation and who adopts it? Industrial and Labor Relations Review 47(2): 173–87. Osterman, P. 2000. Work reorganization in an era of restructuring. Industrial and Labor Relations Review 53(2): 179–96. Ritt, A. 1995. Major minimills broaden their reach. NewSteel 11 (September): 30–33. Sullivan, D. H., and T. M. Smeeding. 1997. Industrial change and the fortunes of workers: An exploratory analysis. Luxembourg Income Study. Wolff, E. N. 1996. The growth of information workers in the U.S. economy, 1950–1990: The role of technology change, computerization, and structural change. Working paper.


Comments on Chapter 6 Nathalie Greenan

Kathryn Shaw’s paper describes the mechanisms by which choices in the fields of ICT investments, organizational changes, and labor demand are closely connected to one another at the plant or firm level. This analysis is backed up by an impressive empirical research that she has conducted with some co-authors on the steel industry. This case study has generated very rich material including facts collected from field work and statistical data on 84 steel mills owned by 40 companies in the United States, Japan, and France during the 1988 to 1997 period. It embeds at the same time some precise information on a comparable segment of the steel process in integrated mills and minimills and a knowledge on the sector itself. Selecting such a field for building up plant-level measures is equivalent to holding constant many variables at the same time. Shaw then broadens the perspective to the manufacturing sector, referring to empirical results found in the economic literature, proposes alternative explanations and a theoretical framework where the formation of a ‘‘knowledge capital’’ plays a central part. This discussion is focused on the worker and on the firm level. It will try to highlight the conclusions given by Shaw with some results stemming from French national labor force and business surveys on the manufacturing sector. These data are different from field work data because they go in much less detail and concern representatives samples of firms. But it is complementary and useful to separate common beliefs nurtured by the managerial discourse and their outcomes in a firm’s everyday life, and large sample statistics are a great help in doing it. A first section stresses and develops some key points in the chapter about the nature of IT impacts on organization and skills. A second section outlines some further research issues

270

Nathalie Greenan

that could be fundamental in order to improve our understanding of mechanisms involved in computerization and to anticipate on future evolutions. These remarks have to do with the way economists conceive organizational change and with measurement issues. ICT, Organization, and Skills ICT Impacts on Work Content and Organization Shaw stresses the fact that computerization and new organizational practices can act as complements in the production process. Computers have changed the job content of steel workers, although the basic steel-coating process is unchanged. Workers do not stand along the line any more but are located in a computerized air-conditioned room above the line where they deal with more information of a formalized type than before, like diagrams or numbers. On the other hand, the steel industry also uses innovative human resource management (HRM) practices that aim at increasing employee involvement through devices such as problem-solving teams, training, job rotation, information sharing, screening of employees, and incentive pay. These practices increase decision-making power for production workers, and they ‘‘fit’’ well with computers that allow for more systematic and synthetic information gathering that helps decision making. Thus computerization and new HRM practices jointly raise productivity more than they would individually. This story of the steel industry seems convincing, although no extensive data on ICT use have been collected. Moreover it is in line with results obtained in other empirical studies (see Brynjolfsson and Hitt, chapter 2, this volume). However, it does not imply that there is a univocal and deterministic relation between computerization and changes in work content and organization. One reason is that causality is difficult to establish. Computerization and adoption of new HRM practices are two technological shocks that have been occurring simultaneously in new mills while old mills remained traditional in their way of doing things. More important, such an interpretation neglects the diversity in information technologies and in organizational changes implemented by firms. As far as ICT is concerned, a distinction could be made between the use of ICT as a production tool and the use of ICT as a knowledge tool. As a production tool, ICT automates the pro-


271

duction process, while as an information tool it changes the way information is processed and gathered throughout the firm. On the other hand, innovative HRM practices group together devices that have to do with higher involvement of direct producers in information processing, but they leave aside other trends in the evolution of modern organizations like growing formalization, increased control, higher work rhythm, downsizing, or reengineering. I will discuss these trends more thoroughly in the second section and just evoke here two example. The first one is a result obtained on a large sample of French manufacturing firms (Gollac, Greenan, and Hamon-Cholet 2000). It seems that in the first half of the 1990s, the investment in computers followed a trend of increased formalization within firms, partly connected with the adoption of management tools such as ISO certification or other quality norms. More recently firms that develop just-in-time practices, outsource some services or subcontact part of their production, seem to benefit from the new connection opportunities opened by network technologies. My second example is an anecdote, gathered from a firm that I visited with Jacques Mairesse in conducting our research on the economic impact of organizational change in French manufacturing (Greenan and Mairesse 1999). This firm is the French subsidiary of an American firm, and we interviewed the plant manager. According to him, the plant had implemented all the modern organizational devices to yield higher involvement of workers (HRM practices), higher quality (quality certification, TQM), and tighter deadlines ( just in time). More precisely, the organization had been modeled according to a ‘‘modern manufacturing’’ program established by scholars from two different American universities at the beginning of the 1990s. This firm was also a heavy user of ‘‘knowledge’’ type of ICT, including mainframes, PC networks, and Internet. The interview was focused on performance issues and what appeared very clearly was the fact that ICT, as well as new organizational devices were very useful to implement both financial and nonfinancial measures of performance. In fact our general impression was that the plant manager had only a very narrow decision-making authority. He spent most of his time calculating performance indicators and sending them in real time to the American headquarters. Decisions in the fields of ICT investments, industrial equipment, product changes, and organizational design were imposed from outside. His

272

Nathalie Greenan

part as a manager was to favor the building up of a learning curve around these imported tools. Of course, this evidence is anecdotal, but it suggests that ICT, as well as new organizational devices, can be used in a vertical or hierarchical perspective. Increased formalization embedded in some of the new practices like quality certification or inventory management systems ‘‘fits’’ well with ICT and allows an easier ‘‘distant’’ control, even in flatter hierarchies. ICT, Organizational Changes, and Skills Another important result stemming from the steel case study is that ICT and organizational shocks have not clearly increased the set of educational skills required in new hires. Although mathematical and computer skills are part of the important criteria, the central change is that firms are looking for different persons: positive attitude toward hard work and reward, and ability to communicate, participate in teams, and think for oneself. We may add here that although these skills are more ‘‘personal’’ than ‘‘educational,’’ they are connected with the ‘‘social’’ and ‘‘cultural’’ capital of the worker that favors success in the scholar system. Thus this type of selection relies more on the social structures than the selection based on levels of education (Gollac and Kramarz 2000). The skill issue should also be thought of in relation with the diversity of ICT and organizational changes. As a production tool, ICT may favor substitution of capital to labor and increase the share of unskilled blue- and white-collar workers among direct producers. As a knowledge tool ICT has much more intricate impacts, both direct and indirect, on employment. In manufacturing, two French empirical studies, using different statistical surveys, one by Goux and Maurin (1997) and the other one by Greenan (1996a), have showed that robots and numerically controlled machine tools were positively correlated with unskilled labor, whereas computer-aided manufacturing systems were correlated with skilled labor. In sum, I would say that ICT is a tool that can be associated with many different kinds of organization and competencies. More precisely, ICT impacts on work content, skills, and employment largely depend on firms’ decisions in the field of organization. In the steel industry, organizational changes are oriented toward greater autonomy for direct producers that favors skill upgrading because direct


273

producers are asked to ‘‘manage’’ operational decisions by themselves. In our French example the plant manager is an engineer who specialized in quality control. It is very likely that his skills in building up performance indicators has been important in his promotion. But we may guess that they would not have been sufficient if the plant had more autonomy in choosing its technology. If we revisit the context of automation in the 1970s, we come to another story on this nexus of relations. We may say that technology allowing automation had a lot to do with ICT (we can describe them nowadays as ‘‘old ICT’’), and they brought about a debate on deskilling versus skill upgrading technical change that has some common features with today’s debate on technological bias. As a matter of fact the 1970s and today have in common the fact that they are periods of learning by using: organizations are finding out what to do with ICT through experimentation. And experimentation requires cognitive, communication, negotiation, and interpersonal skills. If this story is true, we could think that once uses are stabilized (which is an issue), firms could move back to lower skill contents. But saying so, we would miss a central feature that makes Western economies in the 1970s different than in the 1990s: the share of workers in the labor force that is able to read, write, abstract, calculate, and speak languages has been growing, and this structural move toward more education has an influence both on organizational choices within firms and on the type of goods that consumers prefer. Causality is difficult to establish as far as the relations of ICT, organization, and skills are concerned. Nevertheless, the heterogeneity in firm-level databases can give some insights in the causes. In French manufacturing, using a survey on organizational change in production conducted by the Ministry of Industry, I found a correlation between the use of advanced manufacturing technologies and organizational change (Greenan 1996b). But when these two variables were put together in a regression to explain long differences in the occupational structure of firms, only the organization variables showed a significant correlation (Greenan 1996a). Of course, econometric methods have to be improved in order to deal more robustly with the complementary between technological and organizational change, but some progress has been made (Athey and Stern 1998). The careful design of statistical surveys is also very important, and it is backed up by the knowledge accumulated through field work.

274

Nathalie Greenan

Deeper into the Facts We are at a point in research where a set of innovative organizational devices have been identified as forming a set of ‘‘new best management practice.’’ These devices are most of the time described as leading to higher decentralization in operative decisions. The system they make are presented as a response to a trade-off between autonomy and control. But we need more theory on organizational change in order to improve the way we look at the data. As we look closer at these devices, we see that different dimensions are present, and it is not clear whether they really fit well with one another in terms of efficiency. Employee involvement practices are most often referred to as leading, for the shopfloor worker, to more knowledge, more autonomy and more cooperative work. Total quality management leads to more procedures and formalization of work. Delayering, lean production, and just in time generate stronger interdependence among workers and a higher time constraint. In France, organizational change also takes another direction concerning job status: new jobs are more unstable and precarious than old ones, and they are given to new entrants in the labor market, that is, to young workers, even if they are more educated than their colleagues with an insider position. In sum, the evidence is more mixed than what is suggested by a simple trade-off between autonomy and control: How are higher worker participation, demand for more cooperation and team skills, lower job security, and higher time constraints arranged together? The idea of a best practice does not help in understanding the complex trades-off involved and the tensions associated with it. These tensions do not seem to be specific to the French context, even though they may vary according to institutional backgrounds (Cappelli et al. 1997). My last point is about measurement issues in the field of organizational change. Organization is much harder to measure than R&D or ICT investments: no economic value of it can be easily computed, and it is not localized in a special place within the firm, but pervasive. In France, surveys on organizational change are, since the beginning of the 1990s, conducted in two different fashions: through labor force surveys and through business surveys. Business surveys have measured the fact that the internal organization of firms was deeply on the move, bringing about changes in


275

the vertical and horizontal division of labor. In the words of the interviewed firm representatives, they are associated with positive values: autonomy, flexibility, work enrichment, and fitness. On the other hand, labor force surveys, conducted at the workers’ home, lead to different types of results. Between 1984 and 1991 French workers have been reporting their feeling that working conditions were becoming worse: they carry heavier loads, work in environments that are more stressful, have to cope with higher work rhythm, and even increased direct supervision (Gollac and Volkoff 1996). At the same time they have gained autonomy in the sense that they more often fix themselves the various glitches that occur in the course of the production process, and they communicate with an increased network within and outside the boundaries of the firm. Moreover ICT seems to render some of these different work characteristics more compatible (Ce´zard, Dussert, and Gollac, 1992) directly, by shaping the job characteristics of ICT users, or indirectly, through systems that influences the job characteristics of non ICT users (e.g., truck drivers). In France these trends in job profiles have been interpreted as an intensification of work (Gollac and Volkoff 1996). What is surprising is that they do not seem to be associated with higher measured productivity at the firm level, although they are conducive to higher work effort (Greenan and Guellec 1998). Another interesting result can be evoked, coming from a French survey that is close to the British Survey on workplace and industrial relations and called the Reponse survey. In this survey, conducted at the establishment level in 1993, both firm representatives and trade union representatives were interviewed. When asked about the presence in the establishment of employee involvement devices such as quality circles, expression groups, or meetings on a regular basis, about one-third of the firm representatives disagreed with the trade union representative (Coutrot 1996). These different examples show that the measure of organizational change is a matter of point of view. It is likely that broad differences in perceptions are a sign of tensions within the firm and of inconsistencies in organizational choices. In France, a survey,1 called COI (Changements Organisationnels et Informatisation), was conducted in 1997 whereby firm representatives as well as two or three employees per firm were interviewed. Some questions were posed both to employers and to employees (e.g., on autonomy), and others to employers only (strategy) or to employees only (tensions between

276

Nathalie Greenan

assigned objectives). Information about the firm given by employers can enrich employee-level estimations (e.g., wage equations) and answers given by randomly drawn employees within the firm can be used in firm-level econometric studies (Greenan and Mairesse 2000; Mairesse and Greenan 1999). This kind of survey allows one to sample a large number of firms on how the adoption of new management tools changes the way people work. Some early results support the ‘‘knowledge capital’’ argument of Shaw (Greenan and Mairesse 1999). In reorganized firms, blue-collar workers tend to contribute to a larger communication network; they make more frequently suggestions for process improvements and they attend to more meetings. On the other hand, reorganization is also positively correlated with higher ‘‘industrial’’ constraints like the rhythm of work fixed by the machine’s pace and quantitative and qualitative norms or permanent production work within a group. Moreover technology seems to separate blue-collar workers with high communications jobs from blue-collar workers with industrial jobs: using a computer increases the probability of high-level communication work, whereas using an automated machine increases the probability of industrial-type work. Note 1. In 1993 the French Ministry of Industry conducted a survey on organizational change in production. The coi survey was conducted in 1997. The Ministry of Industry, the Ministry of Agriculture, the Ministry of Labor, and the National Institute of Statistics and Economic Studies (INSEE) participated. I provided the scientific guidance for the design of three questionnaires and for the sampling method, and coordinated the work of the statistical agencies involved, first at INSEE and then at the Centre d’Etudes de l’Emploi.

References Athey, S., and S. Stern. 1998. An empirical framework for testing theories about complementarity in organizational design. NBER Working Paper 6600. Cappelli, P., L. Bassi, H. Katz, D. Knoke, P. Osterman, and M. Useem. 1997. Change at Work. New York: Oxford University Press. Coutrot, T. 1996. Relations sociales en entreprise: voir midi a` sa porte. Travail et Emploi 66: 71–86. Ce´zard, M., F. Dussert, and M. Gollac. 1992. Taylor va au marche´: Organisation du travail et informatique. Travail et Emploi 54: 4–19.


277

Gollac, M., and S. Volkoff. 1996. Citius, altius, fortius: L’Intensification du travail. Actes de la Recherche en Sciences Sociales 114: 54–67. Gollac, M., and F. Kramarz. 2000. L’Informatisation comme pratique et comme croyance. Actes de la Recherche en Sciences Sociales 134: 4–21. Gollac, M., N. Greenan, and S. Hamon-Cholet. 2000. L’Informatisation de l’ancienne ećonomie: De Nouvelles machines, de nouvelles organisations et de nouveaux travailleurs. Economie et Statistique 339–340: 171–201. Goux, D., and E. Maurin. 1997. Le dećlin de la demande de travail non qualifie´: Une me´thode d’analyse empirique et son application au cas de la France. Revue Ećonomique 48(5): 1091–1114. Greenan, N. 1996a. Progre`s technologique et changements organisationnels: Leur impact sur l’emploi et les qualifications. Economie et Statistique 298: 35–44. Greenan, N. 1996b. Innovation technologique, changements organisationnels et e´volution des compe´tences. Economie et Statistique 298: 15–33. Greenan, N., and D. Guellec. 1998. Firm organization, technology and performance: An empirical study. Economic of Information and New Technologies 6: 313–47. Greenan, N., and J. Mairesse. 1999. Organizational change in French manufacturing: What do we learn from firm representatives and from their employees. NBER Working Paper 7285. Greenan, N., and J. Mairesse. 2000. Computers and productivity in France: Some evidence. Economics of Information and New Technology 9: 275–315. Mairesse, J., and N. Greenan. 1999. Using employee-level data in a firm-level econometric study? In J. C. Haltiwanger, J. I. Lane, J. J. M. Spletzer, and K. R. Troske, eds., The Creation and Analysis of Employer-Employee Matched Data. Amsterdam: North Holland, pp. 489–512.


Conclusion: Computer, Productivity, and Wages; Reflections on the Economics of the Information Age Timothy F. Bresnahan

The arrival of a new general-purpose technology, with its prospects for widespread changes in product and factor markets, is an occasion to reflect on what we know about the economic impact of technology. The analysis in this volume comes at the time when the commercial internet is at the earliest stages of its role as a GPT, but when computing more generally is at the fifty-year mark. This is a particularly opportune time to confront the empirical puzzles associated with the economic impact of computers so far, and to turn our understanding to the future. In this closing chapter I seek primarily to make that bridge in time.1 This tasks appears daunting because of the puzzles about the past brought to the foreground in this volume. Are computers (and information technology more broadly) a highly valuable general-purpose technology spreading spillout benefits throughout modern economies? Does computerization of the workplace make up the skill-biased technical change we need to explain recent labor market events? I argue that detailed and specific answers to these questions form the only sound basis for using our understanding of the recent past to understand the near future. How might the commercial Internet affect the growth rate of economic well-being, and how might it affect the demand for labor? I will attempt to undertake the leap from past to future with regard to both the productivity and the labor market issues. I focus here on the micro–macro puzzles in the puzzling relationship between computers and the economy. There is an important distinction, revealed in the studies in this volume and elsewhere, between the output and productivity results, which show an important micro–macro puzzle, and the labor market results, which do not. For both productivity and labor, we see a strong relationship in the micro data. Studies of productivity see sharp difference be-

280

Timothy F. Bresnahan

tween firms that use more information technology capital and firms that use less. The more computer-using firms are likely to be more successful on a variety of measures, including productivity. Yet this micro result does not appear as a time series macro results, for measured productivity has not grown rapidly in the era of computerization, with the exception of the late 1990s.2 In contrast, the macro and micro evidence on the labor market appear more consistent. Computer-using firms use relatively more skilled workers, and the era of computerization of work has seen a surprising shift upward in the demand for highly skilled workers relative to those with fewer skills. For the micro productivity results to be right, there need to be some benefits of information technology that can be measured at the firm level but not in the aggregate. Obvious candidates include difficult to measure product quality improvements from computerization, which tend to raise a firm’s demand at the expense of competitors (thereby becoming measurable in micro data) but which elude the quality-adjustment efforts of productivity and output measurement frameworks in the macro data. For the macro productivity results to be right, there must be a measurement problem at the micro level. Obvious candidates include reverse causation based on micro-level heterogeneity. Some firms may be more successful than others for reasons that have nothing to do with computerization. The more productive firms may choose to invest in information technology for whatever reason (perhaps computers are fun for managers, and successful firms have managerial slack?). The labor market results do not immediately bring forward a macro–micro puzzle to reconcile, for there is an effect both at the firm level and the economywide level. They do, however, force us to confront the contradiction between the macro-level results on productivity and those on labor. How did computers manage to shift economywide labor markets without shifting output markets? Further, there is heterogeneity at the firm level in both the output and the labor market results. Why does it wash out at the aggregate level for the productivity measures and not for the labor results? What is the nature of the heterogeneity across firms that affects their labor demand and their productivity; is it a single cause or a coincidence? Reconciling these puzzles of the past involves understanding the forces pulling computing into highly productive uses as well as the limitations on its productivity. Especially important is understand-

Conclusion

281

ing the sources of heterogeneity at the firm level. I argue that they are consequences of heterogeneous adoption costs for a new and valuable technology. First, we should understand the first fifty years of computing as the reasonably rapid diffusion of a series of increasingly better and cheaper technologies. Second, whether we think of these technologies as skill-biased or as productivity enhancing (or both), we should examine the forces that have limited the extent of their diffusion. I argue that the limitation has come at the boundary of organizational computing (as distinct from technical computation or personal computing). As a result I argue that the real promise that comes from the commercialization of the Internet arises from new opportunities to extend organizational computing into new spheres: interorganizational computing. But getting to that result about the future calls for a review of our knowledge of the past. 7.1

Invention in Organizational Computing

I base my explanation of the productivity results and of the labor demand results in the same specific theory of the effects of IT. I focus on organizational computing (OC) and specifically co-invention by firms using IT in business information systems. Business information systems are productive and important applications of organizational computing, which is a new (fifty years old) technology for whitecollar bureaucracies. In sectors like finance, banking, brokerage, and insurance, OC is the technology of production. In much of the rest of the service sector, OC is a crucial technology for control of production processes or is the production process for key subtasks. In the rest of the economy, OC serves as a control technology for many production processes and is the production technology for the buying of inputs and the selling of intermediate outputs. Transacting, remembering, allocating, budgeting, planning, rewarding performance, and managing are all heavily influenced, or accomplished, by organizational computing. Although OC is about fifty years old, the underlying powerful computer, data communications, and software technologies have undergone constant and rapid technical change.3 Adoption of OC began in those firms, and those functions within firms, where there are large-scale white-collar processes. These were in such industries as financial services, and in such functions as accounting in other industries. OC has since diffused to many more kinds of firms and functions.

282


A firm that desires a new business information system and that decides that IT has advanced enough for that system will nonetheless bear large co-invention costs to design and implement that system. By co-invention I mean invention closely complementary to the use of IT that takes place in the IT-using firm itself. For business information systems, co-invention often takes the form of linked improvements in the production process and in difficult to measure elements of output, such as response time, availability, or reliability in provision of a service. Co-invention involves two things that are not conceptually trivial. First, it involves the translation of an underlying technical opportunity—an advance in computer or networking hardware, or in fundamental software such as database management systems— into business systems that produce the same output using fewer resources (sometimes) or which produce a better quality output (often). The translation is itself not trivial, nor is the conceptual leap of understanding how a firms’ customers will value the ‘‘soft’’ output attributes that OC is particularly suited to, such as timeliness, convenience, flexibility, and availability of the output. Second, changing the organization using the business information system often involves re-organization in several senses. Job definitions, division or even firm boundaries, incentive schemes, and other elements of the organization’s structure and functioning will be different in the computerized business information system than in the preexisting one. Effective use of IT in organizational computing calls for substantial co-invention by the using firm, and this co-invention takes time and resources, including flashes of inventive brilliance that become real in changing work systems. Inventive brilliance that changes everyone’s job is not usually the cheapest resource in organizations. Deep understanding of customer valuation of radical changes in ‘‘soft’’ output attributes is often scarce as well. There are at least two important implications of expensive and uncertain (calling for brilliance) co-invention at the firm level. The first implication is that the rate of technical progress in IT-based production processes like OC is slower than the rate of technical progress in IT itself. Economywide advances in computer, software, or telecommunications technology are just one step in a multistep process of invention. To become valuable in use, they need coinvention. Further IT-using firms are very aware of co-invention

Conclusion

283

costs, which are one of the central points of discussion in the professional and business literature of IT users. Because co-invention costs, and the success of attempts at co-invention, vary across firms, there is substantial heterogeneity in the use of IT in the cross section of firms at any given moment. The heterogeneity is driven, in no small degree, by differential success in co-invention and differential ability to co-invent. Co-invention is linked both to productivity (via unmeasured output) and to skill-biased technical change (via organizational change). 7.2

Output Measurement and Productivity

Let me illustrate what I mean about the difficult to measure elements of business information systems’ output contribution with a simple and familiar example, the automatic teller machine. This banking technology is about half as old as computers, just over a quarter of a century, and uses both computer and data communications technology intensively. Its value to the consumer arises largely through flexibility and availability or convenience. Withdrawing money at night, or in a grocery store, and with a quick check on account balances, offer considerable product quality advantages over traditional banker’s hours and locations.4 Banks do not find these service quality improvements easy to value, though they obviously think they are large (partially based on internal studies of higher customer attraction and retention rates). By their technology investments, banks reveal their assessment that the service quality improvements will be beneficial to customers. They hope that these customer benefits will attract customers and contribute, when the returns are not competed away, to profitability. A bank that successfully innovates in a new customer service system— as ATMs were in the 1970s, and improvements to them were later on—would expect to see its market share rise and, typically, its profitability rise as a result. None of this return is captured in the productivity measurement statistics. It would, however, show up in micro-data studies that compare across firms. Some elements of the banking anecdote are systematic. Many business information systems have customer convenience, flexibility, and availability as their primary value. Measuring the private return to those values is difficult—the relevant engineering professional literature talks of the ‘‘problem of metrics.’’ Whereas cost savings would

284


be measurable, at least ex post, the value of service quality improvements is difficult to measure. The anecdotes are not at all limited to the difficult-to-measure sectors of the economies like financial services. Manufacturing firms, for example, compete on delivery terms, availability, and so on.5 Just as is the case with the improvements resulting from the ATM in banking, these returns are not captured by productivity statistics. Further the nature of the returns suggests a substantial spillout to society at large. Customers get much of the return to OC improvements in product quality, particularly after lagging IT-using firms catch up to the leaders in their industry and undercut the leaders’ ability to obtain a price premium for their superior quality. These sources of value from OC in use and the costs interact with the difficulty of co-invention. Because of the latter, there is considerable variety across firms in the degree to which they adopt the latest OC technologies. This variation shows up particularly between firms that are leaders and followers in economic success. After a period of time, however, the adoption process allows the followers to take advantage of the technical advances that leaders had adopted earlier. These could be ordinary mechanisms of technology diffusion, as decreases in the price–performance ratios of IT can make certain technology choices attractive to ever more firms over time. Coinvention that is idiosyncratic to a firm may even spill out to other firms over time. Some mechanisms are immediate and direct, such as in the consulting and custom programming sector which enables OC knowledge to be spread from one client firm to another. Further technical progress may lower the costs of co-invention over time, as in the invention of a database management system that dramatically simplified the creation of many OC applications. Other classes of OC application become embedded in software products, as in SAP’s enterprice resource management products. A firm that acquires such software can avoid co-invention costs to the extent earlier coinvention is embedded in the software. All of these forces mean that there are substantial increasing returns to OC at national and international levels, as the costs of invention and co-invention are both, with a lag, spread out over many firms. There are important limitations on the valuation of IT in use. The first I have already discussed at some length. This is the costs and delays associated with co-invention of business information systems. Those costs and delays result in the benefits of improvements in the

Conclusion

285

underlying hardware, software, and communications technologies lagging several years behind the initial invention, even as much as a half a decade.6 OC has improved in a series of small steps, each setting off a wave of co-invention that diffuses throughout the economy over a matter of years (not decades). The scope of application of OC has also been limited by the fixed costs of designing and implementing business information systems. Within the firm, this means that OC’s benefits are far higher for processes which are repetitive and used by many workers, and lower for processes which are infrequent or limited to individuals. That limitation comes from the large fixed costs associated with OC technologies, such as mainframe computers, themselves, and the fixed costs of co-invention, such as training individual workers to use OC applications. A related limitation arises at the boundaries of the firm. Many valuable applications of OC cross firm boundaries. Traditional organizational computing technologies are better at mediating the transactions between trading partners who deal in standard transactions at a very large volume, such as the automated parts of bourses and the reservations systems that are usable by travel agents (regular trading partners of airlines). They have been less effective for transitory commercial relationships because of the fixed costs of design and training. The personal computer was greeted as the end of these limitations. This turned out to be partially correct. PCs are strong were OC technologies are weak. PCs are smaller, cheaper and easier to use, than mainframes. This permitted use of computing power in tasks such as document production, simple accounting using spreadsheets, and desktop publishing, where the preexisting OC technologies were effective. These ‘‘personal productivity applications’’ are largely separate from the organizational productivity applications built around OC. From a macro perspective, the stand-alone PC is a cheaper form of computing that, while valuable, has diffused to lower-value uses than the earlier OC technologies. From the early days of the PC, those designing business information systems wished for the opportunity to draw on the strongest features of both OC and PC technologies. This ‘‘best of both worlds’’ vision was announced for a number of technical advances, for it was responsive to a very high value demand opportunity, the limitations on OC. Business information systems within the firm or crossing firm boundaries that connected to the central OC applications but

286


that could be used intermittently, causally, or with little training would be highly valuable. For many years, however, the new technologies that were invented to support these kinds of applications failed to deliver on ‘‘best of both worlds.’’ It is the commercial Internet that finally will deliver on that process, a topic to which we will return shortly. In arguing here that there has been a substantial social gain to IT in use, I do not mean to make any argument about the productivity slowdown whatsoever. There is the rest of aggregate technical progress to deal with, and the rest of the macroeconomy’s growth to understand. I argue merely that the social gain to computerization has been very large, that it has covered most sectors of the economy,7 and that it is badly captured by the aggregate productivity statistics. These remarks are only distantly related to the behavior of productivity aggregates over time. In sum, organizational computing has been an important technology in producing gains for consumers. It has also, however, had substantial adjustment costs at the firm level and as a result diffused slowly and unevenly. Important technical limitations have given OC a range of applicability that, while substantial enough to generate a social return to invention at the economywide level, leaves obvious boundaries beyond which applications have not yet penetrated. 7.3

Labor Markets

Over the last several decades, demand for highly skilled labor has grown substantially more than demand for less skilled labor. When not offset by countervailing movements in labor supply, this labor demand shift has been the proximate cause of a widening spread in income distribution. The literature on economywide labor markets identifies skill-biased technical change as the source of large changes in labor demand. OC is a technology that effects enough of labor demand to form an important part of the macro labor market explanation. Much of modern employment is in the white-collar bureacracies where OC is an economywide and substantial piece of technical change. The key to understanding OC as skill-biased technical change lies in the co-invention of business information systems. OC does not merely involve the use of computer and telecommunications technology in the firm. Instead, the firm makes substantial changes in the or-

Conclusion

287

ganization of its workforce. The specifics of these changes vary from firm to firm and industry to industry, of course, but there are strong systematic tendencies toward complementarity and substitutibility relationships with different types of labor. This section examines those relationships with an eye to the question of whether OC is old enough, widespread enough, and sufficiently linked to labor demand to be an important part of the skill-biased technical change seen in the macro labor market. Limited Substitution The first mechanism by which organizational computing has been skill-biased technical change is that of limited substitution. Over the first fifty years of computer use, routine bureaucratic work has gradually become systematized and partially automated through IT-based organizational computing. As a result many of the clerical tasks once performed by low- and medium-skilled workers have been automated. The relevant tasks here are not so much handling of paper, which later becomes handling of data, but the replacement of a great deal of remembering, recording, and rule-based decision making. Is it time for a letter to a customer, reminding him his bill is overdue? That task was once the province of accounts receivable (AR) clerks, now of AR software. There are many such routine tasks in whitecollar bureaucracies. They can be at least partially automated because the essential elements of the tasks can be carried out by people or by business information systems.8 The notion of ‘‘partial automation’’ is important here. It is rare that the computer completely replaces human workers in performing such medium-skill tasks. Instead, OC changes the whole business information system so that work tends to be shared between people and computers in a variety of complex ways. The computer system is quite good at remembering a large number of small things, like how many times customer X has been late with his bill. Continuing to use the AR example to stand for much of OC, I note also that the computer is good at making the same routine decision over and over again, like ‘‘payment is late 30 days, send bill.’’ At a task level then, computers are good substitutes for human workers at repetitive, routine, but care-intensive tasks—the center of work for mediumskilled and low-skilled people in white-collar bureaucracies.

288


As technical progress in OC has advanced over the last half century, the number of such automated tasks has slowly but steadily increased. One driver is that OC has gotten cheaper, so firms find it even more worthwhile to automate tasks. This is partly due to exogenous technical progress in electronics, both computers and telecommunications, as I suggested above, and partly to network effects among firms using OC and scale economies in the invention of OC technologies. Another cumulative piece of technical progress subject to some network effects arises in connection with the automation of work. As experience with OC has grown, computers have gotten better at making more complex decisions because they have been programmed to make more complex decisions as the number and complexity of databases has grown. Relatedly firms have learned over time how to design business information systems to provide better quality service to their customers (the accounts receivable program sends a far sweeter dunning letter to late payers the firm deems valuable customers, the inventory control system directs production toward those customers’ urgent needs, etc.). All these forces have expanded the tasks that can be undertaken by routinized, rulebased systems, permitting more substitution of computers for human cognitive skills. That substitution has historically been limited by perfectly ordinary diminishing marginal returns considerations, as parts of a large complex task are shifted one by one from human to machine function. One source of diminishing returns is at the margin of substitution with human cognitive capabilities. OC has been a more effective substitute for modest bundles of human cognitive skills than for large bundles. The routine and the repetitive have been more automatable than the carefully thought through. As a result there has been far more substitutability for the less highly educated than the more highly educated workers. A second source of diminishing returns is the rest of human skills, the noncognitive ones. Much clerical work has been changed into being the eyes, ears, voice, and fingers of the computer. While the computer can learn that the payment is late, the computer cannot call the customer on the phone to learn why, nor are computers quite as good as (some) people in dealing with an irate customer who is late paying her bill. Thus noncognitive or ‘‘people skills’’ have not been the subject of much substitution.

Conclusion

289

The ‘‘people skills’’ limitation has mattered for the impact on labor demand. As I said above, one of the main purposes of OC-based business information systems has been to improve service quality (broadly understood) and to make bureaucracies more rapidly responsive. This leads to a need for people in the OC-based business information system with good communication skills and other good ‘‘people skills.’’ Further, as complex organizations make decisions faster, the internal organization skills associated with directing and being directed become more valuable. Thus the impact of OC-based production has been to shift labor demand not just from low skill to high skill but to revalue skills of different types. Almost all of this revaluation comes from the co-invention of business information systems, not from the invention of information technology. There is a lively debate about the exact form of the organizational change in firms that is at the heart of all this co-invention from the narrow perspective of hierarchy. The routinization of some aspects of work has led both to increases in the degree to which low- and medium-skilled workers are controlled by the business information system and to increases, in slightly different contexts, in the extent to which the logic of the revised organization calls for increases in autonomy and independence. We do not need to reach a conclusion on this issue to understand that the organizational change at the firm level, understood more broadly than hierarchy, is the primary mechanism by which labor demand is shifted. Complementarities with Skill (in Large Bundles) There is another important mechanism that reinforces the view that labor demand has been shifted from the less skilled to the more skilled. Routinization and regularization and partial automation of the low-paid part of white-collar work is the direct impact of use of OC. This leads to bureaucratic production processes that are more thoroughly understood by the firms that contain them—or at least more systematically and more analytically understood. An OC-based bureaucratic production process is also more directable by the firm that employs it. There is a mechanism for changing the entire system. Further, OC-based bureaucratic production produces a great deal of information—about workers, about departments, about cus-

290


tomers and suppliers and decisions. The information is an input into more analytical decision processes about production and allocation in the firm. The widespread change in bureaucracies toward more analytically understood production processes, toward more directable production processes, and toward a flood of information have all led to an increase in the demand for analytical skills in the firm. The control of OC-based technologies calls for large bundles of cognitive skills, for people who can conceptualize and solve problems, who can understand the goals of the organization as well as its rules, and so on. This is the technological root of a complementarity between computerization and highly skilled workers. It is a source of a rising demand curve for highly educated labor. Counterintuitively, as managerial work grows more analytical, the demand for managers with high levels of human interaction skills grows as well. Interactions with subordinates that are more analytical—perhaps mediated through explicit monitoring and rulebased incentives—make it more difficult to undertake the human part of motivating, leading, counseling, and mentoring. Even more, the change of the organization to a systematized one calls for managing and being managed work that makes the junior people feel valued and motivated. Note that the complementarity arises because the white-collar production process is OC-based, not because the highly skilled worker herself uses a personal computer. In my story, the shift from lowskilled to high-skilled labor demand does not arise because of an effect that arises at the individual worker level. Instead, it arises at the firm or organizational level. The widespread diffusion of the personal computer, and the associated personal productivity applications (e.g., spreadsheets and word processing) does not play much of a role in my story of changed labor demand.9 As with the discussion of economywide productivity and personal computing above, the main point is that the stand-alone PC and the individual productivity application represent marginal use of a technology after it had grown cheap, not the extension of high-value uses to a new domain. The end result of the labor demand shifts associated with OC has two parts. First, there is a shift in the demand for human cognitive skills from the less skilled (less educated, less trained, or simply less smart) to the more skilled. Second, there is an across-the-board shift in the demand for ‘‘soft,’’ ‘‘interactive’’ or ‘‘human’’ skills. To the de-

Conclusion

291

gree that high levels of cognitive skills and of interactive skills present in the same person are particularly rare, this leads to an even larger increase in the price paid for that skill bundle. A potential important result for my analysis of labor markets is this story of failure to unbundle: it is plenty of intelligence combined, in the same person, with plenty of people skill that is scarce and highly rewarded. 7.4

The Future

Despite the early stage of commercialization of the Internet at present, it seems to me that enough can be seen of technical opportunity at this stage, and enough has been learned from the past economic impacts of computing, to make some preliminary forecasts about the size and nature of its economic impacts. Let me begin with a forecast of the direction of technical change. The commercialization of the Internet permits a combination of previously separate elements in information technology. The Internet brings together much of the power, system, and connectedness of traditional organizational computing with the small scale, ubiquity, and ease of use of traditional personal computing. As a result some of the boundaries limiting traditional organizational computing, emphasized throughout this chapter, will now break down. While it is never possible to forecast all of the implications of such an advance,10 enough can now be foreseen to say much about the future economic impacts. One should, in general, be wary of that kind of ‘‘best of both worlds’’ forecast, for it has been heard before. Because of the wellknown limits of organizational computing technology, the users of information technology have been calling out for a combination of OC and PC attributes since the early 1980s. Several not very important technologies have been announced, for marketing purposes or out of general optimism, as bridging the gap between OC and PC. At this juncture, however, we may safely forecast, for there is enough co-invention activity among the users themselves, not merely hopes on the part of technologists, to see some of the utility of the recent technological advances. We are at the beginning of a new wave of organizational computing. This time, the relevant organization is larger than the firm, including suppliers, customers, contractors, and workers. We might

292


call this new wave interorganizational computing. It has a solid foundation in recent changes in networked computing. The technical progress we have already seen in connection with the commercialization of the Internet forecasts the removal of the most important purely technical limitations on the diffusion of organizational computing in the past, difficulty of use and difficulty of low-scale (casual, intermittent, or merely associated with small trading relationships) use. Together with ongoing improvements in computing and communications power, these new technical directions will permit the construction of new IT applications that cut across organizational boundaries. The widespread distribution of the technologies that would permit this kind of valuable advance was cut short by Microsoft’s anticompetitive campaign against commercial Internet technologies. The widespread entrepreneurial invention of technologies to support ‘‘best of both world’’ applications that characterized 1994 to 1998 has been cut off. Nonetheless, Microsoft promises us that it will deliver such technologies in a form that preserves its monopoly position only about ten years later than the entrepreneurs would have. The demand opportunity remains large, so the delay, while costly, does not mean the end of the long-run opportunity for a new positive feedback loop between invention and co-invention in applications that cut across organizational boundaries. The value of automation and systemization in this area is clear. Much of distribution and communication about it has the kinds of obvious inefficiencies that call out for rationalization. Many middleman industries have market power that has accrued along with (one imagines) socially valuable functions such as learning and acting on reputations for reliability in carrying out contract terms. The new technologies permit reorganization and thus removal of the market power in an improved information environment in which achievement of the other functions is undertaken by market participants directly. Many supply relationships are longstanding because of the costs of identifying trading partners who are reliable, honorable, and so on. The resulting inflexibility could be avoided if information relevant to that identification were to arise in a lower-cost way. One could build examples like this at some length. The general point is this: much of what goes on in markets and supply chains is information processing now. Also many of the costs associated with running markets and supply chains are costs familiar to informa-

Conclusion

293

tion economics: the costs of agency, of monitoring, of reputationbuilding, and so on. Information technology, widely deployed in these new domains, has the prospect of being highly valuable. On the other hand, the very nature of that value suggests the need for a great deal of difficult, and sometimes brilliant, co-invention. Market institutions for reputation, assurance, resolution of disputes, and so on, have been built up over long years of trial and error. Replacing a complex web of human relationships and institutions with new business information systems will not be trivial. It is a very safe forecast that the benefits of automating markets, while huge, will be realized by a process that is uneven and sometimes surprising. Co-invention, whether by buyers, sellers, or new intermediaries, will take time, effort, and invention. As benefits are realized in leading examples—leading markets and supply chains in this instance—they will feed back to technological developments in IT. Those in turn will, later, enable more coinvention and adaption in other parts of the economy. There is no reason to suspect that a critical aspect of information technology use in the recent past, heterogeneity between leading and following adapters, will decline. We will continue to see a pattern in which firms vary considerably in their productivity, positively correlated with IT use. The positive correlation will continue to be a blend of rents to successful coinvention and of complementarities between IT and co-invention. The social gains to all this new invention will be substantial in the long run. Much of the dot-com boom was a bet that my last few paragraphs would be wrong.11 The events of 2000 and 2001 show us how those paragraphs are right. Many dot-com firms planned to replace much of the distribution system—intermediaries, or the buying and selling bureacracies in firms—in a radical shift to a new, IT-based form. This was a bet that co-invention was not necessary for interorganizational computing. New entrepreneurial firms would cast aside the knowledge that lived in the existing bureacracies and quickly achieve widespread success with new kinds of organizations. That bet, whether it was ex ante wise or not, has been revealed to be wrong. Instead, we will see that the widespread diffusion of interorganizational computing, just like its predecessor OC, will take time and co-invention. The short-run prospects therefore are of a period in which established firms undertake improvements to the

294


organization of buying and selling, engaging in substantial risky and difficult co-invention. That short-run forecast, however, takes nothing away from the long-run forecast of where that co-invention will ultimately lead, toward highly valuable improvements in the interorganizational computing technology supporting buying and selling. Some of the implications for labor demand seem forecastable. If we look to the production processes that will be affected by interorganizational computing, they are the ones involving buying, selling, and co-ordinating across firm boundaries. Some economists may think that this is a narrow and specific kind of enquiry I am suggesting, to look at specific production processes like those. Yet these production process involve, depending on how one does the definitions, approximately a third of the employment in the rich economies. There is plenty of scope for economywide impact. Further there remains a great deal of unautomated or only partially automated work in the white-collar bureaucracies that border the firm, buying or selling. While many of the production processes at the center of firms and industries have been very successfully automated, the ones at the boundaries of firms and beyond have, so far, changed less. Much of the work in these parts of bureaucracies is very suitable for automation using IT. Much of the white-collar work itself remains simple, suitable for partial automation. Many of the tasks these bureaucracies undertake involve remembering, making sure, coordinating, and controlling, which are exactly the tasks for which IT is quite suitable. The new technical progress opens up the geographic and organizational span of those tasks considerably. Further the prospect of flexibility with control that IT offers is even more important with the new technologies of IT that the commercial internet brings us. Transitory relationships with many suppliers, for example, are more possible after the technology of connecting has lower relationship-specific fixed costs. It seems likely that the forces of limited substitution will still be at play in the future as they were in the past, on that wider geographical and organizational span. These new applications will exhibit a familiar complementarity with analytical management. The reasons will be much the same for the interorganizational computing era as for the OC era, and will if anything grow more important. Interorganizational commercial relationships are more complex than are organizations, as they extend across

Conclusion

295

more domains of knowledge. A typical interorganizational application will involve buyers as well as sellers, and perhaps middlemen as well. This is inherently more complex and cognitively difficult. Extending the reach of organizational computing technologies across the organizational boundary, to smaller and more fleeting economic relationships than those we typically see within the firm, and to a great deal more of economic activity, seems very unlikely to change its cognitive or business-social character. Let me turn first to the business-social elements. It seems to me likely that the extension of people management across boundaries, and the resulting extended span of these new applications, will exhibit a familiar bias toward soft skills or people skills. Negotiating and managing across firm boundaries has at least as many traps and pitfalls as that within. In the present we see firms deploying employees with above-average people skills toward relationships with other firms or with workers or consumers. Further, to the extent that interorganizational computing fosters a shift toward more flexible and transitive economic relationships, it will likely push either toward labor market institutions that are based on flexibility or toward work organizations that have inflexible employment relationships but are highly redeployable. In either case, many of the soft skills and people relationship skills that have recently seen rising wages will likely grow ever more valuable. I would summarize the force of the last three paragraphs as saying the past impact of computing on the demand for labor by skill type will likely continue into the future as to direction. We can expect the relative demand for highly cognitively skilled workers to rise relative to less cognitively skilled workers, and the relative demand for workers with good social or interactive skills to rise as well. How far, how fast? It would be foolish to forecast either the tendency to pay a premium for cognitive skills or for social ones as advancing at any particular pace as the exact size and timing of the relevant technical change—co-invention—is very hard to predict.12 I take that uncertainty as a wonderful research opportunity. The introduction of IT changes the information economics of firms, markets, and supply chains. Yet information economics is notably subtle, and adding information to a system can change its institutions in a variety of directions. As interorganizational computing becomes an important process technology, we have a wonderful opportunity to

296


watch its economic impact, particularly its impact on labor markets. A related research opportunity comes from the opportunity to study the theory of organizations. Much of the successful empirical work in that area has focused on the boundaries of the firm, or on relationships between closely linked firms. There is now the opportunity to see those boundaries redrawn and those relationships redesigned on a wholesale basis—and by business people who would like to be studied, for they have no clear idea where they are going. Notes 1. I am grateful to Jacques Mairesse, Yannick L’Horty, Nathalie Greenan, and several conference participants for helpful comments, and especially to the late Zvi Griliches for discussions full, as always, of gentleness and wisdom. 2. The result does reappear in a simple cross-sectional comparison across countries, but the United States is very different from European countries on so many dimensions as to undercut the value of this observation as a tiebreaker. 3. Sometimes major branches of computing are categorized by the kind of computing used so that organizational computing was for years called mainframe computing. I wish, however, to do a categorization by type of use, and not technology. The partial replacement of mainframes by smaller computers in OC leaves the uses in place, for example. 4. The other benefit of the ATM is as a cost-saving technology, literally an automatic teller. To take advantage of the service-quality improvements, however, ATMs were deployed far earlier in time, and to far more locations, than cost savings alone would justify. 5. Comparing easy-to-measure with hard-to-measure sectors of the economy is thus ineffective as a method for learning anything about the valuation of IT in use. 6. The idea that this is a half century, and that what has been going on so far in the first fifty years of organizational computing is a low-payoff activity that will now finally blossom into a high-payoff one, is laughable. 7. Limiting the value of cross-industry research in understanding the productivity impact of IT, unfortunately. 8. A recurrent debate about whether ‘‘computers can think’’ sometimes confuses onlookers. It is irrelevant to the substitution story. Checking whether a bill is overdue and calculating the overdue penalty can be done either by a person (who is thinking) or a computer (which may or may not be thinking). 9. This analysis applies primarily to such mainstream personal productivity applications as the word processing program and the presentation-slide program. While useful, their impact is limited by the span of the individual worker’s influence. The most common uses, word processing and spreadsheet, do not tend to have dramatic impacts on the productivity of the individual. There are some specialized individual productivity applications that do have such an impact, such as those used by com-

Conclusion

297

puter programmers, graphic designers, and so on. These are a small fraction of workers, however. 10. Broader changes to society, changes in the nature of individuals’ work, changes in the autonomy/control nexis, changes in social interaction, and so on, are very difficult to forecast. 11. I am proud to report that they are printed here just as I spoke and wrote them years ago. 12. Authorial pride once again leads me to point out that sentence pre-dates the dotcom crash by several years.


Index

Adams, J., 198 Allen, S., 199 Alonso-Borrega, C., 198 Amazon.com, 83 AMTs, 68 Analytical management, 294–95 AR software, 287 AT&T, 64 Athey, S., 76 ATMs, 70–71, 83, 283–84 Audric, S., 172 Autor, D., 77–78, 181, 197, 247 Aznar, G., 108 Bailey, T., 245 Bartel, A., 199 Barua, A., 66 Baxter ASAP system, 62 Bell, B., 181 Berman, E., 197 Berndt, E. R., 77 Black, S. E., 67–68, 247 Blanchard, O. J., 173 Blanchflower, D., 114 Blow, L., 52 Boltho, A., 120 Boskin Commission, 7, 23, 26, 29–30, 53 Bound, J., 197 Bresnahan, T. F., 78, 247 British Statistical Office, 53 Brouwer, E., 114 Brown, C., 258 Brynjolfsson, E., 6, 8, 66–69, 71, 73–75, 77–78, 93, 95–98, 247 Bundesbank report, 51–52 Bureau of Economic Analysis (BEA), 8, 26, 29, 31, 35, 37, 40–41, 49

Bureau of Labor Statistics (BLS), 7–8, 26– 27, 40, 49 Caballero, R. J., 173 Campbell, B., 245, 258 Capital One Financial Corporation, 64, 81 Card, D., 161–62, 164, 166 Charleston Conference (1993), 94–95 Chennels, L., 6, 12, 209, 225–27, 258 Clark, J., 116 Clark, K., 249 Co-invention, 282–84, 289, 292 Cohen, D., 164 Cole, R., 240 Compensation theory criticisms of, 105–108, 121 employment and ICTs and, 102–108, 121 imperfection of, 133–35 market compensation mechanisms and, 102–106 technological change and, 182–83 Complementarities of IT, 58–59, 78–79, 289–91 Computer numerically controlled (CNC) technologies, 109 Computer-wage correlation, 12–13, 209 Computers. See also Information technology (IT); Organizational computing (OC) economy and, 279–80 evolution of, 55–58 general-purpose, 56–57, 279 personal, 285, 291 supplier interaction and, 61–63 Computopia, 109

300

Consumer price index (CPI) German, 51–52 U.S., 7, 25–28, 39, 51, 70–71 Consumption ‘‘easy-to-measure,’’ 20, 35–38 elasticities and, 41 food, 32, 35–37 growth rates of, 30–34 ‘‘hard-to-measure,’’ 21, 38, 40–41 housing, 32, 37–38 mismeasurement hypothesis and, 30– 34, 41 motor vehicle, 39–40 CPI. See Consumer price index Crawford, I., 52 Currency Stabilization Law (German), 51–52 Davis, S., 136, 245 Dell Computer, 65 ‘‘Deskilling’’ of workforce, 12, 175 Dewan, S., 67 Dickens, W., 199 Dictionary of Occupational Titles, 247 Diewert, W. E., 24 Differential measurement, 27–28 Diffusion of ICTs causes of, 4–5 employment and, 9–10, 101–102 at firm level, 14–15 at macro level, 15–16 at market level, 15 unemployment in Europe and, 182 Diffusion-based measures, 186–87 DiNardo, J., 209 Doms, M., 198, 209, 247 Dot-com boom, 293 Douglas, P. H., 104 Draper, N., 160, 172 Dre´ze, J. H., 167 Duchin, F., 118 Duguet, E., 198 Dunne, T., 198, 247, 253 Economic growth and ICTs, 1, 3–4 Economic measurement. See also specific types concerns about, 19 problems of, 41–42, 47–48 views of, 47 Efficient consumer response (ECR) programs, 63, 81

Index

Elasticities consumption and, 41 demand, lower, 106 expenditure, 34 income, 34 mismeasurement hypothesis and, 34– 37, 40–41 price, 183 substitution, 161, 171 wage, 171 Eldridge, L. P., 26 Electronic data interchange (EDI), 61–62 Empirical evidence on employment impact of innovation aggregate, 119–21, 126 input-output, 115–19, 126 microeconomic, 113–15 problems with, 111–13 sectoral, 115–19, 126 skill-biased technical change hypothesis and, 12 Empirical models, technological bias and testing hypothesis and, 159–64 unemployment and, 165–66 use of, 158–59 wage rigidities and, 164–65 Employee involvement (EI), 233, 240–42 Employment and ICTs. See also Empirical evidence on employment impact of innovation comments on, 133–37 compensation theory and, 102–108, 121 diffusion of ICTs and, 9–10, 101–102 end-of-work theory and, 108–11, 121, 133 homogenous labor and, 182–83 inequalities, 3 innovation and, 111–21 labor demand and, 255–57 limitations of measuring, 126 skilled versus unskilled workers and, 3–4 structure and, 12–16 summary of research findings, 121–26 technological progress and, 9–10, 101– 102 technology-job relationship and, 209–15 wages and, 12–16, 255–57 End-of-work theory, 108–11, 121, 133 ENIAC (1943 computer), 55 Enterprise resource planning software (ERP), 61

Index

Entorf, H., 114, 209 Equilibrium unemployment model equations in, 149–50 extensions of, 148 firm’s behavior and, 148–49 Europe ICTs in, 3–6 macroeconomic performance, 1 slowdown in productivity in, 29 technological bias in, 141–47, 162–63 unemployment in, 141–47, 165–66, 182 Evangelista, R., 116–17 Everyday low price (EDLP), 63 Excess returns, 72–73, 98 Expenditure elasticities, 34 Firm level co-invention at, 282–83 diffusion of ICTs at, 14–15 ICTs at, 12–15 IT analyses at, 65–79 productivity and IT at, 66–68, 93 Førre, S. E., 114 France end-of-work theory and, 108–109 ICT use in, 5 manufacturing organizational changes in, 271–73 organizational change survey in, 275– 76 unemployment rate in, 141–44, 166 unskilled labor in, 172–73 Fujimoto, T., 249 Garen, J., 179 General Motors (GM), 61, 238 Givord, P., 172 Glyn, A., 120 Gordon, R. J., 52 Gorz, A., 108 Goux, N., 197, 272 Greenan, N., 67, 95, 115, 198, 215 Griliches, Z., 20–21, 69, 197 Gross domestic product (GDP), 27, 80, 120 Gross national product (GNP), 1 Growth accounting calculations, 80–83 Guellec, D., 115, 215 Hall, R. E., 75 Haltiwanger, J., 136, 245 Hammour, M. L., 173

301

Haskel, J., 198 Heden, Y., 198 Hicks, J. R., 104, 107 Hitt, L., 6, 8, 66–69, 71, 73–74, 77–78, 93, 95–98, 247 Hoffmann, J., 51–52 Holzer, H. J., 248–49 Howell, D., 247 HRM practices ICTs and, 13–14 IT and, 242–44 Japanese system of, 240 labor demand and, 236–37 production process and, 239–43 in steel industry, 230, 232–37 Hulten, E., 245 Human resource management practices. See HRM practices ICTs. See also Diffusion of ICTs; Employment and ICTs; Innovation; IT economic consequences of, 1, 3–4, 5–6 at firm level, 12–15 in France, 5 in future, 291–96 HRM practices and, 13–14 inequality puzzle and, 10–16 manufacturing sector and, 116 at market level, 15 organizational change and, 269–70, 272–73 productivity puzzle and, 6–10, 247, 279–81 skill mix and, 12–14 unemployment and, 10–11 wages and, 3–4, 12–16 work content and organization and, 270–72 at worker level, 12–13 Imperfection of compensation, 133–35 Imports, 23–24, 251–52 Information and communication technologies. See ICTs Information technology. See IT Innovation. See also specific types co-invention and, 282–84, 289, 292 employment and ICTs and, 111–21 historical appreciation of, 25 ‘‘locked-in,’’ 107 new economy and, 8, 48–49 in organizational computing, 281–83 process, 102–103, 109–10

302

Innovation (cont.) productivity and, 21, 25–26 rate of introduction of, 25, 48 significance of, 24, 49 technology-jobs relationship and, 183 Input-output empirical evidence, 115–19, 126 Intangible outputs and IT, 68–72 Intel, 186 Internet, commercialization of, 291 IT. See also ICTs; specific types case studies, 58–65 complementarities and, 58–59, 78–79, 289–91 customer relationships and, 63–65 excess returns and, 72–73, 98 firm-level analyses and, 65–79 in future, 291–96 General Motors and, 238 general-purpose, 56–57, 279 homogenous labor and, 182–83 HRM practices and, 242–44 intangible outputs and, 68–72 labor demand and, 229–30, 236–50, 294–95 organizational interrelationships and, 8, 76–79 organizational practices and, 58–65 output growth and, 68–72, 80 productivity at firm level and, 66–68, 79, 93 steel industry and, 230–32 VCRs, 70–71 Walmart and, 238 Jackman, R., 104, 181 Japanese system of human resource management practices, 240 Jensen, J. B., 253 Jorgenson, D. W., 80 Kaldor, N., 182 Kalmbach, P., 118–19 Katz, L., 77–78, 181, 197, 199 Kelley, M. R., 67, 77 Kleinknecht, A., 114 Klette, T. J., 114 Knowledge capital, 254–55, 257, 269 Kong, P., 119–20 Kramarz, F., 209 Kriebel, C. H., 66 Krueger, A., 77–78, 181, 197, 199, 248

Index

Krugman, P., 180 Krusell, P., 160, 162, 172–73 Kurz, H. D., 118–19 Labor demand. See also Employment and ICTs HRM practices and, 236–37 IT and, 229–30, 236–50, 294–95 knowledge capital and, 254–55 organizational change and, 269–70 organizational computing and, 290–91, 295–96 steel industry and, 230–37 wages and, 255–57 Labor force composition changes, 153 Labor markets and organizational computing, 286–91 Ladder effect, 167–68 Laspeyres index number, 29–30 Lawler, E. E., 242 Layard, R., 104, 119, 156–57 Ledford, G. E., Jr., 242 Leontief, W., 118 Levy, F., 247 Lichtenberg, F. R., 66–67, 94–95, 199 Limited substitution, 287–89 Longitudinal Research Dataset (LRD), 197–98, 245, 254 Loveman, G. W., 66 Lucky-Goldstar chaebol, 61 Ludd, Ned, 101 Luddites, 101–102, 109 Lynch, L. M., 67–68, 247 McCully, C., 29 Machin, S., 114, 162, 181, 197–98 Mairesse, J., 67, 271 Malthus, T. R., 105 Manacorda, M., 157, 162–64, 166, 181 Manders, T., 160, 172 Manning, A., 157, 163, 181 Manufacturing sectors, 116, 271–73. See also Steel industry Marconi, G., 56 Mark I (1939 computer), 55 Market level, diffusion of ICTs at, 15 Market share effects on technology-job relationship, 183 Marshall, A., 24, 104 Marx, K., 102, 105, 175 Maurin, E., 197, 247, 272 Meyer-Krahmer, F., 119

Index

Microelectronics, development of, 109– 10 Milgrom, P., 59 Mill, J. S., 105 Millward, N., 114 Min, C.-K., 67 Mincer, J., 199 Minimum wage laws, 165 Mismatch indicators, 11, 156–57 Mismeasurement hypothesis consumption and, 30–34, 41 German economy and, 51–53 growth in productivity and, 26–27 housing consumption and, 32, 37–38 services sector and, 38–39, 50 slowdown in productivity, 7, 11, 19–26, 48–51 Solow productivity paradox and, 7, 11, 21 U.S. productivity and, 29–41 Mohrman, S. A., 242 Mokyr, J., 24 Morrison, C. J., 77 Mukhopadhyay, T., 66 Murname, R., 247 Murphy, K., 181 Neary, J. P., 104 Netherlands, skill structure in, 163 New economy ICTs and, 1, 3 innovation and, 8, 48–49 mismeasurement hypothesis and, 21– 26, 48–49 slowdown in productivity and, 21–26, 48–49 Solow productivity paradox, 3 Nickell, S., 104, 119–20, 181 Noble, D. F., 109 OC. See Organizational computing Organizational change in French manufacturing practices, 271– 76 labor demand and, 269–70 measurement issues in, 274–75 skills and, 272–73 Organizational computing (OC) adoption of, 281 applications of, 284–85 Organizational practices and IT, 58–65 Osterman, P., 242, 249

303

Oswald, A., 114 Output growth and IT, 68–72, 80 Padalino, S., 120–21 Pasinetti, L., 108 Perani, G., 116 Personal computers (PCs), 285, 291 Personal consumption expenditures (PCEs), 27, 35, 41 Petrongolo, B., 157, 162–64, 166 Pianta, M., 116–17 Pigou, A., 104 Pini, P., 120 Pischke, J. S., 209 Pohlmeier, W., 114 Price index. See Consumer price index (CPI) Procter and Gamble (P&G), 63 Production Fordist model of, 107 HRM practices and, 239–43 Productivity. See also Growth in productivity; Slowdown in productivity; Solow productivity paradox ICTs and, 1, 3–4, 7–8 innovation and, 21, 25–26 and IT at firm level, 66–68, 79, 93 mismeasurement hypothesis evidence and, 29–41 nonfarm multifactor (1949 to 1973), 19 organizational computing and, 283–86 of skilled versus unskilled workers, 12 Productivity puzzle and ICTs, 6–10, 247, 279–81 Prost, C., 172 Quality circles, 240 Quality improvements, 24–25, 68–69 R&D expenditures, 94, 184–87, 253 Raff, D. M. G., 25 Randolph, W. C., 37 Reed, S. B., 37 Reijnen, J. O. N., 114 Ricardo, D., 102–104 Rifkin, J., 111 Risager, O., 165 Roach, S., 66 Robbins, L., 104 Roberts, J., 59

304

Roosevelt, F. D., 47, 53 Rosenblum, L. S., 77 Saint Martin, A., 173 SAP R/3, 61 Say’s law, 103, 106 Schankermann, M., 74 Science Policy Research Unit. See SPRU Services sector, 20–21, 32, 38–39 Shadman-Mehta, F., 159, 166, 171–72 Shaw, K., 6, 13, 172, 269 Signet Bank, 64 Sinclair, P. J. N., 104, 119 Sismondi, J. C. L., 105 Skill-biased technological change, 10, 12, 14, 176–79 Skills bias in, labor demand and, 253 changes in, in new and old plants, 246– 50 complementarities with, 289–91 measurement, 187 organizational change and, 272–73 technology-jobs relationship and, 188– 89 Smolny, W., 115 Sneessens, H. R., 6, 11, 159, 166, 171–73 Solow productivity paradox beyond, 97–99 mismeasurement hypothesis and, 7, 11, 21 new economy and, 3 paradox of, 135 skepticism about, 93–95 statistics and, 7–8 Solow, R., 3, 22 Spiezia, V., 6, 9, 133, 136 SPRU, 114, 186 Steel industry economic change and, 13 employment shocks in, 250–52 HMR practices and, 230, 232–37 imports and, 251–52 IT and, 230–32 labor demand and, 230–37 technological change in, 250–51 wage increases in, 252 Stern, S., 76 Stewart, K. J., 37 Stiroh, K., 80 Strassmann, P. E., 66 Structural shocks

Index

mismatch indicators and, 156–57 technological bias and, 151, 153–56 Substitution bias, 29–30 Supplier interactions and IT, 61–63, 292– 93 Technological bias analytical framework, 148–50, 160–67 in Europe, 141–47, 162–63 ICTs and, 10–11 macroeconomic shocks and, 151–53 unemployment and, 10–11, 165–66 in U.S., 141–47, 163 wage rigidities and, 164–65 Technological capital, 176, 198 Technological change biased, 165–66 market compensation mechanisms and, 182–83 model illustrating, 183, 216–17 in steel industry, 250–51 Technological progress biases, 154, 159–66 employment and ICTs and, 9–10, 101– 102 end-of-work theory and, 109 localized, 106–107 market compensation mechanisms and, 102–106 in OECD countries, 162–163 Technology shocks, 180, 254 Technology-jobs relationship econometric models, 184–87 employment and, 209–15 evolution of, 175–76 innovation and, 183 microeconomic perspective of, 216–17 results of research on, 187–91 skills and, 188–99 union effects on, 183 wages and, 199–209 Thesmar, D., 247 Total quality management (TQM), 237– 38, 240–42 Trajtenberg, M., 25 Triplett, J., 6–8, 39, 47–51 Troske, K. R., 247, 253 Turing, A., 56 Unemployment and ICTs empirical models and technological bias and, 165–66

Index

in Europe and, 141–47, 165–66, 182 in OECD countries, 145–47 skills bias and, 180–81 technological bias and, 10–11, 165–66 in U.S., 141–44 U.S. Bureau of Economic Analysis, 8, 26, 29, 31, 35, 37, 40–41, 49 U.S. Bureau of Labor Statistics, 7–8, 26– 27, 40, 49 Van Reenen, J., 6, 12, 114, 181, 197, 209, 217, 225–27, 258 Venables, A. J., 104 Vivarelli, M., 6, 9, 117, 120–21, 133, 136 Wadhwani, S., 114 Wages inequality in, 3, 229, 254–55 rigidities in, 164–65 Walmart, 238, 245 Whitley, J. D., 118 Wilson, R. A., 118 Wolff, E., 247 Workplace Industrial Relations Survey (WIRS), 198 Yang, S., 75 Zamagni, S., 104

305

Productivity, Inequality, and the Digital Economy: A Transatlantic Perspective

The European Productivity Agency and Transatlantic Relations, 1953-1961

Athenian Economy and Society: A Banking Perspective

The Spanish Economy: A General Equilibrium Perspective

The New Economy in Transatlantic Perspective (Routledge Studies in Governance and Change in the Global Era)

Digital Deflation : The Productivity Revolution and How It Will Ignite the Economy

Industrial Organization and the Digital Economy

The digital economy fact book

Digital Integrated Circuits, A Design Perspective

Digital Signal Processing: A Computer Science Perspective

Digital Signal Processing: A Computer Science Perspective

Digital Signal Processing A Computer Science Perspective

Digital Signal Processing: A Computer Science Perspective

A Transatlantic Tunnel Hurrah

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel

a Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel

Global Political Economy in the Information Age: Power and Inequality

Complex Inequality : Gender, Class, and Race in the New Economy

Inequality and the State

The Global Economy in the 1990s: A Long-Run Perspective

Digital Economy:: Impacts, Influences and Challenges

Digital Economy: Impacts, Influences, and Challenges

Digital Economy: Impacts, Influences and Challenges

Race Against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy

Race Against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy

Asian Economy and Finance: A Post-Crisis Perspective

The Singapore Economy: An Econometric Perspective

Productivity, Inequality, and the Digital Economy: A Transatlantic Perspective

The European Productivity Agency and Transatlantic Relations, 1953-1961

Athenian Economy and Society: A Banking Perspective

The Spanish Economy: A General Equilibrium Perspective

The New Economy in Transatlantic Perspective (Routledge Studies in Governance and Change in the Global Era)

Digital Deflation : The Productivity Revolution and How It Will Ignite the Economy

Industrial Organization and the Digital Economy

The digital economy fact book

Digital Integrated Circuits, A Design Perspective

Digital Signal Processing: A Computer Science Perspective

Digital Signal Processing: A Computer Science Perspective

Digital Signal Processing A Computer Science Perspective

Digital Signal Processing: A Computer Science Perspective

A Transatlantic Tunnel Hurrah

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel

a Transatlantic Tunnel, Hurrah!

A Transatlantic Tunnel

Global Political Economy in the Information Age: Power and Inequality

Complex Inequality : Gender, Class, and Race in the New Economy

Inequality and the State

The Global Economy in the 1990s: A Long-Run Perspective

Digital Economy:: Impacts, Influences and Challenges

Digital Economy: Impacts, Influences, and Challenges

Digital Economy: Impacts, Influences and Challenges

Race Against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy

Race Against the Machine: How the Digital Revolution is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy

Asian Economy and Finance: A Post-Crisis Perspective

The Singapore Economy: An Econometric Perspective

Recommend Documents