TAX POLICY
AND THE ECONOMY 16 edited by James M. Poterba
National Bureau of Economic Research The MIT Press, Cambridge, Massachusetts
Tax Policy and the Economy, Volume 16, 2002 ISSN: 9892-8649 E-ISSN: 1537-2650 ISBN: Hardcover 0-262-16210-5 Paperback 0-262-66129-2
Published annually by The MIT Press, Cambridge, Massachusetts 02142 An electronic, full-text version of Tax Policy and the Economy is available from MIT Press
Journals when purchasing a subscription. Subscription Rates
Hardcover/Print and Electronic: $58.00 Paperback/Print and Electronic: $25.00 Outside the U.S. and Canada add $10.00 for postage and handling. Canadians add 7% GST.
Subscription and address changes should be addressed to: MiT Press Journals, Five Cambridge Center, Cambridge, MA 02142-1407, phone 617253-2889; fax 617-577-1545; email
[email protected]. Claims wifi be honored free of charge if made within three months of the publication date of the issue. Claims may be submitted to
[email protected]. Prices are subject to change without notice. In the United Kingdom, continental Europe, and the Middle East and Africa, send back volume orders and business correspondence to:
The MIT Press, Ltd., Fitzroy House, 11 Chenies Street, London WC1E 7ET England, phone 44-020-7306-0603, fax 44-020-7306-0604, email
[email protected] In the United States and for all other countries, send single copy and back volume orders to:
The MIT Press, Five Cambridge Center, Cambridge, MA 02142, toll-free book orders 800-356-0343, fax 617-625-6660, email
[email protected] Copyright Information
Permission to photocopy articles for internal or personal use, or the internal or personal use of specific clients, is granted by the copyright owner for users registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided that the fee of $10.00 per copy is paid directly to CCC, 222 Rosewood Drive, Danvers, MA 01923. The fee code for users of the Transactional Reporting Service is: 9892-8649/02 $10.00. For those organizations that have been granted a photocopy license with CCC, a separate system of payment has been arranged. © 2002 by the National Bureau of Economic Research and the Massachusetts Institute of Technology.
NBER BOARD OF DIRECTORS BY AFFILIATION Officers: Carl F. Christ, Chairman Michael H. Moskow, Vice Chairman Martin Feldstein, President and Chief
Robert Mednick, Treasurer Kelly Horak, Controller and Assistant
Executive Officer
Gerardine Johnson, Assistant
Corporate Secretary
Susan Colligan, Vice President for Administration and Budget and
Corporate Secretary
Corporate Secretary
Directors at Large: Peter C. Aldrich Elizabeth E. Bailey John H. Biggs Andrew Brimmer Carl F. Christ John S. Clarkeson Don R. Conlan George C. Eads
Martin Feldstein Stephen Friedman Judith M. Gueron George Hatsopoulos Karen N. Horn Judy C. Lewent John Lipsky Michael H. Moskow
Alicia H. Munnell Rudolph A. Oswald Robert T. Parry Peter G. Peterson Richard N. Rosett Kathleen P. Utgoff Marina v. N. Whitman Martin B. Zimmerman
Directors by University Appointment: George Akerlof, Cahfornia, Berkeley Jagdish Bhagwati, Columbia William C. Brainard, Yale Michael J. Brennan, Cahfornia, Los Angeles
Glen G. Cain, Wisconsin Franklin Fisher, Massachusetts Institute of Technology
Saul H. Hymans, Michigan
Marjorie B. McElroy, Duke Joel Mokyr, Northwestern Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford Michael Rothschild, Princeton Craig Swan, Minnesota David B. Yoffie, Harvard Arnold Zellner, Chicago
Directors by Appointment of Other Organizations: Mark Drabenstott, American
Richard D. Rippe, National
Agricultural Economics Association Gail D. Fosler, The Conference Board
John J. Siegfried, American Economic
A. Ronald Gallant, American Statistical Association
Robert S. Hamada, American Finance Association
Robert Mednick, American Institute of Certfied Public Accountants Angelo Melino, Canadian Economics Association
Association for Business Economics Association
David A. Smith, American Federation of Labor and Congress of Industrial Organizations Josh S. Weston, Committee for Economic Development
Gavin Wright, Economic History Association
Directors Emeriti: Thomas D. Flynn Lawrence R. Klein
Franklin A. Lindsay Paul W. McCracken
Bert Seidman Eli Shapiro
Since this volume is a record of conference proceedings, it has been exempted from the rules governing critical review of manuscripts by the Board of Directors of the National Bureau (resolution adopted 8 June 1948, as revised 21 November 1949 and 20 April 1968).
CONTENTS Introduction: James M. Poterba vii
Acknowledgments xi ECONOMIC EFFECTS OF MEANS-TESTED TRANSFERS IN THE U.S. 1 Robert Moffitt
TAXES AND HEALTH INSURANCE 37 Jonathan Gruber
DEFINED CONTRIBUTION PENSIONS: PLAN RULES, PARTICIPANT CHOICES, AND THE PATH OF LEAST
RESISTANCE 67 James J. Choi, David Laibson, Brigitte C. Madrian, and Andrew Metrick
THINKING STRAIGHT ABOUT THE TAXATION OF ELECTRONIC COMMERCE: TAX PRINCIPLES, COMPLIANCE PROBLEMS, AND NEXUS 115 Charles E. McLure, Jr.
THE FISCAL EFFECTS OF POPULATION AGING IN THE U.S.: ASSESSING THE UNCERTAINTIES 141 Ronald Lee and Ryan Edwards
POTENTIAL PATHS OF SOCIAL SECURITY REFORM 181 Martin Feldstein and Andrew Samwick
INTRODUCTION James M. Poterba MIT and NBER
The annual NBER Tax Policy and the Economy conference began as a vehicle for communicating current findings from NBER research in the areas of taxation and government spending to policy analysts in both the government and the private sector. It has now been emulated in other
fields, with new research meetings on health policy and innovation policy, designed to achieve similar objectives. Tax Policy and the Economy
papers have addressed a wide range of issues. The six papers in this year's volume are no exception. They touch on both taxation and govern-
ment expenditure programs, and they include discussion of both very current policy topics and issues of long-standing and ongoing interest. The first paper, Robert Moffitt's "Economic Effects of Means-Tested Transfers in the U.S.," provides a wide-ranging analysis of government programs that provide income or in-kind assistance to individuals and families who satisfy particular income constraints. The paper summarizes the results of a recent NBER study on how these programs affect incentives for labor supply, for changes in family status, and for a range of other behaviors. This recent research concludes that these programs have important effects on their potential and actual recipients. In "Taxes and Health Insurance," Jonathan Gruber explores the many channels through which the income and payroll tax system affects the demand for health insurance. The paper begins by describing the key behavioral elasticities that are needed to evaluate potential tax reforms. These are the price sensitivity of insurance demand by individuals, the price sensitivity of employers' decisions about whether or not to offer health insurance, and the price sensitivity of worker's discrete decision whether or not to purchase health insurance conditional on it being
viü Introduction
offered by the employer. Gruber then surveys the existing empirical literature that bears on each of these issues, and he uses mid-range estimates from these studies to describe how potential changes in the
federal income tax might affect health insurance coverage. He concludes
that removing the current income tax subsidy to employer-provided health insurance, while leaving other tax subsidies, such as those through the payroll tax, in place, could reduce the number of persons with health insurance by nearly fourteen million. In "Defined Contribution Pensions: Plan Rules, Participant Choices, and the Path of Least Resistance," James Choi, David Laibson, Brigitte Madrian, and Andrew Metrick analyze the heterogeneity in the retire-
ment saving plans that firms offer their workers. They not only describe the plan-to-plan variation, but they also use this variation to investigate how various plan features affect voluntary enrollment and contribution decisions. They identify a number of plan features that are correlated
with high levels of participation in, and contributions to, retirement
saving plans. These include automatic enrollment programs and allowing a relatively large fraction of salary to be subject to an employer matching contribution. The paper concludes that a significant population of employees simply adopt the default option that their employer offers, with respect to both the fraction of their salary that they contribute to the plan and the asset allocation they choose within the plan. This suggests that it is important for firms to carefully evaluate the default options that they offer in their retirement plans. In the fourth paper, Charles McLure tackles the very current issue of taxing e-commerce. In his paper on "Thinking Straight about the Taxation of Electronic Commerce: Tax Principles, Compliance Problems, and Nexus," McLure places the Internet Tax Freedom Act in context by describing the general problem of taxing sales in an economy with multiple jurisdictions. The need to determine whether a seller has nexus in a given jurisdiction, which is central to the taxation of electronic commerce, arises in many other settings as well. The paper outlines general rules for designing a sales tax that minimizes economic distortions, and then explains the set of tax rules on Internet commerce that would be most consistent with these principles. The last two papers address issues related to social security policy. Ronald Lee and Ryan Edwards, in their paper on "The Fiscal Effects of Population Aging in the U.S.: Assessing the Uncertainties," present important evidence on the uncertainties associated with demographic projections. They note that most current projections of how aging will affect federal spending rely on scenario analyses that make a fixed set of assumptions about key parameter values, such as birth rates and migration
Introduction
ix
rates, and then evaluate the long-run fiscal consequences of various government policies. Lee and Edwards present an alternative methodology for addressing these issues, using stochastic simulation. Their results highlight the potential uncertainties associated with any single-scenario description of the long-run prospects for a given fiscal policy. They also suggest that it is risky to advertise policy reforms, such as an increase of a
given amount in the payroll tax, as "fixes" for problems such as the
potential insolvency of the Social Security Trust Fund. The uncertainty of future birth rates and death rates means that there is still a chance that such policies will result in an excess, or a shortfall, of resources for public programs. Finally, in the last paper, "Potential Paths of Social Security Reform," Martin Feldstein and Andrew Samwick consider a number of potential reforms of the current social security system. Each reform combines a
system of personal retirement accounts with a variant of the current defined benefit system. The paper evaluates how each of the various reforms would affect the benefits available to future retirees. It also considers how the reforms would affect national saving and the solvency of the social security trust fund. The paper addresses the risk of individual account investments and presents evidence on the range of possible benefit outcomes under different assumptions about the long-run average return on corporate equities. The papers in this volume illustrate the type of policy-relevant research that is carried out by the affiliates of the NBER Public Economics Program. Each of the papers provides important background informa-
tion for policy analysis, without making recommendations about the merits or demerits of particular policy options. I hope that each of these
papers wifi provide useful input to various participants in the policy process who are concerned with tax and expenditure program design.
ACKNOWLEDGMENTS In planning and organizing this year's Tax Policy and the Economy conference and the associated volume, I have incurred debts to many individuals. Martin Feldstein, President of the NBER, has been an active sup-
porter of this conference throughout its sixteen-year history. Jennifer Dayton did an outstanding job in updating the invitation list and in disseminating information about the conference to potential participants. The members of the NBER Conference Department, notably Lita Kimble, Rob Shannon, and Conference Department Director Carl Beck, handled conference logistics with efficiency and good cheer. Helena Fitz-Patrick oversaw the publication process with outstanding attention to detail and with exceptional speed and efficiency. Jam also grateful to R. Glenn Hubbard, the Chairman of the President's Council of Economic Advisers, for delivering a fascinating set of luncheon remarks at the conference at which these papers were presented. Glenn explained how the tragic terrorist attacks of September 11 had raised new questions about the appropriate role of government in the economy, and
he sketched many individual contexts in which there is ongoing debate about the government's role. Finally, I wish to thank each of the authors whose papers are included in this volume for striving to communicate their important research findings in a readable and clear fashion. I appreciate their efforts and willingness to participate in this very important opportunity for interchange between academics and policymakers.
ECONOMIC EFFECTS OF MEANS-TESTED TRANSFERS IN THE U.S. Robert Moffitt Johns Hopkins University and National Bureau of Economic Research
EXECUTIVE SUMMARY The system of means-tested transfers in the U.S. has evolved in important ways over the last decade, with significant expansions of Medic-
aid, the Earned Income Tax Credit, and the Supplemental Security Income program, and with significant contraction in Aid to Families with Dependent Children, now titled the Temporary Assistance for Needy Families program. To determine where we are in our understanding of each of these programs, as well as the other major programs in the system of means-tested transfers, a volume is under preparation by the National Bureau of Economic Research that surveys the current structure and historical evolution of each of these programs and
that synthesizes the results of the research that has been conducted on their economic effects. In addition to the AFDCTANF, Medicaid, EITC, and SSI programs, reviews have been conducted for the Food Stamp program and for housing, child care, job training, and child support programs. This paper summarizes the results of those reviews and highlights the large number of important findings from existing research. The author would like to thank the Smith-Richardson Foundation for support for the conference and NBER volume upon which this paper is based. Comments from Robert LaLonde, Edgar Olson, and Karl Scholz are appreciated.
2
Moffitt
Reform of the system of means-tested transfers in the U.S. continues to be an important topic for public policy as well as an area of continued
research by economists. Policy and research interest have been kept particularly high by significant transformations in the means-tested transfer system over the last decade. The most important structural changes have taken place in three programs. One is the Aid to Families with Dependent Children (AFDC) programnow named the Temporary Assistance for Needy Families (TANF) programwhose generosity has been significantly reduced and whose eligibility conditions have been restricted to those who can and are willing to comply with work requirements and other new rules. A second is the Medicaid program, which has been significantly expanded to cover more families and children off the AFDC-TANF program. The third is the Earned Income Tax Credit (EITC), whose benefits have greatly expanded and whose expenditures now exceed those in the traditional AFDC-TANF program. A
fourth program which has undergone significant expenditure and caseload expansion, although without major structural change, is the Supplemental Security Income (SSI) program.
To determine where we stand in our understanding of each of these programs, as well as the other major programs in the system of meanstested transfers, a volume is under preparation by the National Bureau of Economic Research that surveys the current structure and historical evolution of each of these programs and that synthesizes the results of the research that has been conducted on their economic effects (Moffitt, forthcoming). In addition to the AFDC-TANF, Medicaid, EITC, and SSI programs, reviews have been conducted for the Food Stamp program and for housing, child care, job training, and child support programs. This paper summarizes the results of those reviews. The paper first provides a brief background discussion of trends in expenditures on means-tested transfers as a whole. It then goes on to discuss each of the major programs individually.
1. OVERALL TRENDS IN EXPENDITURES IN MEANS-TESTED TRANSFER PROGRAMS Figure 1 shows trends since 1968 in per capita expenditures in the eighty
largest means-tested transfer programs in the country. The figure reveals that there have been four phases of spending growth: an expansionary phase beginning in the 1960s and running through the early or mid-1970s, a contractionary (or stationary) phase beginning in the mid1970s and running until the mid-1980s, another expansionary phase
Economic Effects of Means-Tested Transfers in the U.s.
3
FIGURE 1. Real Per Capita Expenditures on Means-Tested Transfers, 1968-1998
1965
1970
1975
1960
1955
1990
1995
2000
Year
Sources: Burke (1999, Tables 4 and 5), U.S. Dept. of Commerce (2000, Table 2, Population).
running from the late 1980s to the mid-1990s, and another contractionary (or stationary) phase beginning in the mid-1990s. The first phase saw an increase in AFDC benefits; enactment of a
major piece of welfare legislationthe 1967 Social Security Amendmentswhich raised earnings disregards in the program (i.e., lowered the tax rate on earnings); and the creation of the food stamp and Medicaid programs and, later in the period, the Supplemental Security Income
program. Caseloads grew rapidly in all four of these programs. This period was later termed the era of the "welfare explosion" and set the modern framework of means-tested transfers. The second phase saw a steady decline in real AFDC benefits; enactment of a major piece of AFDC legislationthe 1981 Omnibus Budget Reconciliation Actwhich effectively eliminated the earnings disregards enacted in 1967 and consequently cut thousands of families with earnings from the rolls; and witnessed an increasing interest in work requirements and mandatory training programs for welfare recipients among federal policymakers. Declining real AFDC benefits were accompanied by slow but steady growth in the number of single-mother families, and the offsetting effects of these two forces left AFDC expenditures more or less unchanged in real terms.
4 Moffitt
The third phasewhich is not always recognized, for it is often presumed that the system has been in steady contraction since the 1970s saw a dramatic expansion of the Earned Income Tax Credit (EITC); major
expansions of eligibility in the Medicaid program, primarily to nonAFDC families; and sizable expansions of the caseload in the SSI program, arising mostly from increased numbers of disabled adults and children. The Family Support Act of 1988, although occurring in the third phase and seemingly contractionaryit mandated work and training for AFDC recipients more heavily than in the pastis best viewed as neutral, for not only was it never effectively implemented, but it also could be interpreted as expansionary inasmuch as it required new expenditures on work programs for AFDC recipients. The runup of expendi-
tures in this period, although not quite as large in magnitude as that resulting from the welfare explosion of the late 1960s and early 1970s, occurred much more quicklyessentially occurring in a five-year period between 1990 and 1995.
The fourth phase, which is continuing at this writing, is a combined result of 1996 welfare legislation, which contracted the AFDC-TANF program, and a robust economy which has led to declining caseloads in many programs, thereby slowing expenditure growth. The Food Stamp and Medicaid programs, as well as AFDC-TANF, have seen declining caseloads.1
Table 1 shows, in more detail, the sources of expenditure growth in the third, expansionary period. AFDC expenditures actually declined, presaging the further decline which has occurred subsequent to the 1996 legislation. The Food Stamp program expanded by 42 percent, however, indicating robust growth. A very large percentage expansion occurred in the Medicaid program, which grew by 88 percent. As wifi be discussed further below, the Medicaid program covers different types of recipients, and the growth over this period came not only from expansions of expenditures for single mothers and their children, but also from increased expenditures on the disabled. While single mothers and their children represent the largest fraction of the Medicaid caseload, expenditures are greater for the disabled because of their greater medical needs. The largest percentage expansion in Table 1, however, occurred in the EITC program, whose expenditures almost tripled over the period. As wifi be discussed below, major expansions of the size of the credit re-
sulted in this growth. Housing programs grew modestly during the The unemployment rate appears to have started to increase in late 2000 or early 2001, possibly indicating the beginning of a recession. Whether this will signal the beginning of a fifth phase or a modification of the fourth remains to be seen, and wifi depend on legislative developments and on the course of expenditure growth over the next few years. 1
Economic Effects of Means-Tested Transfers in the U.S.
5
TABLE 1
Change in Real Expenditures' in Six Major Programs, FY1990 to FY1996 Expenditure (million 1996 dollars)
Program AFDC
Food Stamps Medicaid EITCb
Housingc SSI
1990
1996
24,758 20,654 84,658 8,092 16,922 20,125
23,677 27,344 159,357 24,088 19,877 32,065
Share of
Change
growth
(%)
(%)
4
1
42 88 198 17 59
7 60 13
4 10
Sources: Burke (1993, Table 15; 1999, Table 3, 12).
'Federal and state combined totals. blncludes reduction in tax liability, not just refundable portion. cSum of expenditures on public and Section 8 housing.
period, but the SSI program grew by a large amount, 59 percent, reflecting, as in Medicaid, increases in expenditures on the disabled.2 The last row of Table 1 shows the shares of total expenditure growth in the largest eighty means-tested transfers from 1990 to 1996 accounted for by each of these six programs. Medicaid expenditure growth, while not the largest in percentage terms, is the largest in dollar terms and accounts for the largest fraction, 60 percent. The EITC and SSI together account for another 23 percent. Altogether, these six programs accounted for 93 percent of the overall increase in means-tested expenditures in the 1990-1996 expansionary phase. Finally, Table 2 shows the expenditures and caseloads in the ten largest means-tested transfer programs in FY 1997. The largest is Medicaid,
as expected, and the next fiveSSI, EITC, Food Stamps, TANF, and subsidized housingare of the same general magnitude but at a large distance from Medicaid. The TANF program, which in the 1960s was the largest of the six, is now a distant fifth in rank. The evolution of means-tested transfers which has led to the ranking in Table 2 reflects several trends. One is the gradual decline of unrestricted cash transfers like AFDC relative to in-kind transfers like Medicaid, Food If medical care prices are used to deflate Medicaid expenditures instead of a general price index, Medicaid expenditure growth amounted to only 34 percent. Which index should be used depends on whether the goal is to value expenditures from the point of 2
view of the taxpayer or the recipient.
Moffitt
6
TABLE 2
Annual Expenditures and Caseloads in Ten Largest Transfer Programs, FY1997 Expenditures Program
($ million)
Medicaid
167,359 32,395 28,800 24,772 23,179 19,336 9,220
SSI
EITC FS
TANF
Subsidized housing Medical care for veterans without service disability Foster care Social service Federal Pell grants
6,794 6,400 5,660
Caseload (thousands)
Expenditures per recipient ($)
153
4,138 4,638 495 1,024 2,120 4,481 60,261
289
23,509
40,446 6,984 58,143 24,200 10,936 4,315
NA
NA
3,665
1,544
Source: Burke (1999, Table 12).
Stamps, and housing. Voters and legislators appear to prefer to make transfers tied to specific consumption items rather than open-ended cash
transfers. A second is the increasing narrowness of the targeting of transfers: the programs which have seen the largest growth in the last decade are tied to specific eligibffity groups. The EITC is specifically targeted to families with earnings, the SSI program is targeted to the disabled and elderly, and Medicaid is targeted to the disabled andin the expansions
that have occurredmainly to single mothers and their children off TANF. This development represents a continued, if not increased, cate-
gorization of the nation's welfare population into a system in which different demographic groups are judged to be needy not just on the
basis of income but on the basis of some other specific characteristic that leads them to be deserving in the eyes of the public. This also explains why the EITC and SSI programs, which provide tied cash transfers, have expanded while the AFDCTANF program has not. As a consequence of these developments, the great expenditure expansion of the late 1980s and early 1990s increased total transfers to the low-income population but also changed the distribution of those transfers. Families off welfare
with earnings and the disabled gained, for example, relative to lowincome single-mother families as a whole, particularly those on welfare or not working.
The transfer programs reviewed in the forthcoming NBER volume include the six largest programs shown in Table 2. In addition, several
Economic Effects of Means-Tested Transfers in the U.s.
7
smaller but important programs are covered. These include child care programs (approximate FY 1999 expenditures of $17 billion across all programs), programs for child-support enforcement ($3 billion in 1996), and job training programs for the disadvantaged (expenditures of $1 billion in 1998). These nine programs wifi be discussed in this chapter roughly in order
of their total expenditures. For each program, the discussion first covers the structure and rules of the program and its historical evolution, followed by a discussion of trends in expenditures and caseloads and recipient characteristics, then followed by a review of research findings.
2. MEDICAID The Medicaid program, as noted by Gruber (forthcoming), is really four separate programs rolled into one. One supports the medical expenses of low-income single mothers and their children. The other three provide public insurance for portions of medical expenditures not covered by Medicare for the low-income elderly, support medical expenses for the low-income disabled, and provide coverage of nursing-home expenditures of the institutionalized elderly. The first program has a majority of the recipients, but the other three programs are responsible for a majority of the expenditures. Medicaid was created in 1965 by the same legislation that created the Medicare program. It is administered by the states, which must operate under federal regulation, and the federal government pays a fixed share of state expenditures (the state share, determined by a formula involving various state characteristics including median income, is approximately 43 percent). The program was initially aimed at providing medical benefits to traditional welfare populationslow-income single mothers and children, and the aged, blind, and disabled. However, over time eligibility has been expanded to other groups. Early in the program some coverage was extended to low-income children in two-parent families, and a Medically Needy program was instituted which provided care for low-income families (usually single-mother families) with income too high for welfare eligibility, albeit with numerous restrictions on eligibility. Beginning in 1984, and accelerating after 1987, more significant expansions were first allowed, and then mandated, requiring states to cover children in families with incomes below 133 percent of the poverty line, or higher at state option. Pregnant women were also covered, but otherwise there was no expanded coverage for adults. These expansions are part of the reason for the expenditure increase discussed in the last section. A further major expansion took place in 1997 with the creation
8
Moffitt
of the Children's Health Insurance Program, which provides a capped federal match for state creation of programs to cover groups outside of existing Medicaid eligibility or with higher incomes. Some states have chosen to expand their Medicaid programs, while others have created wholly new programs to cover these additional groups. Medicaid mandates a specific list of services that states must provide to all "categorically needy" recipients. States may go beyond this at their option, but few do, and when they do, most states cover the same types of extra services; as a result, there is substantial uniformity in the service package across states. Reimbursement rates, on the other hand, are given much more leeway, and there is major cross-state variation. Reimbursement rates are generally quite low and below those of Medicare and private payers. States are allowed waivers to experiment with different options for care provision, and the major direction states have pursued is the use of managed care for their Medicaid caseloads. By 1998,54 percent of Medicaid recipients were in managed care plans. Gruber shows that both expenditures and enrollment in the Medicaid program have increased enormously over the last decade, as indicated in the background discussion above. The major enrollment growth has been among the disabled and among children under 21. Enrollment growth has slowed in recent years, possibly because of 1996 federal welfare reform legislation which contracted the AFDCTANF program. Calculations of participation rates in the Medicaid program have been computed only for children and pregnant women, because they are the only groups for which eligibility has been calculated; no estimates are
available for the elderly and disabled. Eligibility has expanded greatly, as already noted, but takeup has slipped behind significantly, resulting in declines in participation rates. Whereas participation rates among eligible children were close to 100 percent prior to 1989, by 1996 they had fallen to 73 percent. There has been a great deal of research by economists on the Medicaid
program. One issue concerns reasons for the declining participation rates just noted. Research has shown that much of the explanation is that the increased eligibility arising from coverage expansions was to groups with higher than usual incomes, groups with less need for insurance; to groups outside the AFDC program and who therefore do not have the relatively easy institutional access to the program that welfare recipients do; and to groups who already are covered by other forms of
insurance. This last finding is related to a significant area of research on Medicaid crowdout, which occurs when Medicaid expansions result in substitutions of Medicaid coverage for private insurance coverage. There are a variety of empirical estimates of the extent of crowdout, some
Economic Effects of Means-Tested Transfers in the U.s.
9
indicating relatively small and some quite large effects. For example, in the latter category estimates have indicated that approximately 50 percent of those who have taken up Medicaid would have been privately covered otherwise. Research continues in this area in an attempt to resolve the differences in the magnitude of the effect. Another area of research concerns the effect of the Medicaid expansions on health and health outcomes. The studies which use nationwide data rather than data from individual states typically show significant positive effects of the Medicaid expansions on infant mortality, prenatal care utilization, and child preventative care, and that they led to more hospitalizations (but fewer "avoidable" ones). Research indicates that the positive effects are larger for those in demographic groups with typically worse health, such as black families, immigrants, and those with low educational levels. Effects are also larger for targeted expansions which are aimed at low-income mothers and children than for broad expansions which reach further up the income distribution. The effect of the Medicaid program on the labor supply of recipients and on their AFDCTANF participation decisions has been another focus on research, concentrating on single mothers and their children rather than the elderly and disabled. There is a range of research using different methodologies, and virtually all of it shows that the close historical tie of Medicaid eligibility to AFDC receipt tended to increase AFDC participation rates, and that the Medicaid expansions which loosened that tie also tended to reduce AFDC participation rates. Also, because AFDC has some work disincentives, the historical link has tended to decrease labor supply, while the Medicaid expansions have tended to increase it. Although the magnitudes of these effects are not precisely estimated in the literature, their direction is supported by most studies. Related work on the effect of Medicaid expansions demonstrates that they lowered savings and increased consumption, consistent with the notion that welfare recipients engage in less precautionary savings when they know that the program wifi support them should their income decline. Finally, there has been research on the effects of reimbursement policy and long-term care provision in Medicaid. The literature on physician reimbursement rates mostly shows that higher reimbursement rates lead to somewhat increased participation by physicians in the program, increased access to care, and occasionally better health outcomes, although the linkage between reimbursement policy and utilization and health is far from simple. With regard to long-term care, research indicates that Medicaid recipients are often on long waiting lists and have less access to care than private pay patients, but also that increases in subsidies to nursing home care raise overall nursing home utilization.
10
Moffitt
Some other research raises the issue of whether nursing home quality might be reduced as nursing homes increase the percentage of their patients who are minimum-pay Medicaid recipients.
3. THE SUPPLEMENTAL SECURITY INCOME PROGRAM As described by Daly and Burkhauser (forthcoming), the SSI program is a federal program which pays cash benefits to low-income individuals who are 65 or older, or who are blind or disabled. It was enacted in 1972 and was to a large extent a product of the proposals by the Nixon administration for a negative income tax. Eligibility requires not only low income and assets but also, for the blind and disabled, a medical test. The test is most complex for disabled adults and involves a multistep process meant to ensure that the individual is incapable of working, for the goal of the program is to serve only those who are totally disabled. Assessments by medical examiners as well as more general determinations of the nature and severity of the disability and capacity for employment are conducted. An earnings test is also used, which requires that applicants earn less than a fixed dollar amount. All in all, about 63 percent of the applicants are denied by this process. Eligibffity determination for children is different because the employment test is inappropriate; it is instead based on the presence of a severe functional limitation. In 1990, a court decision (the Zebley decision) required that children also be given a particular additional functional assessment test which, when later implemented, effectively lowered eligibility standards by allowing children onto the rolls who did not pass the more formal medical tests. In the same 1996 legislation that restructured the AFDC program, Congress narrowed the basis for SSI child eligibility and moved it back towards the pre-199O standard in breadth. The legislation also denied SSI eligibility for noncitizens. The SSI program has work incentive features that reflect its origins in discussions of a negative income tax. After eligibility has been established and individuals begin receiving benefits, earnings (after disregards) are reduced by only 50 cents for every extra dollar of income, thus providing some incentives to work. However, despite these incentives, the percentage of SSI recipients with earnings has always been very low. Only 4.4 percent of the caseload had earnings in 1996, and the proportion had never exceeded 4.7 percent in the history of the program. In
addition, special incentives allowing working beneficiaries to retain Medicaid coverage after their incomes exceed normal eligibility levels
Economic Effects of Means-Tested Transfers in the U.s.
11
have been taken up by only 1.3 percent of the caseload. Consequently, work incentives are still a major issue in the program, and mechanisms for increasing work are still under active discussion. Caseload and expenditure growth in the program has been positive since its inception in 1974 but was exceptionally high in the early 1990s, as noted previously. This growth was disproportionately concentrated among the blind and disabled, children, and non-citizens (rather than the elderly). Growth in the number of recipients who qualified on the basis of mental impairments was particularly strong. Since the 1996 welfare legislation, the child and non-citizen caseloads have declined. Nevertheless, despite the strong caseload growth in the 1990s, there has been continued concern that many eligibles do not participate in the program. Estimated participation rates among the eligible elderly population, for example, range from 45 to 60 percent. Research on the SSI program has focused on a number of issues. One of the most heavily studied focuses on the reasons for the caseload growth. Much of this research has examined historical fluctuations in applications and awards for the nonelderly. These fluctuations, shown in Figure 2, have been very large. Increases in applications in the late 1970s, subsequent declines in the early 1980s, and revived growth in the mid-1980s can be explained largely by administrative changes in screening stringency over the period, as perceived by the eligible population. The rise in applications in the late 1980s and early 1990s and a portion of the decline after 1994 have been shown to be heavily affected by the business cycle. This is an important finding, because it establishes that labor-market participation is a realistic alternative for many disabled persons, contrary to the
notion that only those incapable of working are in the program. The business cycle is also responsible for much of the very large increase in applications in the early 1990s, but only for some eligibility groups. Caseload growth among the disabled with mental health and musculoskeletal (e.g., back pain) conditions, and among children, were equally the result of relaxation of screening and eligibility rules. Declines in applications subsequent to 1996 can also be partly attributed to the 1996 federal legislation as well as the business cycle.
Another factor in the growth of the child caseload identified in the research is the relationship between SSI and AFDC benefit levels. For AFDC families with children who can qualify for SSI, the greater benefit levels in SSI than in AFDC provide an incentive to move children from the latter program to the former. The gap between benefits has also been growing over time. Research has shown that this has made some contribution to the growth in the child SSI caseload. In addition, related research has shown that work disincentives for single mothers accompany
Moffitt
12
FIGURE 2. SSI Applications and Awards among Population Aged 18-64 Apptcations 10
9 8 .2
0.
C
-
Awards
0.
a a
2
0
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
Year
Source: Daly and Burkhauser (forthcoming).
this shift, as the availability of SSI benefits allows single mothers to participate less in the labor market than they otherwise would. Some research has also been conducted on exploring reasons for the low participation rates of SSI-eligible elderly. Lack of information about the program does not seem to be a factor, but financial need does, for many nonapplicants have alternative sources of income. Nevertheless, the research on this subject has failed to clarify sufficiently why so many eligibles fail to apply for the program. Research on the work incentives of the SSI program has yielded rather discouraging results to date for the disabled. Disabled SSI recipients appear to be relatively unresponsive to financial incentives, and experimental tests of programs which offer financial and other incentives to
undergo additional training or vocational rehabilitation have experienced very small takeup. In addition, as noted previously, very few recipients take advantage of the less than one-for-one benefit reduction rate in the program. A major and continuing policy challenge in the program is the search for mechanisms to encourage and allow disabled recipients to fulfill their employment potential.
4. THE EARNED INCOME TAX CREDIT The EITC, as noted by Hotz and Scholz (forthcoming), has been one of the fastest-growing means-tested tax programs in the country. Its popularity
Economic Effects of Means-Tested Transfers in the U.s.
13
stems from its emphasis on rewarding families that have significant levels
of employment and earnings. The program provides a refundable tax credit, which can be as high as $3,800 a year (1999), to families with earnings. A small credit to childless families is also available. The program was introduced into the tax code in 1975, but did not see significant expansion in generosity until the 1980s, when the size of the subsidy was increased and then indexed to inflation. Tax bills in 1990 and 1993 increased the amount of subsidy greatly and have led to the sizable growth in expenditures in the 1990s which was noted earlier. The subsidy is obtained by filing a tax return and reporting the number of qualifying children in the household and the earnings of the father and mother. The size of the tax credit is proportional to earnings up to some maximum level, and then is phased out as earnings increase. In 1999, for example, a two-child family could receive a credit equal to 40 percent of their earnings up to an annual earnings level of $9,540, and the credit was phased out at a 21-percent rate until the credit fell to zero, which occurred at an earnings level of $30,580. Thus families fairly high in the income distribution were eligible for benefits. Fourteen states and the District of Columbia have state EITCs which provide for further tax credits. An issue in the administration of the EITC has been overpayment of subsidies, which in 1995 were estimated to be 25 percent of tax expenditures. Most of these result from inaccuracies in the claim for qualifying
children. While the overpayments are high, it is often noted that
17
percent of taxes are not paid to the IRS overall and more than 25 percent of taxes are not paid for some forms of capital income and income from the informal sector. In addition, despite the overpayments in the EITC, participation rates of EITC eligibles appear to be less than 100 percent, sometimes much less so. For example, participation rates among eligible
single mothers, who historically have low tax filing rates, have been estimated to be in the range of 42 to 54 percent. Much of the research on the EITC has concerned its effects on work incentives, since this is one of the main appeals of the program. While the EITC should increase labor force participation, 77 percent of families eligible for the credit fall into the flat or phaseout region of the credit, where there are more likely to be work disincentives than work incentives. Most studies have indicated that there is a strong and significant
positive effect on the labor-force participation rates of single-mother households. The participation rates of such households have risen markedly over the last decade, and the FITC is one of the leading causes of that increase. At the same time, however, research has suggested that the program may have had a slight negative effect on the employment
14
Moffitt
rates of married women, for many women are married to men who earn sufficiently high wages that additional earnings from the wife fall into the phaseout region of the EITC (that is, the region where additional earnings actually reduce the amount of the credit). In addition, there is
some evidence that, while increasing employment rates overall, the EITC may have lessened the hours of work of men and women in twocareer families. There has also been some concern that the EITC may discourage marriage because men and women in certain earnings ranges can receive a greater EITC sum by not marrying and filing separate returns than by marrying and filing joint returns. The empirical evidence to date, however, suggests little effect of this incentive on actual patterns of marriage. There has also been some research on the advance payment option in the EITC by which recipients can receive their credit over the tax year in question, as they earn wages, rather than in a lump sum at the end of the year or in the following spring. Some observers believe that the work incentives of the EITC would be greater if recipients could see the link between their earnings and the credit more quickly and immediately. Hotz and Scholz point out the high administrative costs of making this option more widespread, however, and describe the potential for noncompliance and fraud which would make monitoring procedures necessary. The advance payment option is used in the United Kingdom in a somewhat different program, but it appears that little monitoring for noncompliance is conducted there.
5. FOOD AND NUTRITION PROGRAMS As discussed by Currie (forthcoming), the Food Stamp program (FSP) is
only one among several programs that support food expenditure and nutrition among low-income families. The FSP is the largest, but also important are the Special Supplemental Nutrition Program for Women,
Infants, and Children (WIC), the National School Lunch Program (NLSP), and the School Breakfast Program (SBP). Expenditures on the latter three
programs are over 50 percent of those of the FSP, thus constituting a
sizable additional amount of spending. All of these programs are federally financed and uniform across the states. Also, there are many other
smaller programs that support food and nutrition in the U.S., covering smaller and more specialized populations. The FSP provides food assistance to families and individuals, regardless of family structure, who meet income and asset conditions. Families on TANF, SSI, or general assistance are automatically eligible. The nominal tax rate on earnings in the programthe amount by which benefits
Economic Effects of Means-Tested Transfers in the U. S.
15
are reduced for each dollar increase in incomeis 30 percent, but this rate is affected by the presence of deductions and exemptions. Benefits have historically been paid by the issuance of paper coupons but recently have been increasingly paid by electronic transfers using debit cards. This change has been thought to reduce the incidence of fraudulent selling of paper coupons, although the extent of that activity has never been accurately determined. The FSP began as a small pilot program in 1961 and was gradually expanded over time, finally being mandated for all counties in 1974. It was indexed to inflation in the 1970s, thus preventing the decline in real benefit amounts experienced in the AFDC program. The program was largely untouched by the 1996 federal welfare legislation that restructured AFDC, although work requirements and eligibility for certain categories of single adults as well as immigrants were restricted. The WIC, NSLP, and SBP are quite different. WIC provides financial assistance for the purchase of nutritious foods, nutrition education, and access to health services for pregnant or lactating women and children under 5. It is thus aimed specifically at improving nutrition among women and young children. Eligibility requires not only low income and assets but also that the women and children be at "nutritional risk," such
as having inadequate or inappropriate nutritional intakes, specific nutrition-related health deficiencies, large weight-for-height, or a number of other measures that are set by the states. The NLSP and SBP allow children in low-income families to receive reduced-price or free school lunches or breakfasts. They are thus like the FSP in subsidizing food expenditure per se but like the WIC in having a specific target population. In addition, the meals provided to the children must meet USDA nutritional guidelines, although there has been some concern recently that the meals remain high in fat and low in certain nutrients. The NLSP
is the far larger program of the two, having almost five times larger expenditure than the SBP. Caseloads in the FSP rose in the late 1980s and early 1990s, but have fallen since the enactment of 1996 welfare reform legislation. The data show that this is partly the result of an improving economy as well, but partly a result of the decline in AFDC-TANF caseloads, for participation rates among eligibles have also declined. Expenditures and caseloads in the WIC and NSLP programs, on the other hand, having risen in the late 1980s and early 1990s, have continued to rise in the mid-1990s, albeit at a slower rate. There has been a considerable amount of research on the FSP, WIC, and NLSP programs. One area of research has focused on the effects of these programs on food expenditures, nutrient availability, and nutrient
16
Moffitt
intake. Research indicates unequivocally that the FSP increases food expenditures, although not dollar for dollar, implying that recipients re-
duce some food expenditure out of their own income and spend it on other goods. It also appears that the program increases nutrient availabilitythat is, the nutritional content of the foods purchased or brought into the homebut the evidence on nutritional intake (i.e., taking account of wastage and food eaten away from home) is much weaker. Evidence on the WIC program generally indicates favorable effects on child birth weight but also that the program tends to discourage breast feeding, which is generally preferable to using infant formula. The latter effect arises because the WIC program gives free formula to participating mothers. The effect of WIC on infant outcomes is more variable, but the evidence does indicate increases in nutrient consumption and reductions in the incidence of anemia. Research on the NLSP indicates that it improves nutrient intake.
There have also been a number of studies on the determinants of participation rates in the FSP, for such rates are generally in the range of 60 percent, and thus not all eligibles are in the program. The research indicates that three factors are important in explaining nonparticipation in the program: lack of information about eligibility for the program, transactions costs which make participation onerous, and the stigma of being a welfare recipient. Research on the WIC program indicates that administrative barriers to participation are an important factor in explain-
ing lack of takeup. In the NLSP, an additional factor is the nutrient content of the lunches offered, for it appears that the higher the nutrient content, the less likely students are to participate. Steps to make the school lunches both nutritious and appealing to students have been discussed. Two other areas of research on these programs concern whether the
FSP should be cashed outthat is, whether cash should be provided to recipients instead of food couponsand whether the programs have a negative effect on work incentives. The first of these issues is motivated in large part by the rather low levels of FSP coupons relative to private food expenditures of the poor, suggesting that the coupons simply sub-
stitute for private food spending and hence are no different than cash welfare to the recipient. Interestingly, both econometric evidence and evidence from cashout demonstrations indicate that Food Stamp coupons have a greater effect on food expenditures than does cash, creating a puzzle that has not been adequately explained. The second of these questions concerns the traditional issue of whether a welfare program such as the FSP, which provides assistance even to those who do not work, has work disincentives. The several studies on this issue show
Economic Effects of Means-Tested Transfers in the U.S.
17
relatively little labor-supply response to the program, perhaps because its benefits are small relative to other forms of income received by the household.
6. The AFDCTANF Program Despite its decline, in terms of caseloads and expenditures, to a point where it is only the fifth-largest means-tested program in the country, the AFDCTANF program continues to receive the most attention from policymakers, the general public, and researchers. In his review of past and current research developments in the program, Moffitt (forthcoming) charts its growth and decline over the last three decades and reviews the research conducted on it. The program was created by the 1935 Social Security Act, was targeted at low-income children living with only one biological parent, and was intended to support widows with children. The caseload grew slowly through the 1950s and then accelerated in the 1960s and early 1970s. Subsequent to the 1970s, benefit levels in the program declined in real terms and an emphasis on work requirements steadily grew. The 1988 Family Support Act mandated employment programs in all states but required a human-capital, education-and-training emphasis to be part of the program mix. But the 1996 Personal Responsibility and Reconciliation Act (PRWORA) changed the program in more
fundamental ways, by devolving the responsibility of major program design elements as well as financing to the individual states, imposing strict work requirements in order to qualify for federal aid, and imposing lifetime limits on the number of years of benefit receipt which could be paid to a parent out of federal funds.
Table 3 shows the major elements of the 1996 Act and how they changed the program. The legislation converted the previous matching grant to a block grant and removed much of the federal regulatory authority over the design of the program, leaving the states free to set the benefit level, tax rate, income limits, asset requirements, and even the form of assistance (cash or in-kind services). In addition, no federal definition of who is to be included in the assistance unit is imposed; states can cover two-parent families at their own discretion, for example. The entitlement nature of the program is abolished, and states are not required to serve all eligibles. At the same time, however, the law imposed new federal authority in a few specified areas. Federal funds are not to be used to pay adults for more than 60 months of TANF benefits over their lifetimes (although states are allowed an exemption from this requirement for 20 percent of their caseloads), and new work requirements are imposed which require that states engage much greater
18
Moffitt
TABLE 3
Comparison of the AFDC and TANF Programs TANF
AFDC
Characteristic Financing
Matching grant
Block grant
Eligibffity
Children deprived of support of one parent or children in low-income twoparent families (AFDC-
Children in low-income families as designated by state; AFDC-UP abolished. Minor mothers must live with parents; minor mothers must also attend school
UP)
Immigrants
Illegal aliens ineligible
Aliens ineligible for five
Form of aid
Almost exclusively cash
payment
States free to use funds for services and noncash benefits
Benefit levels
At state option
Same
Entitlement status
Federal government required to pay matched share of all recipients
No individual entitlement
Income limits
Family income cannot exceed gross income limits
No provision
Asset limits
Federal limits
No provision
Treatment of earnings disregards
After 4 months of work, only a lump-sum $90 deduction plus child care expenses; nothing after 12 months
No provision
Time limits
None
Federal funds cannot be used for payments to adults for more than 60 months lifetime (20 percent of caseload exempt)
JOBS program
States must offer a pro-
JOBS program abolished
gram that meets federal law
years after entry and longer at state option
Economic Effects of Means-Tested Transfers in the U.s.
Work requirements
Parents without a child under 3 required to participate in JOBS
19
Exemptions from work requirements are narrowed, and types of qualified activities are narrowed and prespecified (generally excludes education and classroom training) and must be 20 h/week rise to 30 h/week for single mothers
Work-requirement participation requirements
JOBS participation requirements
Child care
Guaranteed for all JOBS participants
No guarantee, but states are given increased child care funds
Sanctions
General provisions
Specific provisions mandating sanctions for fail-
Participation for work requirements rises to 50% by FY 2002
ure to comply with work requirements, child support enforcement, schooling attendance, and other activities
Child support
States required to allow first $50 of child support received by mother to not reduce benefit
No provision
Source: Burke (1996).
fractions of the caseload and which exempt many fewer families (as many as 50 percent of single-mother recipients and 90 percent of twoparent families must comply). Recipients involved in general education and training cannot be counted toward these participation requirements; most activities require direct work. The most important new features are the time limits and work requirements. Lifetime time limits are a new concept in U.S. transfer programs and are based on a quite different philosophy of the aims of public assistance than has been the case heretofore, namely, that families are only entitled to temporary assistance. States have embraced time limits with vigor: half of them have chosen to adopt time limits even shorter than the federal five-year maximum. The work requirements in the new legislation are much stronger than in previous law and change its orientation
20 Moffitt
from education and training to work per Se. Indeed, most states have
adopted a work-first approach in which recipients and new applicants for benefits are moved as quickly as possible into work of any kind, with a deemphasis on education and training. The law also allows states to impose sanctions on recipients for failure to comply with the work requirements, sanctions which are much stronger than in past law and which have been actively enforced. With the aim of reinforcing the effect of these work requirements on employment, states have generally lowered their tax rates to encourage work as well, a feature that historically has been strongly supported by economists who believe they will provide work incentives. Another new goal of welfare programs in the 1990s has been to reduce the rate of nonmarital childbearing and to encourage marriage. Although there were few provisions of the PRWORA legislation that were directly aimed at these family-structure outcomes, the provisions aimed at reducing the amount of government assistance and encouraging women to sustain themselves off welfare were thought to implicitly encourage marriage and discourage nonmarital childbearing. As shown previously in Table 1 and as discussed earlier, caseloads in the AFDCTANF program have been falling for several years, and real expenditures have been declining. The per capita TANF caseload is now below what it was in 1970. The decline began prior to 1996 but accelerated thereafter. Expenditures have also changed in composition, as a smaller fraction is devoted to traditional cash expenditures and a larger fraction is devoted to noncash expenditures on services such as child care and other social services, reflecting a preference by states to support families in those ways. Real benefits also fell from the 1970s until the mid-1990s, when the decline abated and benefit amounts leveled off. It is important to note that the recent reforms have contracted the program in many ways, but reductions in benefit amount have not been one of them. The characteristics of AFDCTANF recipients have changed in some ways over time, but not dramatically. The one major change has been a shift in the types of single mothers on the rolls. Whereas the program began with a caseload composed mostly of widows, it shifted in the 1960s and 1970s to one composed mostly of divorced women. It then shifted again in the 1980s to one composed mostly of never married single mothers who have had children out of wedlock. These trends partially parallel larger trends in the society. They may also partially explain the decline in popularity of the program among voters. Economic research on the AFDC and TANF programs has been large in volume. The most heavily researched issue is the effect of the program on
Economic Effects of Means-Tested Transfers in the U. S.
21
labor supply and work effort. The research on this issue indicates that transfer programs like AFDC which provide open-ended support reduce work effort and that providing unrestricted benefits to those who do not work has work disincentives. At the same time, research on the effect of
reducing tax rates on recipientsby increasing earnings disregards
shows it to have a much smaller positive effect on overall labor supply than expected, because tax rate reduction has the offsetting effect of drawing additional women onto the rolls and inducing them to decrease their work effort. One way to reduce this offset is to provide additional work incentives to those off welfare as well as on, that is, to provide a more universal work subsidy (to low-income families) that is not tied to welfare.
The EITC is one program of this kind. Other programs that have been proposed allow women to "take" their subsidy off the rolls and to continue to obtain earnings supplements afterward. Under TANF, most states have increased their earnings disregards but the existing evidence therefore suggests that this will have little effect on overall work effort. However, the new work requirements are more likely to have a positive effect. Theoretical research on work requirements strongly suggest they wifi increase work effort, as should be expected, with a possible cost if some recipients who need assistance cannot comply and leave the welfare rolls. However, work requirements require that the welfare system be converted to a categorical program which divides recipients into those who can and cannot work, and imposes the work requirements only on the former. Dividing the caseload up in such a way is difficult and opens the door to possible inequitable treatment as borderline cases are assigned to one group or another and either gain or lose as a consequence. The theoretical literature in this area cannot answer the practical question of how well categorization can be implemented but it does highlight the tradeoff between better targeting of assistancethat is, providing work requirements to some recipients but benefits without requirements to othersand the costs of making that separation. Another area of economic research has focused on the relative merits of the work-first approach embodied in TANF and more education, human-capital oriented approaches. There have been a long series of evaluations of different types of employment and training programs in the AFDC program which, overall, show that modest positive effects on earnings can be achieved with relatively low-cost job search and job assistance programs even for the very unskilled population that the program covers. Typical gains, for example, are in the range of $300-600 per year. Research to date on work-first programs indicates that they have a more immediate effect on employment and earnings than human
22
Moffitt
capital programs, but one which fades out over time. Human capital programs appear to have effects that are more long-lasting. No clear winner emerges in this comparison, and many researchers have taken the rather different tack of investigating whether different programs might be tailored to different individuals, commonly called a mixed strategies approach, in which some recipients are deemed sufficiently
job-ready that a work-first approach is best while others are seen to be in more need of basic skills training. The other major feature of TANF, time limits, has also been the subject of considerable research attention. Researchers have noted that, while time limits should eventually force recipients off the welfare rolls with consequent increases in employment and earnings, recipients may also take action to leave the rolls early in order to bank their benefits for a later time when they are most in need of them. Indeed, the most important development thus far is how few recipients have hit time limits. The
massive reduction in the caseload, whether it has been the result of a good economy or of welfare reform, and regardless of whether some of it has been the result of banking behavior, has had the result that many families have not used up their years of eligibility even though five years have passed since 1996. However, this may change as the economy slows down and as states with five-year time limits experience more families hitting those limits. A large volume of research has been devoted to estimating the overall effects of the 1996 PRWORA legislation and of the creation of the TANF program. Descriptive evidence, for example, reveals that employment rates of single mothers have increased and that the incomes of all but a
small lower tail of the distribution have risen since the legislation. A sizable body of research has sought to disentangle the effects of the economy from those of welfare reform in explaining these trends. The majority of results from this literature indicate that the law has indeed had a significant and large additional effect beyond that of the strong economy. Prima facie evidence for this view is that, after 1996, the decline in the national unemployment rate slowed but the decline in the AFDCTANF caseload accelerated.
Another area of research has focused on the effects of welfare benefits on family structure and other demographic outcomes. Research on the
effect of AFDC on these outcomes is quite large in volume and has gradually moved, over the past decade, to a consensus that there are some non-zero effects of this kindthat is, that variation in AFDC benefits across states, received primarily by single mothers and not twoparent families, tends to be positively correlated with the rate of single motherhood. Research into the causes of the time-series increase in sin-
Economic Effects of Means-Tested Transfers in the U.S.
23
gle motherhood in the U.S., on the other hand, suggests that that increase is primarily the result of deeper economic and social forces such as the rise in female job opportunities and the decline in unskilled male wages. The TANF program, while having few direct provisions relating to marriage or fertility (aside from family caps), nevertheless was intended to have a positive effect on marriage and a negative effect on nonmarital fertility. The research evidence to date, however, is mixed at best in its results. There is little sign in the data of a strong effect of welfare reform per se on these outcomes, again, perhaps, because they are so driven by larger social, cultural, and economic forces. While the research evidence on the overall effects of welfare reform is
by now reasonably large in volume and has yielded important new findings, most of the research on the effects of detailed individual provi-
sions of TANFtime limits, work requirements, and so onhas unfortunately foundered on difficulties of evaluating their effects. By and large, researchers have not been successful in using the variation in programs across states to isolate the independent effects of these individ-
ual components of reform, and to estimate how much of the overall effect would have occurred if all elements of reform had been enacted except each of these components, in turn. The cross-state variation under TANF is sufficiently great, and the types of program variation so complex, that the effects of the components per se have not thus far been sufficiently isolated. Random-assignment evaluations could in principle do better, for they could be designed to alter each component while holding the others fixed, but they have not been designed in that way thus far. Finally, there has been considerable research attention paid to the effects of the block grant system put in place by PRWORA. The conventional view based on existing research on the effects of matching grants, which were used for AFDC, and block grants, which are in place for TANF, is that they have different price effects because the latter does not subsidize state expenditures above the block grant at all, whereas the former does. This should curtail spending over the block grant amount. Theoretical research has also shown the possibility of a "race to the bottom," as states facing a high price of expenditures lower their benefits to avoid immigration of the poor from other states, causing a cascading series of benefit reductions by all the states. However, to date none of these effects have occurred because the block grant allocations made to the states are generally much in excess of what states are spending, primarily because of the marked decline in the caseload and consequent reduction in state spending. Ascertaining whether the block grant structure wifi lead to restricted state spending or to benefit reductions around
24 Moffitt
the country will have to await a period when welfare spending rises up to the block grant level, where it wifi become binding.
7. HOUSING PROGRAMS The set of housing programs for low-income families in the U.S. consti-
tute a complex mix of programs with different features. As noted by Olsen (forthcoming), new programs have been added to the system over time and the rules of existing programs have changed frequently. These programs are much more expensive than commonly realized because they rely to a much greater extent than other welfare programs on indirect subsidies that do not appear in the records of the administering agency. Programs divide up into those that are project-based (owned either by
the government or by private contractors who are subsidized by the government) and those which are tenant-based, in which eligible families receive subsidies to defray the rent in private housing. The public housing program, begun in the 1930s, is the best-known project-based program; it offers rental units to low-income families in newly constructed projects owned and operated by the government. Beginning in 1954, the government began in addition to contract with private parties to construct low-income housing or to rehabilitate existing housing for this purpose, in most cases insuring the mortgages of the contractors. The Section 8 New Construction Program established in 1974 is the
largest program of this type. Under this program, the government subsidized the construction costs of privately built housing for low-income families and provided monthly rental payments. In 1983, Congress halted additional commitments under HUD's new construction programs except for small programs for the elderly and disabled. Today the largest housing subsidy program is tenant-based.3 The Housing Choice Voucher Program enacted in 1998 consolidated the two variants of the Section 8 Existing Housing Program that had operated simultaneously for fifteen years. This program pays a portion of the rent of eligible low-income
households that locate housing in the private market that meets the program's minimum housing standards. Although HUD or USDA programs have produced few new units in recent years, the IRS's Low Income Housing Tax Credit, enacted in 1986, will soon become the However, as Olsen discusses, the Low Income Housing Tax Credit, enacted in 1986, pays 70 percent of the development costs of projects for low-income families and has become the second-largest housing program in the country.
Economic Effects of Means-Tested Transfers in the U.s.
25
second-largest housing program in the country and it is growing much more rapidly than any other program. Eligibffity for these housing programs is based on a number of factors, the first being the requirement that adjusted family income fall below certain thresholds determined by family size and the median income in the locality. However, because fixed budgets are authorized for these programs and there is excess demand for subsidies, they must be rationed; that is, housing programs are not entitlements. Local housing authorities and owners of private projects, operating under general guidelines from Congress, determine their preferences in granting assistance to individual families by giving weight to characteristics of the families. Once assistance is granted, families in project-based programs are offered specific units and families in tenant-based programs are authorized to locate eligible units in the private market. A substantial majority of assisted families participate in programs that require them to pay 30 percent of their income toward rent. However, many assisted families pay rent that is independent of income. Research on housing programs has addressed a number of different topics. One concerns the cost-effectiveness of different program types. The studies are unanimous in finding that tenant-based assistance provides housing equal in quality to that of project-based assistance at a much lower total cost. Another key issue is whether the programs indeed increase the housing consumption of their recipientscertainly a main goal of the program, but not one guaranteed to occur, at least for projectbased housing. The literature indicates indeed that housing consumption is raisedthat is, that families occupy higher-quality housing than they would in the absence of the programboth in housing projects and housing occupied by voucher recipients. There appears to be some leakage in the subsidy, for consumption of nonhousing goods rises as well, although this should be expected if part of the goal is to enable families to reduce what are often very high housing expenditures. The housing programs also appear, according to the research, to increase housing consumption more than would a pure cash grant, consistent with the rationale for housing assistance. The Housing Allowance Supply Experiment conducted in the 1970s studied the market effects of an entitlement housing voucher program similar to the limited enrollment Section 8 Voucher Program that operated between 1983 and 1998. The Supply Experiment operated for ten years in two small metropolitan areas with very different initial vacancy rates and minority populations. About 20 percent of the families in the two counties were eligible to receive assistance. Participation rates of
26
Moffitt
eligibles never exceeded 50 percent, partly because subsidies for those with moderately high incomes were not large enough to outweigh the costs of moving and participating in the program. The results showed that an entitlement program of tenant-based assistance would produce a substantial increase in the supply of dwelling units meeting minimum housing standards but would have little effect on rent levels. There have been a few research studies on other topics as well. Some examine the work disincentives of housing programs, on the presumption that, like all welfare programs, the reduction in the subsidy with an increase in earnings wifi reduce the incentive to work. The results show that such work disincentives probably exist but that they are quite small. Another set of studies examine the relative effects of public housing and tenant-based housing on the choice of neighborhood, finding that public housing exacerbates economic and racial segregation while tenant-based subsidies ameliorate them to some extent. Findings from the recent Moving to Opportunity Experiment that offered randomly selected families in public housing vouchers on the condition that they move to neighborhoods with very low poverty rates indicate improvements in the educational attainment of the children involved and reductions in their violent criminal behavior. It also increases the earnings of adults in these families.
8. CHILD CARE PROGRAMS Blau (forthcoming) describes the structure of means-tested child care pro-
grams in the U.S. He notes that such programs have at least three
different goals, not always mutually compatible. One is to increase the rate of employment of low-income women, particularly when operated through the AFDC-TANF program or when aimed at assisting lowincome parents in general to work. A second is to increase the quality of child care for low-income children, and a third is to assist in the development of disadvantaged children through early education programs such as Head Start. These goals may conflict, as they do for programs which encourage low-income women to work through the provision of inexpensive child care.
A variety of programs serve one or more of these goals. One of the largest is the Child Care and Development Fund (CCDF), which provides funds to states to subsidize child care for low-income families and is intended to support employment of low-income parents. It was created in 1996 by Congress and consolidated a number of prior programs, some of which had served primarily the AFDC population and some of which had served the "working poor," meaning low-income families not on AFDC. When requiring such consolidation, Congress also required
Economic Effects of Means-Tested Transfers in the U.s.
27
that minimum percentages of the grant be spent on the AFDCTANFbased population (not only current recipients, but also families who had
recently left AFDC or are at risk of going on) and that minimum percentages be expended on the working poor. A second program, the Title XX Social Services Grant, provides states with funds to expend on a variety of social services for the poor, including child care; states spend approximately 15 percent of their funds on that service. The Dependent Care Tax Credit, a nonrefundable tax credit in the federal income tax, also
provides a subsidy for child care, which declines as income rises. Finally, three programs are intended for early education and child development, and are not tied to parental employment. These include the Head Start and Title I-A programs, which provide early education for disadvantaged children, and the Child Care and Adult Food Program, which provides subsidies for nutritious meals in child care settings for lowincome children. All the programs are federally financed and have uniform national rules, except the CCDF and Title XX Social Services Block
Grant, which give considerable discretion to the states on operation
within federal guidelines. In terms of expenditure on low-income families, the CCDF and Head Start programs are currently the largest at about $5 billion each, followed by the Dependent Care Tax Credit at approximately $3 billion. The number of children served is the largest in the Dependent Care Tax Credit, followed by the CCDF; Head Start is one of the smallest. Expenditures per child are essentially inversely related to size: the Dependent Care Tax Credit gives $720 per child, while Head Start gives $5,759 per child. Eligibility in the three federal early-education programs is related to various measures of low income and is nationwide, while eligibility in the CCDF and Title XX Grant is set by the states within federal guidelines. Title XX funds must be spent on children in families with income below entirely state-chosen limits, while CCDF funds must be spent on families with income no greater than 85 percent of state median income. States are free to set a wide variety of subsidy mechanisms in their CCDF-funded programs, with fees that have maximums and minimums or are waived
for certain groups, with vouchers or direct contracts with providers, and with flexibility in setting reimbursement rates for providers. Child
care facilities must meet state licensing and regulatory requirements. Research on the effects of child care programs has been concentrated on a few selected issues. One is whether child care subsidies in general increase the employment of mothers. Based on evidence from demonstrations and random-assignment trials as well as from nonexperimental studies which use variation in child care price to estimate employment responsiveness, the literature strongly indicates that child care subsidies
28
Moffitt
increase employment and hours of work. However, the magnitude of the effect is quite uncertain and varies considerably across studies. In his review, Blau finds that the studies with the best data and which account most realistically for the child-care market find relatively low price elasticities of employment response, but ones that are still statistically significant.
Research on several other issues has been conducted as well. One
study of the effect of price on the quality of care chosen by parents using formal day care centers found that child care subsidies led parents to use more care but care at lower-quality centers, as measured by childstaff ratios and staff training. Another study examined the effect of child-care
subsidies on the probability that a single mother would be on AFDC, and found that such subsidies lower that probability, presumably by allowing AFDC mothers to go off welfare and work. Blau also reviews the large literature on the effect of early childhood education on child outcomes, finding that the evidence supports an effect of such education on some outcomes for some programs. Whether the effects fade out over
time or persist is more controversial, although some studies do show persistent effects.
9. EMPLOYMENT AND TRAINING PROGRAMS As discussed by LaLonde (forthcoming), the main omnibus employment and training program in the U.S. at the present time is the Workforce Investment Act (WIA). Passed by Congress in 1998 and taking effect on July 1, 2000, WIA replaced the Job Training Partnership Act (JTPA),
which was the main program for employment and training in the U.S. from 1982 to 1998. WIA provides block grants to the states to fund employment and training programs for adults and youth. WIA contains several titles with different programs and different services. These indude Title I.B.5, which covers adults; Title 1 .B.4, which covers youth; and Title 1 .C, which covers the Job Corps, a high cost training program for disadvantaged youth. Except for the Job Corps, states have great freedom to design their own WIA-funded programs but must meet certain federal requirements. The adult programs are not restricted to low-income persons, but priority must be given to cash welfare recipients. Training programs for youth, on the other hand, both the Job Corps and other youth programs, do require that the recipient have low income and other measures of economic disadvantage. All adults are eligible for job search assistance, but more comprehensive services require that the recipient be unemployed and be unable to find a job or otherwise need intensive services
Economic Effects of Means-Tested Transfers in the U.S
29
to maintain employment. Training is primarily provided through individual training accounts, which allow the individual to choose from a list of
acceptable providers, and thus retains some features of a voucher. A system of "one-stop shopping" is required by the legislation, allowing WIA enrollees to go to only one agency, provider, or location to be directed to all services. There are three generic types of training programs typically provided. One general category is aimed at enhancing skill development, and in-
cludes both classroom training and on-the-job training. A second is "work experience," which involves temporary placement in an actual job. A third is employability development, which includes job search assistance and career counseling. The first is aimed at increasing the individual's long-run labor-market skill level, while the second and third are aimed more at encouraging immediate employment. Typically, an
individual's needs are first assessed, and then he or she is assigned to one of these types of programs. There is no research on the WIA program, because it has been put in place so recently. However, there is a large body of research on JTPA and related training programs, which should still be quite relevant to WIA, given that the basic types of programs are unlikely to change markedly. The majority of the research surveyed by LaLonde is from random-assignment evaluations, where the effects of the training program in question are measured as the difference in outcomesusually earnings and employmentbetween an experimental group and a control group. These experiments typically estimate training effects separately for adult women, adult men, and youth. Adult women are always separated because they include a high fraction of welfare recipients and,
indeed, many of the training programs are explicitly targeted at that group. The findings are quite different by group. For adult women, there is consistent evidence of positive effects of a variety of types of training programs on employment and earnings. The programs include weifareto-work programs tested by individual states on their welfare recipients, high-cost programs for disadvantaged women in general (such as Sup-
ported Work), and the JTPA program. The effects persist for several years and occur for all program types (job search assistance, work experi-
ence, and employability development). For the job search assistance programs, the magnitudes of the effects are modest in sizeranging up to $500 per year, typicallybut are also very modest in cost, leading to very favorable cost-effectiveness ratios. Programs that provide classroom instruction or which add work experience on top of job search assistance have somewhat greater effects. The high-cost programs, such
30
Moffitt
as Supported Work, have even greater effects, ranging up to $1,000 per year. Whether they are cost-effective depends crucially on whether these effects are permanent or fade over time. In the favorable event that they are permanent, even these programs have strong cost-effectiveness ratios (e.g., 15-percent rates of return).
The estimates for adult men and youth are more mixed. For men, most evaluations show little effect on employment and earnings overall. Some programs appear to have positive effects for certain subgroups of men, but the pattern does not have any clear explanation. For youth, effects estimated in the Supported Work experiment as well as JTPA and other training programs are typically very small. However, for youth,
the Job Corps has traditionally been thought of as the main program showing favorable results, based on past evaluations. A new experimental evaluation confirms that its effects on employment and earnings are positive. Four years after enrollment, annual earnings were on the order of 12 percent higher. The effectiveness of the Job Corps in comparison
with some of the other youth programs is thought to arise from the comprehensiveness of its training services as well as the relatively large expenditures on it. However, LaLonde also shows that subgroup analyses present a mixed picture of Job Corps effects, which vary markedly by age and ethnicity. Indeed, for some subgroups (e.g., 16-24 year old Hispanics) the program appears to have no effect. There also is some support in the data for effects being greater for young adults than for teenagers, but even this result is not completely uniform. LaLonde concludes his review by emphasizing the positive findings for adult women and youth. For adult women, low-cost training programs have a fairly large effect relative to cost and constitute what ap-
pears to be a worthwhile investment. He notes that the cost of these programs is far less than the cost of a year of formal schooling, for example, and should not be expected to have dramatic impact as a result. Higher-cost programs may be cost-effective as well, but this depends on the size of their long-term effect, about which little is known. For youth, it appears that only high-cost comprehensive training programs are likely to be productive social investments. The U.S. spends far
less than other countries on training programs, and this evidence suggests that a greater expenditure in that direction could increase the earnings of many groups in the disadvantaged population.
10. CHILD SUPPORT The child support system in the U.S., while not formally a means-tested program or a public transfer program at all, nevertheless plays an impor-
Economic Effects of Means-Tested Transfers in the U.S.
31
tant role in policy discussions on transfers to the low-income population and to single mothers in particular. Lerman and Sorensen (forthcoming) note that the Child Support Enforcement (CSE) system, the governmental program aimed at enforcing private child support obligations, is concentrated on the low-income population. In their chapter, Lerman and
Sorensen review the structure of the present system and the research that has been conducted on it. The CSE program was established by Congress in 1975 to provide matching funds to states to collect child-support obligations, establish paternity, and obtain support awards. States were required to provide child-support enforcement services to AFDC recipients and to any nonAFDC family that requested them. The statute also required that AFDC recipients assign their child-support rights to the statethat is, that any child support payments they received be taken by the state and used to compensate for the AFDC benefitand to cooperate in establishing paternity and securing support. Thus reducing welfare costs and increasing child support were both goals of the system, goals that have remained to the current time. Through legislation, Congress has steadily increased pressure on the states to strengthen the child-support enforcement system in many ways since 1975. In 1988 it set numerical goals for the states to establish paternity for children, and later required that states establish voluntary pater-
nity acknowledgement procedures in hospitals. In 1984 and again in 1988, Congress increased pressure on states to require judges to adhere to state child-support guidelines governing the setting of child-support awards, which are generally tied to the income of the noncustodial parent. This was aimed at preventing judges from setting child-support awards that were too low. Over the 1980s, Congress also increased requirements on states to use wage withholding to obtain payments from non-custodial parents, and in 1996 went further by requiring that every new hire be reported to the CSE agency in order to locate such noncustodial parents who were delinquent in their payments and had not
been locatable by the agency. The fraction of low-income custodial mothers who receive any child support at all was only 24 percent in 1997, and of those that receive child support, even fewer receive the full amount that has been awarded by the court. These low figures, despite the years of increased stringency of child-support enforcement, attest to the difficulty of the problem. The
fraction receiving any support is, however, larger than it was twenty years ago, when it was only 17 percent. The increase has arisen from a greater percentage of poor custodial mothers who actually have an award, which is no doubt partly a result of governmental efforts at
32
Moffitt
establishing paternity and encouraging awards. The increase would have been larger had it not been for a decline in the fraction of mothers who actually received anything even if they had an award. Part of the reason for this decline, though not all, has been a shift in the composition of poor custodial mothers from those who are divorced or separated to those who have never been married; the latter have always received less support than the former. The fraction of single mothers on AFDC receiving child support is approximately 17 percent, even lower than that of all poor custodial mothers. Research on child-support issues has had several purposes. One is determining the incomes of poor noncustodial fathers in order to determine how much they are capable of paying. This is a difficult task, because there is no ready data set to identify noncustodial fathers and their incomes, so most estimation is indirect. Estimates indicate that, overall, noncustodials fathers could pay 3 to 4 times more than they are actually paying, given their incomes and given customary guidelines for how child-support awards are based on income. However, no estimates are available for low-income fathers alone. Evidence from ethnographic
studies indicates that poor noncustodial fathers have high rates of
nonemployment, low levels of education, little work experience, and poor health, and often have criminal histories and unstable housing arrangements. Another area of research focuses on the effect of child-support collections on AFDC participation and on the work effort of welfare mothers. Because states collect most of the child support received by women on AFDC, an increase in child support paid by the noncustodial father has no impact on a woman's income while on welfare, but it increases income off welfare. This should therefore decrease AFDC participation and increase the labor supply. Although the evidence is not as strong as it could be, it does suggest that this is the case. Increases in CSE reduce AFDC caseloads, according to the evidence, and increases in child support reduce rates of AFDC participation and increase employment rates. However, there is also some evidence that increases in child support reduce the work effort of custodial mothers not on AFDC, for in this case the extra income allows them to reduce their hours of work or work effort overall.
Research in this area suggests that, in principle, child-support payments and CSE in particular might reduce the work effort of noncustodial fathers, as they are required to pay a percentage of their income toward support. However, the little empirical evidence available indicates little response of this kind. This may be because noncustodial
Economic Effects of Means-Tested Transfers in the U.S.
33
fathers have inelastic labor supply curves, but it may also be because only in a minority of cases do courts update award amounts as incomes of fathers change. Typically, award amounts are set in relation to income at the time of the initial court judgement, but no adjustments are made thereafter. Nevertheless, ethnographic evidence does suggest that child-support enforcement tends to drive many men into the underground economy, where income is not reported. Indeed, much of the research discussion of the incentives faced by noncustodial fathers focuses on the lack of incentives to pay child support given the fact that all payments go to the government instead of to the children if the mother is on welfare. An additional problem is that many men have accumulated large amounts of child-support debt, which are very difficult to work off. Attention has also been paid to the effects of child-support payments on marriage, divorce, remarriage, and nonmarital childbearing. The predictions of the effect are in most cases ambiguous, because, while increased child support gives men an incentive not to marry, remarry, or have children out of wedlock, it increases the custodial mother's incentives in the other direction by making single motherhood less financially onerous. The little evidence on the issue suggests that there are indeed effects in this direction, with child support appearing to reduce remarriage, nonmarital childbearing, and divorce, but their magnitudes are uncertain. Finally, there has been considerable research on the effectiveness of child-support enforcement policy itself on increasing paternity establishment, award rates, and payment of child support. The evidence sug-
gests that it has had an effect, particularly on the first of these. This is consistent with the time-series evidence mentioned earlier. Thus CSE policy has been shown to have an effect, and for this reason it continues to enjoy strong support as a public policy.
11. CONCLUSIONS Economic research on the effects of the nation's system of means-tested transfers has yielded a large volume of important findings. One of the most basic is the repeated finding that the programs are, by and large,
attaining their central goals of increasing the consumption of lowincome families of medical care, food, housing, child care, and other targeted goods. Another is that there has been an increased redirection of support toward the disabled, both adults and children, both for the receipt of cash support and for medical assistance, and toward needy
34
Moffitt
children off TANF, another worthy goal. A third is that the EITC has been successful in raising the employment rate of low-income single mothers, a long-sought goal of transfer policy in the U.S. At the same time, research has demonstrated that the attainment of other goals of these programs is still a challenge. Designing the transfer programs to provide strong work incentives which are acted on is still an issue in the SSI program, for example, and the EITC has some work disincentives for groups other than single mothers. The AFDC-TANF reforms have been successful in raising employment among single mothers, but the effects on their incomes are less unambiguously positive. The child-support system in the U.S. has made great improvements in support for low-income children, but too little support is still received by low-income mothers, yet the burden on low-income fathers is already onerous by many accounts. Effects of all transfer programs on family structure have become an important topic, but no program has been successful in making a major improvement. Designing reforms to address these and other issues wifi continue to make this a fruitful area of research.
REFERENCES Blau, D. "Child Care Subsidy Programs." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. ted Income: Burke, V. (1993). "Cash and Noncash Benefits for Persons with Eligibifity Rules, Recipient and Expenditure Data, FY 1990-92." Washington: Congressional Research Service. (1996). "New Welfare Law: Comparison of the New Block Grant Program with Aid to Families with Dependent Children." Report 96-72OEPW. Washington: Congressional Research Service.
(1999). "Cash and Noncash Benefits for Persons with Lted Income:
Eligibility Rules, Recipient and Expenditure Data, Fl 1996-Fl 1998." Washington: Congressional Research Services.
Currie, J. "U.S. Food and Nutrition Programs." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Daly, M., and R. Burkhauser. "The Supplemental Security Income Program." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Gruber, J. "Medicaid." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Hotz, V. J., and J. K. Sholz. "The Earned Income Tax Credit." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. LaLonde, R. "Employment and Training Programs." Forthcoming in MeansTested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Economic Effects of Means-Tested Transfers in the U.S.
35
Lerman, R., and E. Sorensen. "Child Support: Interactions Between Private and Public Transfers." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Moffitt, R. (ed.) (forthcoming). Means-Tested Transfer Programs in the U.S. Chicago: University of Chicago Press. "The Temporary Assistance for Needy Families Program." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Olsen, E. "Housing Programs for Low-Income Households." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. U.S. Department of Commerce, Bureau of the Census (2000). Statistical Abstract of the U.S.: 1996. Washington: Government Printing Office.
TAXES AND HEALTH INSURANCE Jonathan Gruber MIT and NBER
EXECUTIVE SUMMARY A common prescription for reducing the number of uninsured is to increase the tax subsidization of health insurance in the U.S. Yet, we already provide over $100 billion per year in tax subsidies to health insurance. This paper provides an assessment of the past and potential impacts of taxation on health insurance coverage and costs. I begin by reviewing the central facts on health insurance and taxation. I then provide a framework for assessing the impacts of tax policies on health insurance coverage and costs, and I review the existing empirical evidence on the key behavioral parameters required to model these impacts. I conclude with the policy implications of these findings for tax policies to expand insurance coverage.
1. INTRODUCTION Uninsurance is one of the worst social problems in the U.S., and it has continued to worsen as the economy has improved throughout the 1980s and the 1990s. In 1987, 14.8 percent of non-elderly Americans were without health insurance. Over the next decade, the non-elderly population without insurance coverage grew by nearly 25 percent, to 18 percent, before falling for the first time in two decades last year. Still, I am grateful to Jim Poterba for comments and to Robin McKnight, Tracey Seslen, and Avi Ebenstein for excellent research assistance.
38
Gruber
despite the recent good news, over 42 million Americans lack health insurance. The problem of the uninsured has been a major focus of policy debate throughout the 1990s. The most prominent example was the proposed national health insurance plan of the Clinton Administration, which was resoundingly defeated in 1994 (Cutler and Gruber, 2001). Tax incentives to expand health insurance coverage have also been considered continually through the last decade, and received particular attention during the presidency of George H. Bush and again in the presidential election campaign of 2000. In recent years, proposals have been made to offer tax
credits to individuals to buy insurance in the non-group market (by
President Bush); to expand those credits to cover the cost to individuals of their group insurance policies (by the Progressive Policy Institute); and to provide credits to firms to induce them to offer insurance (by a number of members of Congress). But the existence of these proposals should not be taken to imply that the U.S. doesn't already dramatically subsidize the provision of insurance in the workplace. In fact, the exclusion of employer (and some employee) health insurance expenditures from the income tax base costs the government over $100 billion in lost revenues annually (Shiels and Hogan, 1999). Indeed, some tax-based proposals would end or limit this exclusion of health insurance from the income tax base, and use the resulting funds to offer tax subsidies to individuals for insurance purchase. Disentangling the costs and benefits of these alternative approaches to tax subsidization of health insurance is difficult, and revolves centrally around a series of behavioral parameters that determine how individuals and firms will respond to changes in the tax treatment of health insurance. An incomplete list of such parameters includes: the price sensitivity of the decision of firms to offer health insurance; the price sensitivity of the takeup decision, conditional on offering, of employees; the price sensitivity of insurance demand among those not offered insurance; the influence of subsidies on the structure of employer-provided insurance plans, such as employee premium sharing. Despite the importance of these issues, however, we have remarkably little evidence on the key behavioral elasticities, and the evidence that does exist is often contradictory. The purpose of this paper is to lay out a framework for researchers and policymakers to think about how tax policies might affect the level and distribution of health insurance coverage in the U.S. I begin by reviewing the key facts that are relevant to thinking about health-insurance policy. I then turn to a discussion of the central parameters that we need to know to fully model both the effect of the existing tax subsidy and the effects of tax-based approaches to increasing insurance coverage in the
Taxes and Health Insurance 39
U.S. I then discuss what we know about these parameters. Finally, I discuss the implications of the facts and our existing knowledge for the design of tax policy toward health insurance in the U.S.
2. THE FACTS There are a number of key facts that must be considered when modeling the effect of taxes on health insurance in the U.S.
2.1 90% of Insurance Coverage Is Employer-Based The predominant source of insurance coverage in the U.S. is employerbased insurance. This is shown in Table 1, from EBRI (2000), which shows the sources of insurance coverage in the U.S. over time. A fairly constant feature of insurance coverage has been that more than 9 in 10 of
those who are privately insured derive their insurance from an employer, generally their own, their spouse's, or their parents'. Why is the employment setting the predominant source of insurance coverage? There are three potential reasons. First, there may be substantial economies of scale in administering insurance which increase the value of pooling mechanisms. Second, the major problem facing providers of insurance is adverse selection, so that insurers are constantly
searching for means of pooling large groups of individuals along dimensions exogenous to health in order to ensure a predictable distribution of medical costs. Workplaces provide just such a pooling mechanism. Finally, the U.S. tax code subsidizes health insurance purchase through the firm relative to the non-group market by excluding the value of that insurance from an individual's income, for both income and payroll tax purposes. This leads to a very large effective subsidy to the cost of health insurance for workers. The result of this subsidy is that there is a lower "tax price" of insurance:
TP -
1 - Tf - T
- T55 - Tmc
1 + T55 + Tmc
where Tf is the federal income tax marginal rate, ; is the state income tax marginal rate, ; is the marginal payroll tax rate for the OASDI program (the 6.2-percent tax rate that is levied equally on employees and employers); and Tmc is the marginal payroll tax rate for the Medicare HI pro-
gram.1 I differentiate the last two programs because, beginning in the
1 The reason that the payroll tax rate is additive in the denominator is that the employer is indifferent between purchasing one dollar of benefits and paying wages of 1/(1 + ;, + Tm,), since each dollar of wages requires a payroll tax payment as well.
TABLE 1
Dependent
35.4 35.0 34.7 33.8 33.5 32.9 30.7 30.9 31.1 31.2 31.5 31.7 32.4
Own name
33.8 33.9 33.9 33.1 32.8 31.8 32.9 32.7 32.7 32.9 32.8 33.1 33.4
Total
69.2 69.0 68.6 67.0 66.3 64.7 63.5 63.6 63.8 64.0 64.2 64.9 65.8
Year
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Employment-based
6.9 6.8 6.7 6.5 6.6
6.1 6.5 7.3 7.1
6.7 6.3 6.6 6.5
mdiv. purchased 13.3 13.3 13.2 14.5 15.5 16.0 16.7 16.9 16.6 16.0 14.8 14.3 14.2
Total
1.7 1.6 1.6 1.8 2.0 2.0 2.0 2.0
12.1 11.0 10.4 10.4
14.8 15.5 15.7 16.1 16.3 17.0 17.3 17.1 17.4 17.7 18.3 18.4 17.5
4.0 3.8
8.6 8.7 8.8 10.2 11.1 11.8 12.7 12.5 12.5 1.4 1.5 1.5 1.6 1.6
3.3 3.3 3.8 3.2 2.9 2.8 2.9 2.7
3.6 3.6 3.5
None
Tncare/ CHAMP VA
Medicaid
Public
Medicare
Proportion of population (%)
Insurance Coverage of Non-elderly Population over Time
Taxes and Health Insurance
41
early 1990s, the taxable maximum for the HI program was increased above that for the OASDI program (and was eventually removed altogether); the marginal rate is zero above the taxable maximum for payroll taxation. For a typical worker in the 15-percent tax bracket, facing a 5-percent state tax rate and a 15.3-percent combined payroll tax rate, this tax price is roughly 0.65; a dollar of health insurance costs 35 cents less than a dollar of other goods purchased with after-tax wages. In addition, there are also tax subsidies available to employees for their spending on employer-provided health insurance, under section 125 of the Internal Revenue Code. Section 125 generally provides that an employee in a cafeteria plan wifi not have an amount included in gross income solely because the employee may choose among two or more benefits consisting of cash and qualified benefits. A qualified benefit generally is any benefit that is excludable from gross income under the
Code, including health insurance, group-term life insurance, 401(k)
plans, child care, and adoption assistance. While employee contributions can therefore be excluded from taxation with a section 125 plan, protection of employee contributions is far from complete. The data on the prevalence of such arrangements are sketchy. The most recent available data, from Kaiser Family Foundation (2000), suggest that half of all workers are in firms that offered such flexible benefit plans. The reason for less than full coverage of this generous tax benefit is unclear, but some of it may have to do with extensive IRS regulation of these arrangements to ensure that they are not abused. For example, the regulations state that no more than 25% of the benefits of a plan can be attributed to any "highly compensated" employee, essentially ruling out the availability of section 125 plans for very small firms. Moreover, there are strict and complicated rules that limit the flexibility of employees to switch sources of insurance coverage during the year if they are paying their health insurance contributions on a pre-tax basis. Perhaps as a result of these inherent advantages of the group insurance market, the non-group insurance market has not provided a very hospitable environment for insurance purchase. Load factors in this market are high, and the generosity of the typical policy is much lower than in the group market (Gruber and Madrian, 1996). A recent study by the Kaiser Family Foundation found that, for those individuals in less than perfect health, it was often not possible to get coverage that was fully comprehensive, the particular illness of the individual often being underwritten out of the policy. Prices were also very variable in this market, making it difficult to effectively anticipate the cost of insuring oneself. There is an existing tax subsidy to the non-group market itself, for a particular group: the self-employed. Beginning in 1986, the self-employed
42
Gruber
were allowed to deduct 25% of their insurance premiums from their taxable income. This share has grown over time and is slated to reach 100% by 2003.
2.2 Employer-Based Coverage Has Been Eroding, with Public Insurance Picking lip Some of the Reduction The notable time trend in Table 1 has been the steady erosion of employer-provided insurance coverage in the post-1987 period. The share of the population with private coverage fell from 69.2 to 63.5 percent in 1993, before rising again slowly to 65.8 percent. This trend was partially offset in 1987-1993 by a sharp rise in the share of the population with Medicaid coverage, due to extensive expansions in that program, particularly for children. But the slow rise in employer coverage after 1993 has also been offset by a sharp decline in public coverage, particularly from 1996 to 1998, which may attribute to an unintended consequence of welfare reform (see Gruber, 2001, for a review of Medicaid program issues). A striking feature of the erosion in employer-provided insurance coverage is that it was not occurring through a decline in employer offering of health insurance, but rather through a decline in employee take-up of that insurance, conditional on it being offered (Cooper and Schone, 1997; Farber and Levy, 1999). A central, and unresolved, mystery is why we saw this decline in take-up. This period was marked by a rise in the
share of health insurance costs borne by employees (Gruber and
McKnight, 2001). But all available evidence, as I wifi review below, suggests that take-up of insurance by employees is not very sensitive to its
price. Indeed, takeup of employer-based insurance in general remains quite high, as I point out below.
2.3 Most of Those Offered Insurance Take It An key fact for designing tax policy towards the uninsured is that most of those who are offered insurance by their employers take it up. This is illustrated in Table 2, which represents tabulations of insurance take-up rates from a 1997 Robert Wood Johnson survey of employers. This table cross-tabulates firm size against average earnings in the firm, and in each cell lists the take-up rate of insurance, which is computed as the number of covered employees divided by the number of employees eligible for coverage. What is striking about this table (in particular in reference to the next table we wifi see) is the high and uniform rate of take-up across cells.
Taxes and Health Insurance 43 TABLE 2
Takeup Rates by Firm Size and Average Earnings Categories Takeup ratea
Average earnings ($) 50,000
1-9 employees
10-24 employees
26-49 employees
50-99 employees
>100 employees
0.81 0.84 0.88 0.89
0.70 0.83 0.84 0.84
0.70 0.79 0.83 0.89
0.71 0.78
0.76 0.82 0.88 0.88
0.90 0.75
a From author's tabulations of 1997 Robert Wood Johnson Foundation survey of employers.
While take-up does rise somewhat with firm size and with earnings, it is quite high in every cell in this table. If employees are offered insurance, they appear to take it up uniformly at quite high rates.
2.4 Insurance Offering Is Highly Correlated with Firm Size and Average Wage The offering of insurance, on the other hand, does very quite significantly across firms, along both of these dimensions. Table 3 crosstabulates (using somewhat different categories for emphasis) firm size and average earnings again, this time summarizing in each cell the average rate of offering health insurance. Offering of insurance is in fact quite low for the smallest firms, even at high wage levels, and for larger firms at the very lowest wage levels. The correlation with firm size corresponds to the non-tax arguments made above for why firms would offer insurance: economies of scale and predictability of insurance expenditures both rise with firm size. Indeed, the Congressional Research Service (1988) reports that the loading factors on insurance are roughly 35% higher for the smallest than for the largest firms. The correlation with average wages may reflect preferences across firms for insurance offering; but it is notable that when insurance is offered in these low-age firms, employees then take it up at a high rate. This suggest some disconnect between the offering and takeup decisions, which I will return to in the evidence section below. Given these last two pieces of evidence, it should not be surprising the the primary correlate of being uninsured is not being offered insurance. Data from the Current Population Survey data show that over threequarters of the uninsured are not offered health insurance on their jobs.
44
Gruber TABLE 3
Insurance Offer Rates by Firm Size and Average Earnings Categories Offer Ratea
Average earnings ($) 30,000+
1-9 employees
10-24 employees
25-49 employees
50-99 employees
>100 employees
0.24 0.32 0.43 0.50 0.55 0.61
0.45 0.55 0.70 0.77 0.83 0.88
0.63 0.76 0.83 0.86 0.92 0.94
0.81 0.88 0.93 0.95 0.91 0.95
0.95 0.93 0.98 0.97 0.98 0.98
From author's tabulations of 1997 Robert Wood Johnson Foundation survey of employers.
2.5 The Uninsured Are Quite Mixed with the Insured Finally, any solution to address the problem of the uninsured, tax-based or not, must recognize a fundamental conundrum: the uninsured are not
an isolated and easily identified sub-population. This is illustrated in Table 4, also from EBRI (2000), which shows the income distribution of the uninsured. The second column shows the number of uninsured in each income category listed in the first column; the third column shows the percentage of the uninsured in each income category; and the final colunm shows the percentage of that category that is uninsured. The last column is the least surprising: the share of any income group that are uninsured declines with increasing income. What is more striking is the second column: there are many uninsured who are not poor or even near poor. Indeed, almost a quarter of the uninsured live in families with incomes over $50,000 per year. This table highlights the difficulty of targeting programs to cover the majority of the uninsured. To cover the majority, you must go fairly high up in the income distribution. But, as you do so, you enter ranges where only a small fraction of the group is uninsured.
3. HOW DOES TAX POLICY AFFECT INSURANCE COVERAGE? To fully understand how alternative tax policies might affect insurance coverage requires knowledge of a wide variety of key behavioral parameters. In this section, I review the parameters that must be measured to fully assess the range of tax policy effects.
Taxes and Health Insurance 45 TABLE 4
Income Distribution of the Uninsureda Income category ($) 50,000 Total
Number of uninsured (in millions)
Percentage of uninsured in income category
Percentage of income category that is uninsured
4.8 3.3 4.5 4.5 6.8 5.3 3.3 9.7 42.1
11.4 7.8 10.6 10.7
44.3 31.0 34.6 32.0 24.6 19.4 13.9 8.5 17.5
16.1 12.5
7.9 22.9 100
a From EBRI (2000).
Figure 1 summarizes the channels through which tax policies might affect coverage, and the resulting behavioral responses that are key to assessing tax-policy effects. The leftmost part of this chart shows the three possible types of tax subsidies to insurance: subsidies to employers to offer coverage (either reforming the existing subsidy, or offering new subsidies); subsidies to employees to take up coverage (once again, ei-
ther reforming the existing section 125 option, or offering new subsidies); and subsidies to individuals for non-group insurance coverage. Each of these types of tax subsidies has effects on both firms and on individuals, as shown in the middle of the figure. Moreover, the decisions of both firms and individuals feed back to each other. Finally, for sizable tax interventions, there may be effects on the insurance market itself which wifi affect the decisions made by firms and workers. In terms of firm decisionmaking, the key element of response is the decision to offer insurance in response to an increased subsidy to employer coverage, or to drop insurance in response to a reduction in the net subsidy to employer coverage. For increases or reductions in the existing tax subsidy to employer-provided insurance, these responses are likely to be symmetric. But a key unresolved question is the extent to which changes in other subsidies would impact employer decisions. For example, would a 10% subsidy to employees for their expenditures on employer-provided insurance have the same impact on insurance offering as an additional 10% subsidy to employers for their insurance spending? In terms of simple economic theory, the answer is clearly yes. But there may be differences in practice that make these responses asymmetric.
46
Gruber
FIGURE 1. Channels for Effects of Tax Policies on Coverage Tax Subsidy to Employers
Offer/Drop Coverage
Firms
Raise/Lower Contributions Raise/Lower Plan Generosity
Tax Subsidy to Employees
Takeup/Drop Employer Coveragi
Takeup/Drop Nongroup Coverag Takeup/Drop Public Coverage Tax Subsidy to NonGroup
Takeup/Drop Spousal Coverage
Markets
For example, subsidies to employers may be targeted by firm size or some other firm characteristic, but are available to the firm as a whole. But subsidies to employees that are targeted will likely leave some employees in a firm eligible and others not. While the subsidy amounts may be the same on average (e.g., a 50-percent subsidy to the firm or a 100percent subsidy to 50 percent of the workers), in practice the impacts may be quite different, depending on how employee preferences are aggregated in the determination of benefits. Similarly, in terms of economic theory, subsidies to non-group and to group insurance should have similar and opposite effects. But, once again, in practice the reactions of firms may not be symmetric, since workers may not view the (currently) inhospitable non-group market as an effective substitute for the group market. Firms also have other margins of response besides the decision to offer (or not to offer) insurance. One important margin is the decision on how much to contribute to insurance costs. Currently, firms pay about 75 percent of the costs of insurance, but this has fallen dramatically over the past 15 years (Gruber and McKnight, 2001). If higher employee premiums lead employees not to take up their employer coverage, then this shift in costs from employee to employer could be significant. Similarly, employers may react to tax subsidies by reducing the generosity of their insurance plans along a variety of dimensions, such as shifting from fee-
Taxes and Health Insurance
47
for-service to managed care plans, raising patient copayments, or restricting benefit coverage. Indeed, it is this subsidy on the margin to insurance costs that led Martin Feldstein to criticize the tax subsidy to employer-
provided insurance in the early 1970s. He claimed that this subsidy would lead to overly generous insurance coverage, which through moral hazard would then raise spending on medical care, increasing further the demand for insurance coverage, and thereby leading to a spiral of rising medical costs.
Individuals can also respond directly to tax subsidies either to employee purchases of insurance or to non-group subsidies. There are four dimensions along which individuals can respond. The first is to change their take-up of employer-provided insurance, conditional on its being offered. The second is to move into or out of non-group insurance coverage. The third is to move into or out of public insurance coverage. This channel may seem more controversial, but the majority of those made eligible for public insurance over the past 15 years have been eligible as well for employer-provided insurance, so this is a margin of potentially active substitution. Finally, married couples can shift insurance coverage from one spouse's job to the other's as the relative subsidy to one spouse or the other changes. In addition, the decisions of employers and workers can have feedback effects on each other. As employers change their offering of insurance, this will affect employee take-up. Likewise, changes in employerprovided insurance generosity can affect take-up and decisions to move across spouses or into public or non-group coverage. Moving the other direction, when economists model firms' benefit decisions, they do not
think of a firm as a distinct entity, but rather as an aggregation of its workers. So individuals' responses to tax interventions can also feed back to firms' decisionmaking. An important research question, alluded to above, is how worker preferences are aggregated; once again, this aggregation may differ across types of tax subsidies and along the margin of employer response (e.g. offering vs. employee contributions). I discuss the scant evidence on this question below. Finally, tax policy can affect the insurance market directly. So long as the market supply for insurance is upward sloping, any major intervention that increases demand wifi lead to a partially offsetting pre-tax rise in insurance prices. On the other hand, many have argued that the high
and unstable prices in the non-group market reflect the "thinness" of this market, and that a major subsidy to non-group policies which led to more purchase could reduce inefficiency and lower prices in that market. These changes in pretax prices will obviously mitigate or exacerbate any direct response by firms and individuals to tax incentives.
48
Gruber
Thus, there are an enormous number of margins that must potentially be modeled to assess the effect of a full range of tax subsidy options on insurance coverage. In principle, there are many direct channels from tax policy to individual and firm behavior, as well as a large number of potential feedback effects from firms to individuals, from individuals to firms, and from the market to both. Given the impossibility of estimating all of these responses, a key question is the extent to which response symmetry can be called upon to apply behavior from one type of subsidy to another.
4. WHAT DO WE KNOW ABOUT KEY BEHAVIORAL PARAMETERS? There is a large literature devoted to estimating some of the behavioral parameters that were discussed in section 3. In this section, we review that literature. We begin by reviewing what is known about firm behavior, then turn to individual behavior.
4.1 Firm Offering Decisions The behavioral response on which there has been the most work is the elasticity of insurance offering by firms. There have been several approaches to estimating this elasticity. The first approach discussed here is
to use variation in the premiums faced by firms to identify the price sensitivity of their offering decision. Two examples of this work are Feldman, Dowd, Leitz, and Blewett (1997), who use information from 1993 for a sample of small firms in Minnesota to estimate price elasticities of 3.9 (single coverage) to 5.8 (family coverage); and Marquis and Long (1999),
who use data from 1993 for 10 states to estimate a much smaller price elasticity of only 0.14. A key problem with this approach, however, is that one only observes premiums for the firms that do offer insurance, and they must be imputed to firms that do not. Thus, instruments must be found that are correlated with the price of insurance but not firm demand, and previous articles have not used firm characteristics that are likely to meet this criterion (e.g. whether the firm is unionized). The third approach is to use variation in taxation to identify the price elasticity of offering, in essence asking whether those firms with higher tax-related subsidies to insurance purchase are more likely to offer insur-
ance. Leibowitz and Chernew (1992) use variation in tax rates across states to examine the effect of after-tax prices on insurance offering by small firms, as well as using variation in premium quotes across locations obtained from small-group insurers. They separately estimate the
Taxes and Health Insurance
49
response to premiums and subsidies, and obtain an elasticity between 0.8 (premiums) and 2.9 (subsidies). Royalty (1999) also uses crossstate variation in marginal tax rates to estimate an elasticity of firms' insurance offering across all employers at 0.63. Gentry and Peress (1994) study cross-city differences in the average share of workers offered health insurance benefits, as a function of cross-state differences in after-tax prices of insurance. They find that for each percentage-point increase in the price of health insurance, the fraction of blue-collar workers covered by employer-provided insurance declines by 1.8 percentage
point, implying an elasticity of 1.4; however, there is no statistically significant effect for white collar workers. These types of studies have the advantage that differences across cities and states in tax rates should be independent of insurance-offering deci-
sions. But they may not be entirely independent: cities and states with substantial taste for insurance may be the ones that offer the largest tax breaks, which would lead to a strong relationship between price and offering. This criticism is addressed in recent work by Finkelstein (1999), who studies the removal of the large (25%) tax subsidy to supplemental private health insurance in Quebec in 1993, and finds an elasticity of 0.42 to 0.54 for employer offering. But it is somewhat unclear how to apply the elasticity of offering of supplemental insurance for a national health insurance scheme to the decision of U.S. firms to offer full private health insurance plans. A third approach comes from running small-scale subsidy pilot programs for small businesses and evaluating the response of firms to subsi-
dized prices. These pilot programs have the advantage of essentially providing a randomized intervention. Two such pilot programs are evaluated in Helms, Gauthier, and Campion (1992) and in Thorpe et al. (1992). The former study finds a wide variety of price responsiveness across sites, with sites such as Utah offering 40% discounts and seeing only 4% enrollment among uninsured firms (an elasticity of only 0.1) and other sites such as Arizona offering 10% discounts and seeing 411% enrollment (an elasticity of 0.4 to 1.1). The latter finds very weak
response to a program that provided a 50% subsidy to the price of insurance for small firms in New York, with an elasticity of only 0.07 to 0.33. But it is unclear whether the small elasticities estimated here are because of the temporary experimental nature of these subsidies; firms may be reluctant to set up insurance plans based on subsidies that wifi only last for a short time. There could be much larger responses to more permanent changes in the after-tax price of insurance. A final approach is to use responses of firms to hypothetical questions about changes in the price of insurance. Morrisey, Jensen, and Merlock
50
Gruber
(1994) use the response to such hypothetical questions to estimate a price elasticity of insurance offering among small firms of 0.92. But it is unclear whether firms respond in the same way to hypothetical questions as they do when faced with an actual insurance purchase decisions. A final issue with this literature is that, with firm-level data, one does
not observe the characteristics of the employees to which the firm is responding in making its benefit decisions: who is the marginal worker whose preferences determine the firm's decisions? One article which attempts to address both the identification and marginal-worker issues is Gruber and Poterba (1994). They study how the self-employed responded to the Tax Reform Act of 1986, which introduced a subsidy to the insurance purchases of the self-employed. This natural experiment provides exogenous variation in the after-tax price of insurance. More-
over, for the self-employed, there is no issue of deciding who is the marginal worker. They find significant increases in the insurance coverage of the self-employed relative to the employed over this period, with an implied price elasticity as large as 1.8. Unfortunately, however, it is unclear how generalizable these results are to firms, which must aggregate the preferences of all their workers in making benefit decisions. A more recent approach to surmounting these problems is Gruber and
Lettau (2000). In that paper, the authors use data from the Employee Compensation Index (ECI) dataset, which collects information both on firm insurance provisions and on the characteristics of a sample of workers in the firm. The latter feature allows them to directly measure the
distribution of tax rates within the firm, and to assess which tax rate seems to be most central in driving benefit provision decisions. They also introduce a new identification strategy, extending the previous taxbased work to rely on state tax progressivity and changes in state taxes over time, while controlling for mean differences across states that are likely to be correlated with tastes for insurance. They estimate an elastic-
ity of insurance offering of 0.3 to 0.4, towards the lower end of the previous literature. Gruber (2001) recently applied the identification strat-
egy of Gruber and Lettau's paper to data from the Current Population Survey, which gathers data on a random sample of workers but not on the distribution of workers in a firm. He finds a higher elasticity of offering of 0.7 in these data. But the ECI estimates seem more reliable in view of the higher quality of the data.
4.2. Employer Insurance Spending There is also a sizable literature on the effect of after-tax prices on employer insurance spending. Estimates of this elasticity come from three types of studies. The first is time-series evidence on how total spending
Taxes and Health Insurance 51
on employer-provided health insurance responds to changes in federal tax rates, presented in Long and Scott (1982), Vroman and Anderson (1984), and Turner (1987). These studies typically yield estimates of the price elasticity of demand between 0 and 0.5. But the results are hard to interpret, as there are many things changing in the time-series data; for example, the fact that health insurance coverage fell in the 1980s may be the result of declining marginal tax rates, but it may also be the result
of a shift in the job base towards service jobs that are less likely to provide insurance. A second set of studies, including Taylor and Wilensky (1983), Holmer (1984), and Sloan and Adamache (1986), analyze cross-sectional data on individuals or firms and ask whether those with higher tax-related subsidies to insurance purchase spend more on insurance coverage. But a potential problem with these studies is that differences across individu-
als in their tax rates arise in part from differences in the underlying behavior of individuals or firms, such as differences in labor supply, family structure, or the nature of the work force. It is impossible to tell whether differences in observed insurance coverage are due to taxes or these behavioral differences. A wide range of estimates emerge from these studies; Pauly (1986) summarizes the consensus range as 0.2 to
more than 1.0.
The final approach that attempts to overcome the problems inherent in the previous cross-sectional literature examines how demand for insurance responds to plausibly independent legislated tax differences. Woodbury and Hammermesh (1992) analyze all fringe-benefit expenditures around the Tax Reform Act of 1986 in a panel data set of colleges and universities. They conclude that tax reform substantially reduced the demand for fringes, with an estimated elasticity in excess of 2. But this is not focused on health insurance spending per se, so it is difficult to disentangle the impact on health insurance. Gruber and Lettau (2000), in the study described above, also examine the impact of tax subsidies on employer-provided insurance spending, using similar variation across states in their tax systems to Woodbury
and Hamermesh. They also find a quite large elasticity of insurance spending: 0.94.
4.3 Employer Contributions to Health Insurance A particularly important margin of response to tax subsidies is how employers change their contributions to health insurance for employees.
This response would be subsumed in the spending elasticities cited above, but it is important to break it out distinctly due to the potential impact of changing contributions on employee take-up.
52
Gruber
There is only one study of which I am aware on this response. Gruber and McKnight (2001) use the Current Population Survey to study the impact of tax changes on the decisions of employers to pay all of the cost of their health insurance plans. The share of employers paying all those costs fell from 44 percent in 1982 to 28 percent by 1998. They model this variable as a function of a variety of actors, including the tax subsidy to employer-sponsored insurance. They find that the tax subsidy did have a very important impact on employer contribution policy. Their central findings suggest that for each 10-percentage-point reduction in the tax subsidy, the share of employers paying all of the costs of health insurance falls by 1.7 to 3.8 percent.
4.4 Employee Take-up of Employer-Provided Insurance A central parameter for evaluating the effects of tax subsidy policy is the
elasticity of employee take-up of employer-provided insurance if it is offered. As highlighted earlier, take-up among those offered is fairly high for all firm sizes and average firm wage levels. But does take-up respond to the prices charged to employees for insurance? The answer, to date, appears to be no. Two studies have examined how employee take-up responds to the price charged for insurance (Chernew, Frick, and McLaughlin, 1997; Blumberg, Nichols and Banthin 2001). Both papers find that firms that charge more for insurance have no lower take-up of their insurance policies. These papers do run into the possible problem that firm premium decisions are endogenous to employee tastes. The direction of bias here is unclear, and depends on whether firms set low employee premiums when there are tastes for insurance, or when there are not (because of paternalism or to satisfy insurance-company conditions for high employee take-up). Gruber (2001) also investigates this question by assessing whether the existing employer tax subsidy affects coverage, conditional on offering, as it should for those with a section 125 plan (roughly half of employees by the most recent estimates). But I find no such effect. The lack of elasticity of employee take-up is very striking in view of the time-series trends discussed earlier. Over the mid-1980s through the late 1990s, the trend was towards rising employer contributions and falling employee take-up. This is shown in Figure 2, from Gruber and McKnight (2001), which shows the share of workers who have employer-sponsored insurance over time, and the share of workers whose employers pay all of the costs of insurance over time. As those authors note, there is a striking correspondence between these series, and a price elasticity of employee takeup of 0.4 would explain the time-series trend in take-up. But this appears to be well above the best estimates to date of the take-up elastic-
Taxes and Health Insurance
53
FIGURE 2. Group Health Insurance vs. Employer Pays All 0 % with Group I-Il
% whose Employers Pay All I
I
.803
.44
I 0
0
a
.73 -
.28 I
1979
1983
I
1987 year
1991
1995
1998
ity. Thus, the cause of this trend towards declining take-up remains a mystery. On the other hand, these findings are consistent with a growing body of evidence which suggests that it is the decisions that employers make for their workers that are most important in determining worker benefit provision, not active decisions taken by those employees. The most striking example is Madrian and Shea (2001), who find that when a firm moved the default investment option and contribution level for its 401(k) plan, the vast majority of workers did not move from that default, despite it being a worse choice for many of them.
4.5 Substitution between Forms of Insurance Coverage As the lower part of Figure 1 illustrates, a key issue for modeling the
effect of tax policy on individual insurance coverage is the substitutability across different forms of insurance coverage. The margin of substitutability for which there is the best evidence is substitutions between private and public coverage. There is a large literature on crowdout over the past 5 years which examines the question of whether those made eligible for public insurance will drop their private insurance to take it up. The earliest estimates of crowdout suggested it was quite large, with one person losing private insurance for every two gaining public insurance in the late 1980s and early 1990s (Cutler and Gruber, 1996). But
54
Gruber
subsequent estimates suggest that the effect may be more modest, on the order of 10-20 percent; see Dubay (1999) for a review. There is less evidence on the other key margins of substitutability,
between own employer coverage, non-group coverage, and spousal coverage. Gruber (2001) did recently extend the analysis of Gruber and Lettau (2000) to consider the impact of changing the employer-provided insurance tax subsidy on all insurance margins, not just on the margin of insurance offering. We find that a larger subsidy to employerprovided insurance causes a reduction in public insurance, consistent with the crowdout literature. But we find little substitutability with nongroup coverage.
To date, there is little evidence on the substitutability of own and spousal insurance coverage, On net, I find that the effect on offering is almost directly translated to a bottom-line effect on being uninsured, suggesting that employer offering is the key margin of response to existing subsidies.
4.6 Substitution across Spouses There is enormous scope for substitution of health insurance coverage across spouses. For example, of the 28 million male married workers aged 21-64 who are insured on the job, 15 million have wives that are offered health insurance. More than 8 million report that their wives actually take up the coverage on their jobs. Of the 16 million married female workers aged 21-64 who are insured on the job, 12 million have husbands who are offered insurance, and almost 9 million of those husbands take up that offer. With such a large number of spouses who are jointly offered insurance, it would seem that substitution across spouses as policy changes would be a real possibility. There is only one article that investigates this issue (Monheit, Schone, and Tayor, 1999), using 1987 data, and their results suggest that the decision to take double coverage (health insurance coverage through both the husband and wife), conditional on both spouses being offered insurance, is very sensitive to incentives. For example, they find that the odds of having double coverage rise by more than a third if both plans are free to the husband and wife. They also find that double coverage take-up is higher among those families in ifi health, and that it is more common when it results in a more comprehensive set of benefit coverage than does take-up by one spouse alone. This set of findings suggest that substitution across spouses who are offered insurance may be quite fluid as tax policy changes. How much
substitution there would be depends on how universal the policy changes are, and how similar the jobs are that wives and husbands hold.
Taxes and Health Insurance
55
For an income-targeted subsidy, for example, it is possible that a husband who works with many low-income workers wifi lose his offer of health insurance, while his wife who works with many high-income workers wifi not.
4.7 The Big Mystery: Take-up among Those Not Offered At least three-quarters of the uninsured are not offered health insurance. Thus, in focusing on the effects of tax subsidies on non-group insurance, the key question is how price-sensitive this group will be in their take-up decisions. Unfortunately, we have essentially no evidence on this critical question. The one relevant paper is by Marquis and Long (1994), who estimated the demand for non-group insurance coverage among workers not offered employer-sponsored coverage as a function of the area-
specific price of non-group coverage. They estimate an elasticity of non-group coverage of 0.3. The problem with this approach is that the price of insurance reflects not only true price differences in insurance (differences in the load factor, or the premium cost relative to expected claim expenses), but also differences in medical costs and differences in underlying tastes for insurance. Both of these latter two factors will bias
downward any estimate of the effect of area insurance prices on demand. Thus, we are left with only a lower bound on the elasticity of insurance take-up among those not eligible for employer coverage.
4.8 How Are Employee Preferences Aggregated? As noted above, appropriate modeling of the implications of subsidies to employer-provided insurance requires an understanding of the mechanism for aggregating employee preferences in the firm's benefits decisions. The best discussion of this issue in the context of benefit provision
is in Goldstein and Pauly (1976). They conclude that the equilibrium benefit determination could arise in one of two ways. One is through the collective choice of the existing set of workers, through an insider-
outsider or union mechanism. In this case, through standard voting arguments, the benefits chosen wifi reflect the tastes of the median worker. The second is through the choices of employers, whose goal is to minimize their total labor costs, and wifi therefore design their benefits packages to reflect the average preferences of their workers. If there is a perfect Tiebout equilibrium across firms, with workers sorted completely by their tastes for insurance, then the average and median tax prices wifi be everywhere the same and the distinction between these models wifi not be important. However, as Gruber and Lettau (2000) discuss, there remains considerable dispersion between these measures within firms; almost 10 percent of firms have a median
56
Gruber
and a mean tax price which differ by 5 percentage points or more. If there is imperfect sorting, then these models can have very different implications, depending on the difference between the median and the mean. Further complications arise when considering the fact that both mobility and influence within the firm differ across workers. Firms may not consider the average of all workers' preferences in making benefits, but may weight more highly the preferences of either mobile workers or the "most influential" workers in the workplace. Assessing the appropriate model of within-workplace benefit determination is not purely an academic concern. Modeling the implications of tax reform may depend critically on appropriately capturing the structure of how tax prices throughout the firm affect benefit decisions. If, for
example, the median tax price is the only one that matters, then tax reforms such as that in 1993 which raised tax rates only at the top of the distribution wifi have essentially no impact on insurance decisions. But if other movements of the distribution matter, then such reforms may have larger impacts. Gruber and Lettau investigate this issue by drawing on the strength of their ECI data to include several moments of the distribution of tax prices in their insurance demand model. They find that the median tax price explains benefit determination significantly better than the average. But they also find that there is an important additional role for the highestpaid worker in the firm (the worker with the lowest tax price). So the
appropriate voting model in their context appears to be one with a decisive median voter but some extra influence for the highly paid worker. Unfortunately, however, it is not clear if these results, estimated in the context of an unlimited tax deduction, extend to other tax-subsidy structures, such as targeted tax credits. Consider a very generous non-group
credit that pays the full costs of insurance for 49 percent of a firm's
workers, but is zero for those at the median and above. It seems unlikely that the firm would not respond at all to such an outside option. In these types of cases, it may be the mean incentive across all workers which better captures how preferences are aggregated.
4.9 Market Responses Another mystery area is how markets wifi respond to tax interventions. As noted above, tax subsidies could lead to rises or falls in group or nongroup market insurance prices. A particularly critical question is whether, for non-group subsidies of a given value (such as the Bush plan), nongroup plans will emerge that are targeted to that dollar value. Advocates
Taxes and Health Insurance
57
of non-group subsidies point to the availability of very cheap insurance plans over the Internet. But, even if such plans are offered, there is the question ofwhether they wifi be demanded. After all, even in the face of existing low-cost policies, there remain over 42 million uninsured persons. An outstanding mystery in insurance markets is why there is not more demand for low-cost, catastrophic policies, given that they insure the substantial risks we should be most worried about but can reduce costly moral hazard. But given this lack of demand, even the offering of such low-cost policies tied to subsidy amounts may not significantly increase take-up.2
5. POLICY IMPLICATIONS The discussion of section 3 and the evidence presented in section 4 have several clear implications for tax policy towards insurance. Before presenting these implications, it is important to consider the goals of subsidizing insurance through the tax code. Presumably, the most important goal is to raise insurance coverage in the U.S., and in the most costefficient manner. Thus, the discussion below wifi focus on the efficiency of various alternative policies, measured as their cost per person newly insured. But a secondary goal of tax policy may be horizontal equity, or redistribution to those already paying for their insurance without the tax subsidies available to others. This is not a goal that will receive attention in the discussion below, but it is important to recognize that if it is the goal of policy, it may be well served by some of the alternatives I dismiss as "inefficient" below.
5.1 Reforming the Existing Tax Subsidy I first consider the implications of reform of the existing tax subsidy to employer-provided health insurance. One alternative here would be to end this subsidy, perhaps redistributing the dollars to other forms of insurance subsidization. Gruber and Lettau (2000) perform some simulations using their estimates to assess the implications for employer offering and insurance spending. The results are summarized in Table 5, which shows the impact of a several alternative reforms on the rate of insurance offering, the level of insurance spending conditional on offering, and the overall level of spending. The ranges of estimates reflect the 2
Moreover, the low prices of catastrophic policies in today's market may reflect positive selection: such policies wifi only be demanded by the healthy, so that prices can stay low. When subsidies are available that make such policies cheaper, then there may be more demand for them by the sick, which would lead to price rises.
58
Gruber
TABLE 5
Implications of Reform of Existing Tax Subsidy
Reform
Remove all tax subsidies Remove income tax subsidy 10% tax rate cut
Change in
Change in spending among
offeringa
Reduction in employerinsuredb
(%)
(million $)
(%)
14.1
22.3
35.4
50.3%
8.6 0.9
13.6
23.7
32.8%
1.4
2.8
3.7% [$95]
offereda
Total
change in spendinga
[$1231]
[$815]
'From Gruber and Lettau (2000). b Result of applying reductions in preceding column to national total for employer-insured.
alternative models estimated in this paper; to be conservative, in this discussion I focus on the lower-bound estimates.3 This paper finds that there would be very significant implications of reducing the existing tax subsidy. Completely removing the existing subsidy, with respect to all (federal and state) income taxes as well as with respect to the payroll tax, would lower the rate of employer insurance offering by 14%. Assuming (following the results in Gruber, 2001) that all of those dropped by their employer would become uninsured, this implies that 22 million Americans would become uninsured. This would represent a roughly 50% rise in the number of uninsured. Moreover, this paper also finds a very large reduction in spending on those who have insurance, with the net result being a 50% reduction in the dollars spend on employer-provided insurance.4
These are enormous impacts, and the potential for 22 million more uninsured Americans should lead anyone to pause before removing the existing employer subsidy. But it is also important to remember that the existing subsidy costs over $100 billion per year in forgone revenue. So this says that we are spending about $5000 per person to insure these 22 million persons, a quite high level. So the question becomes whether It is of course important to recognize that these projections are only as precise as the underlying estimates; the central estimates on which they are based are a probit coefficient on insurance offering of 5.424 (2.017), and an OLS coefficient on log spending of 1.465 (0.404).
The calculations in the final column are conjectural in that they assume that the firms that stop offering insurance are spending the average amount on insurance before dropping. It seems likely that those firms were spending less than average, so that there is a smaller reduction in overall spending than is implied in Table 5.
Taxes and Health Insurance 59
other alternatives are available at a more reasonable cost to insure large numbers of Americans. Table 5 also shows the impacts of less dramatic reforms to our existing system. The first is to remove the existing subsidy in the income tax, but in for the payroll tax. This would still lead to very large reductions in insurance offering and spending, with the potential for 13.6 million more uninsured. The final column illustrates that even very modest reforms to the tax code can have large effects on employer-provided insurance. This column shows the impact of cutting all tax rates by 10%, roughly akin to the original proposal of the Bush Administration. Even this very small change could lead to 1.4 million more uninsured, and a reduction in total spending on employer-provided insurance of almost 4%. One potential reform which would may less dramatic implications for coverage would be not to reduce the existing employer subsidy but to cap it, for example at the mean or median cost of a group insurance plan. As discussed in more length in Gruber and Poterba (1996), such a reform might temper the inflationary aspects of the tax subsidy highlighted by
Feldstein, and significantly raise government revenues, but without causing a major displacement of the employer-insured. In principle, administering such a cap would be straightforward: employers would simply be asked to report, for tax purposes (either income tax alone or payroll tax as well), any spending they make on an employee's behalf for insurance beyond some cap level. But, in practice, caps run into important administrative and political difficulties. First, there are very large regional disparities in the cost of health insurance, which would ideally be reflected in the cap level. But there has never been a regionally adjusted tax credit, and efforts in other arenas (e.g. poverty measurement) to have regional adjustment have run into daunting political difficulties. Second, a cap would penalize workers for having high-cost co-workers, since it would reflect average and not individual-specific insurance spending. This would lead to general redistributions from older, higher-cost industries to newer, lower-cost ones, raising further daunting difficulties.
5.2 New Subsidies to Employers A more likely direction for reform is to offer new subsidies to employers. But doing so immediately runs into the type of efficiency considerations highlighted above. The majority of employers already offer health insur-
ance. For this group, new subsidies are just redistribution, with no impact on insurance coverage, except through feedback effects on employees through reduced employee contributions or more generous insurance levels.
60
Gruber
Thus, to be cost-effective, new subsidies to employers must be well targeted. Table 2 illustrates how such targeting would be most effective:
if subsidies were largest for the smallest and lowest-wage firms. But there is a trade-off with targeting: the more targeted are subsidies, the more distortionary "cliffs" there may be to firms' decisions on pay and hiring. This was in fact an important criticism of the proposed subsidies to small firms embedded in the Clinton HSA plan. But such "cliffs" need not in fact arise, so long as there are smooth phaseouts with respect to wages and firm size. Consider for example the following subsidy structure:
There is a maximum subsidy rate of 0.4 for firms of 10 employees or fewer with average annual earnings of $10,000 or less. The subsidy is reduced by 0.01 for each extra employee above 10, so that it reaches zero at 50 employees. The subsidy is also reduced by the amount that earnings rise above $10,000 per year. This reduction factor is proportional to firm size that is, the bigger the firm, the faster the subsidy is reduced as wages rise. The formula is: Firms of 10 employees or fewer: subsidy reduced by 0.0222 for each $1000 rise in earnings above $10,000 per year.
Firms of more than 10 employees: subsidy reduced by (firm size)/ 450 for each $1000 rise in earnings in hourly wages above $10,000 per year. For example, for a firm of 25, the subsidy is reduced by 0.0555 for each $1000 rise in earnings above $10,000 per year.
I have simulated the effect of such a policy in the RWJF data used to compute Tables 2 and 3, drawing on the evidence presented above. The results suggest that for an annual cost of $3.8 billion, 2.2 million persons could be insured per year, for a cost of $1720 per newly insured. And there are very low implicit tax rates embedded in this gradual phaseout structure. On average, for every additional $100 paid out in wages, firms only lose $1.2 in subsidy; even at the maximum, the subsidy loss is only $15 per $100 paid out in wages, a modest distortion by the standards of our tax system. Similarly, for each worker hired, the subsidy loss is on average only $82; the maximum possible subsidy loss per hire is $720. So distortions need not be enormous to offer very targeted subsidies. But one interesting feature is that the efficiency of such firm subsidies diminishes as they get larger. For example, an expanded version of the
above subsidy plan that delivers a maximum subsidy of 50 percent
Taxes and Health Insurance 61
would cost 175 percent as much ($6.7 billion/year), but only cover 50 percent more persons (3.3 million), so that the cost per newly covered rises to $2070. This is a common finding in all simulation work on tax subsidies: their efficiency is inversely related to their scope. This is because as the subsidies get larger, they necessarily become less targeted and more attractive to those who are already providing insurance. Nevertheless, the efficiency of these firm subsidies seems quite high relative to the other policies considered below.
5.3 New Subsidies to Employees The second major alternative discussed earlier was new subsidies to employees to take up employer coverage. The first point to note about such subsidies is that, as a device for targeting take-up per se, they are likely to be very inefficient. This is because fewer than 10% of those offered insurance are actually uninsured. Moreover, the work reviewed above suggests that the take-up decision is not very elastic with respect to price. These two facts are a recipe for an inframarginal subsidy that wifi serve only as redistribution and not to increase insurance coverage.
But, in a general economic model, such subsidies should also increase employer insurance offering. Indeed, there is no economic rationale for not treating them symmetrically with a subsidy to employers:
both are subsidies to offering insurance through the workplace. In practice, however, their effects might differ somewhat. Employer subsidies are targeted to the characteristics of a firm, while employee subsidies are targeted to employees. This has the advantage that it may be possible to do better income targeting with employee subsidies, since even low-wage firms have high-wage workers. But it has the disadvan-
tage that it may be harder to target low-wage firms by simply giving subsidies to low-wage employees. Many low-wage employees work in high-wage firms that already offer insurance. So even a very tightly targeted subsidy to employees is likely to result in little new offering, as most of the dollars flow to those already in firms offering insurance.
5.4 Non-group Subsidies The final alternative is to subsidize non-group purchase of insurance. The prototypical proposal here, which is quite similar to that proposed by President Bush, would be: $1000 credit for individuals; $2000 for families
Usable for non-group insurance purchase only Refundable and advanceable Phased out for upper-middle- and upper-income families
62
Gruber
There are several difficulties faced by non-group credits. The first is a simple fact: roughly half of the uninsured do not pay taxes (Gruber and Levitt, 2000). This makes the tax code a problematic mechanism for
delivering subsidies to the uninsured. In principle, this can be addressed by making credits refundable. In practice, this raises two difficulties. First, refundability is a very difficult political goal in recent years, as many conservatives view refundable credits as akin to cash welfare. There were enormous battles over the refundability of the child credit in 1997 and again in 2001, with the result being only very partial refundability of this credit. Similar battles are likely to occur with these tax credits. Second, and more fundamentally, the usefulness of refundability is quite limited without effective advanceability. Insurance premiums are
due from the beginning of year t, but tax refunds do not arrive until spring of year t + 1. Thus, the uninsured must have sufficient resources to advance-fund their insurance purchases if credits are to be effective;
but most uninsured, indeed, most Americans, do not have sufficient liquid assets to do such advance funding. In principle, this could be surrounded by legislating an advanceable credit. But our one experience
with such a feature is not encouraging. Individuals can claim their Earned Income Tax Credit (EITC) throughout the year rather than the next spring, and it would be sensible for most claimants to do so; but fewer than 1% of claimants take advantage of this option (Liebman, 1998). The reasons for this low take-up are unclear, but the main conclusion is that low-income taxpayers appear to be reluctant to take any risk that they wifi end up facing a tax liability, rather than receiving a refund, on April 15. This will limit the effectiveness of advanceability for a nongroup credit as well. Finally, the major difficulty faced by such a credit is that non-group insurance is very expensive. Today, the typical non-group policy for a family costs $6,000 to $7,000 per year. Thus, even a sizable $2,000 credit leaves the family with $4,000 to $5,000 in costs to pay, which is enormous relative to the incomes of the working poor uninsured. This point ties to the earlier issue about market responses. If the insurance market can respond to the availability of this credit by delivering low-cost insurance products that are demanded by the public, then these affordability barriers may be overcome.
As a result of these limitations, my previous work suggests only modest impacts of nongroup credits on net insurance coverage (Gruber and Levitt, 2000; Gruber, 2000). This work is based on a major micro-
simulation model that takes as its base the 1997 Current Population
Taxes and Health Insurance
63
Survey, and imposes on that base a complicated set of behavioral equations that make assumptions on all of the relevant margins in Figure 1 (as detailed in the appendix to Gruber, 2000). The base case is a $1000 or $2000 refundable credit which is not effectively advanceable, due to the limitations noted above. I assume that this credit is available to singles with incomes up to $75,000 and families with incomes up to $100,000. It is worth noting that subsidies that are more tightly income-targeted are more efficient, as for incomes above the median there are relatively few
uninsured. But, once again, there is an important political question about whether it is feasible to have a truly targeted new credit in today's environment; all of the middle-class entitlements of the past 5 years have extended to income ranges similar to those used here (or higher). I find that for this base-case policy there is a cost of $13.3 billion per year, and net reduction in the uninsured of 4 million persons, for cost of $3300 per newly insured, well above the costs cited above for employer subsidies. The effectiveness of the credit rises significantly if it is advanceable, but the cost per newly insured remains above $2500. Thus, non-group credits do not appear nearly as effective as group credits, due to the problems noted above.
6. CONCLUSIONS Tax policy towards health insurance is likely to remain a topic of vigorous debate in the years to come. This paper has laid out some key facts, economic evidence, and policy simulations to help guide this debate. The key conclusion that I draw from existing facts and evidence is that policies targeted to firms are more likely to be effective than are policies targeted to individuals, as firms appear quite price-responsive in their insurance-offering decisions, and the actions of firms appear to be directly translated to individual coverage. But the prevalence of offering means that to be cost-effective such subsidies must be tightly targeted to the firms least likely to offer insurance: small and low-wage firms. The more important conclusion to be drawn from this paper is that we still know remarkably little about a number of key parameters that determine the effectiveness of tax policy towards insurance coverage. Most notable among these is the responsiveness, to new subsidies to buying insurance, of the existing uninsured who are not offered insurance. But there are a variety of other unanswered questions as well that must be addressed by future research if we are to draw fully informed conclusions as to the efficacy of tax policy.
64
Gruber
REFERENCES Blumberg, Linda, Len Nichols, and Jessica Banthin (2001). "Worker Decisions to Purchase Health Insurance," mimeo, Urban Institute. Chernew, Michael, Kevin Frick, and Catherine G. McLaughlin (1997). "The Demand for Health Insurance Coverage by Low-Income Workers: Can Reduced Premiums Achieve Full Coverage?" Health Services Research 32:453-470. Congressional Research Service. Costs and Effects of Extending Health Insurance Coverage. Washington, DC: U.S. Government Printing Office, 1988.
Cooper, Phillip, and Barbara Schone (1997). "More offers, fewer takers for employment-based health insurance: 1987 and 1996," Health Affairs, 16, 142149.
Cutler, David, and Jonathan Gruber (2001). "Health Policy in the Clinton Era: Once Bitten, Twice Shy," forthcoming in Jeffrey Frankel and Peter Drszag, eds., Economic Policy During the 1990s.
Cutler, David, and Jonathan Gruber (1996). "Does Public Insurance Crowd Out Private Insurance?," Quarterly Journal of Economics, 111(2), May 1996, 391-430.
Dubay, Lisa (1999). "Expanding Public Insurance Coverage and Crowd-Out: A Review of the Evidence," in Options for Expanding Health Insurance Coverage: l'Vliat Difference Do Different Approaches Make?, co-edited by Judith Feder and Sheila Burke (Washington DC, Henry J. Kaiser Family Foundation).
Employee Benefits Research Institute (2000). "Sources of Health Insurance and Characteristics of the Uninsured: Analysis of the March 2000 Current Population Survey." Washington, D.C.: EBRI. Farber, Henry, and Helen Levy (1999). "Recent Trends in Employer-Sponsored Health Insurance Coverage: Are Bad Jobs Getting Worse?" Journal of Health Economics. 19, 93-119.
Feldman, Roger, Bryan Dowd, Scott Leitz, and Lynn A. Blewett (1997). "The Effect of Premiums on the Small Firm's Decision to Offer Health Insurance," Journal of Human Resources 32:635-658.
Finkelstein, Amy (1999). "The Effect of Tax Subsidies to Employer-Provided Health Insurance on Workplace Pooling: New Evidence From Canada." Forthcoming in Journal of Public Economics.
Gentry, William and Eric Peress (1994). "Taxes and Fringe Benefits Offered by Employers." NBER Working Paper no. 4764. Goldstein, G. S.; Pauly, M. V. (1976). "Group Health Insurance as a Local Public Good" in The Role of Health Insurance in the Health Services Sector, Richard N. Rosett, ed. New York: Neale Watson Academic Publications, p. 73-114. Gruber, Jonathan (2000). "Medicaid," forthcoming in Means Tested Transfer Programs in the U.S., Robert Moffitt, (ed.) University of Chicago Press, (still forthcoming). Gruber, Jonathan (2001). "The Impact of the Tax System on Health Insurance Decisions." MIT. Mimeo. Gruber, Jonathan, and Michael Lettau (2000). "How Elastic Is the Firm's Demand for Health Insurance?" NBER Working Paper no. 8021. November. Gruber, Jonathan, and Larry Levitt (2000). "Tax Subsidies for Health Insurance: Costs and Benefits." Health Affairs 19:72-85. Gruber, Jonathan, and Brigitte Madrian (1996). "Health Insurance and Early Retirement: Evidence from the Availability of Continuation Coverage," in Ad-
Taxes and Health Insurance
65
vances in the Economics of Aging David Wise, ed. University of Chicago Press, pp. 115-143. Gruber, Jonathan, and Robin McKnight (2001). "Why Are Employee Health Insurance Contributions Rising?" MIT. Mimeo.
Gruber, Jonathan, and James M. Poterba (1994). "Tax Incentives and the Demand for Health Insurance: Evidence from the Self-employed," Quarterly Journal of Economics 109:701-733.
Gruber, Jonathan, and James Poterba (1996). "Tax Subsidies to EmployerProvided Health Insurance." In Empirical Foundations of Household Taxation, Mar-
tin Feldstein and James Poterba (eds.). Chicago: University of Chicago Press. Helms, W. David, Anne K. Gauthier, and Daniel M. Campion (1992). "Mending the Flaws in the Small-Group Market." Health Affairs 11:8-27. Holmer, Martin (1984). "Tax Policy and the Demand for Health Insurance." Journal of Health Economics 3:203-221.
Kaiser Family Foundation (2000). Employer Health Benefits Survey. Menlo Park, CA: Kaiser Family Foundation.
Liebman, Jeffrey B. (1998). "The Impact of the Earned Income Tax Credit on Incentives and Income Distribution," in James Poterba, ed. Tax Policy and The Economy 12, 83-120.
Liebowitz, Arleen, and Michael Chernew (1992). "The Firm's Demand for
Health Insurance." In Health Benefits and the Workforce, U.S. Department of Labor. Washington: U.S. Government Printing Office. Long, James E., and Frank A. Scott (1982). "The Income Tax and Nonwage Compensation." Review of Economics and Statistics 64:211-219.
Madrian, Brigitte C., and Dennis F. Shea (2001). "The Power of Suggestion:
Inertia in 401(k) Participation and Savings Behavior." Quarterly Journal of Economics. Vol. 116, No. 4, pp. 1149-4187. Marquis, M. Susan, and Stephen H. Long (1994). "Worker Demand for Health Insurance in the Non-group Market." RAND Corp. Mimeo. Marquis, M. Susan, and Stephen H. Long (1999). "To Offer or Not to Offer: The Role of Price in Employer Demand for Insurance." RAND Corp. Mimeo. Marquis, M. Susan, and Charles E. Phelps (1987). "Price Elasticity and Adverse
Selection in the Demand for Supplementary Health Insurance." Economic
Inquiry 25:299-313. Monheit, Alan, Barbara Steinberg Schone, and Amy Taylor (1999). "Health Insur-
ance Choices in Two-Worker Households: Determinants of Double Cover-
age." Inquiry 26:12-29.
Morissey, Michael A., Gail A. Jensen, and Robert J. Morlock (1994). "Small Employers and the Health Insurance Market." Health Affairs, 13, 149-161. Pauly, Mark (1986). "Taxation, Health Insurance and Market Failure in the Medical Economy," Journal of Economic Literature, 24, 629-675.
Royalty, Anne Beason (1999). "Tax Preferences for Fringe Benefits and Workers' Eligibility for Employer Health Insurance." Journal of Public Economics. Vol. 75, issue 2, February 2000, pp. 209-227. Sheils, John, and Paul Hogan (1999). "Cost of Tax-Exempt Health Benefits in 1998." Health Affairs 18:176-181.
Sloan, Frank, and Killard Adamache (1986). "Taxation and the Growth of
Nonwage Benefits." Public Finance Quarterly 14:115-139. Taylor, Amy, and Gail Wilensky (1983). "The Effect of Tax Policies on Expenditures for Private Health Insurance." In Market Reforms in Health Care: Current
66
Gruber
Issues, New Directions, Strategic Decisions, Jack Meyer (ed.). Washington: Ameri-
can Enterprise Institute.
Thorpe, Kenneth L, et al. (1992). "Reducing the Number of Uninsured by Subsidizing Employment-Based Health Insurance: Results from a Pilot Study." Journal of the American Medical Association 267:945-948.
Turner, Robert (1987). "Are Taxes Responsible for the Growth of Fringe Benefits?" National Tax Journal 40:205-220.
Vroman, Susan, and G. Anderson (1984). "The Effect of Income Taxation on the Demand for Employer-Provided Health Insurance." Applied Economics 16:33-43.
Woodbury, Stephen A., and Daniel S. Hammermesh (1992). "Taxes, Fringe Benefits, and Faculty." Review of Economics and Statistics 73:287-296.
DEFINED CONTRIBUTION PENSIONS: PLAN RULES, PARTICIPANT CHOICES, AND THE PATH OF LEAST RESISTANCE James J. Choi Harvard University
David Laibson Harvard University and NBER
Brigitte C. Madrian University of Chicago and NBER
Andrew Metrick University of Pennsylvania and NBER
EXECUTIVE SUMMARY We assess the effect on savings behavior of several different 401(k) plan features, including automatic enrollment, automatic cash distributions, We thank Hewitt Associates for their help in providing the data. We are particularly grateful to Lori Lucas and Jim McGhee, two of our many contacts at Hewitt. We also thank James Poterba and Olivia Mitchell for comments. Choi acknowledges financial support from a National Science Foundation Graduate Research Fellowship. Laibson and Madrian acknowledge financial support from the National Institute on Aging (ROl-AG-16605 and R29-AG-013020 respectively). Laibson also acknowledges financial support from the MacArthur Foundation and the Sloan Foundation.
68
Choi, Laibson, Madrian & Metrick
employer matching provisions, eligibility requirements, investment options, and financial education. We also present new survey evidence on individual savings adequacy. Many of our conclusions are based on an analysis of micro-level administrative data on the 401(k) savings behavior of employees in several large corporations that implemented changes in their 401(k) plan design. Our analysis identifies a key behavioral principle that should partially guide the design of 401(k) plans: employees often follow the path of least resistance. For better or for worse, plan administrators can manipulate the path of least resistance to powerfully influence the savings and investment choices of their employees.
1. INTRODUCTION Over the last 20 years, defined-contribution pension plans have gradually replaced defined benefit pension plans as the primary privately-
sponsored vehicle to provide retirement income. At year-end 2000, employers sponsored over 325,000 401(k) plans with more than 42 million active participants and $1.8 trillion in assets.1 The growth of 401(k)-type savings plans and the associated displacement of defined benefit plans have generated new concerns about the
adequacy of employee savings. Defined contribution pension plans place the burden of ensuring adequate retirement savings square on the backs of individual employees. However, employers make many decisions about the design of 401(k) plans that can either facilitate or hinder the employees' retirement savings prospects. Although the government places some limits on how companies can structure their 401(k) plans, employers nonetheless have broad discretion in their design. Making good plan design decisions requires an understanding of the relationship between plan rules and participant choices. In this paper, we
analyze a new data set that enables us to carefully assess many such relationships. The data set is compiled from anonymous administrative records of several large firms that collectively employ almost 200,000 individuals. Many of these companies implemented changes in the design of their 401(k) plans. These plan changes enable us to evaluate the impact on individual savings behavior of institutional variation in 401(k) plan rules. A list of the companies studied in this paper, along with the plan changes or other interventions that we analyze, appears in Table 1 Appendix A gives a brief description of the data analyzed for each company. 1
See EBRI Databook on Employee Benefits at http://www.ebri.org/facts/12OOfact.htm.
To maintain the anonymity of the companies described in this paper, we refer to them with letters. 2
TABLE 1.
30,000
20,000 10,000
40,000
50,000
Office equipment
Insurance
Food
Utility
Consumer packaged goods
Insurance
B
C
D
E
F
G
Change in eligibility
Change in eligibility Instituted employer match
Increased match threshold
Automatic enrollment
Automatic enrollment Financial education seminars
Automatic enrollment
Savings survey
Plan change or intervention
January 1997
July 1998 October 2000
January 1997
January 1998
January-December 2000
April 1998
January 1997
January 2001
Date of Change or Intervention
'Number of employees (rounded to the nearest 10,000) on December 31, 2000 (Companies A, B, D, 5, F), June 30, 2000 (Company C), or December 31, 1999 (Companies G, H).
30,000
10,000
Food
A
Sizea
Industry
Company
Companies and Their 401(k) Plan Changes or Other Interventions
70
Choi, Laibson, Madrian & Metrick
Because low employee savings rates have motivated plan administrators to adopt many of the 401(k)-plan changes that we discuss in the rest of the paper, we start off in Section 2 with a discussion of savings adequacy. Using new data from a survey that we designed, we find that two-thirds of employees believe that they are saving too little and that one-third of these self-reported undersavers intend to raise their saving rate in the next two months. By matching survey responses to administrative records, we show that employees who report that they save too little actually do have low 401(k) saving rates. However, almost none of the employees who report that they intend to raise their saving rate in the next two months actually do so. This finding introduces a theme that we return to throughout the paper. Specifically, at any point in time employees are likely to do whatever requires the least current effort: employees often follow the path of least resistance. Almost always, the easiest thing to do is nothing whatsoever, a phenomenon that we call passive decision. Such passive decisionmaking implies that employers have a great deal of influence over the savings outcomes of their employees. For example, employer choices of default saving rates and default investment funds strongly influence employee savings levels. Even though employees have the opportunity to opt out of such defaults, few actually do so. In section 3, the heart of our paper, we discuss the impact of changes in seven different types of plan rules. In section 3.1, we show that automatic enrollment in a 401(k) plan dramatically raises participation rates, but that the vast majority of employees accept the automatic-enrollment default contribution rate and investment allocation. By contrast, before automatic enrollment was instituted, few employees chose to invest at these defaults. In section 3.2, we discuss the effects of automatic cash distributions for terminated employees. We argue that automatic cash distributions, which are given to terminated employees with balances below $5,000, undercut retirement wealth accumulation. Most employees with balances below $5,000 who receive such automatic distributions consume the proceeds. By contrast, most employees with balances above $5,000 leave their money in the 401(k) plan. Hence, the automatic cash distributions seem to play a critical causal role in the consumption of these lowbalance 401(k) accounts. In section 3.3, we discuss different interventions designed to raise employee contribution rates. Benartzi and Thaler (2001b) have shown that employees are willing to commit to automatic schedules of slow 401(k) contribution rate increases, and that committing to such a sched-
Defined Contribution Pensions
71
ule wifi result in substantially higher 401(k) savings rates after only a few
years. We report an experiment of our own that shows that a savings intervention that does not include such an automatic commitment component is not successful. In section 3.4, we discuss the effects of the employer match rate and the employer match threshold (the maximum employee contribution that the employer matches) on savings outcomes. We show that adopting an employer match can increase 401(k) participation, and that the match threshold is an important focal point in the selection of employee contribution rates. We also show that increasing the match threshold can raise the contribution rates of individuals with low saving rates. In section 3.5, we discuss the impact of changes in eligibility waiting periods on the 401(k) participation profile (i.e. participation rates plotted against tenure at the job). We show that an increase in the length of time before 401(k) eligibility truncates, but does not shift, the participation profile.
In section 3.6, we discuss mutual-fund menus and the role of employer, or "company," stock. We argue that the menu of asset allocation options and the choice of the default asset allocation influence actual asset allocation decisions and portfolio diversification. Finally, in section 3.7 we discuss the role of financial education in the workplace. Using data that link employees' receipt of financial education to their actual saving behavior, we show that although many seminar attendees plan to make 401(k) savings changes, very few actually do so. Thus, while financial education does improve savings outcomes, its effects are modest at best. We see passive decisionmaking in many of the behavioral patterns described above. Passive decisionmaking partially explains the powerful influence of defaults, the anchoring effects of the match threshold, the remarkable success of automatic schedules of slowly increasing contribution rates, and the influence of mutual fund menus on asset allocation decisions.
We conclude the paper by encouraging employers to implement 401(k) plans that work well for decisionmakers who often use passive strategies like those described above. Employers and policymakers need to recognize that it is difficult to present a neutral menu of options for a 401(k) plan. Framing effects will influence employee choices, and passive employee decisionmaking implies that the default options will often carry the day. Sophisticated employers will choose these defaults carefully, keeping the interests of both employees and shareholders in mind.
72
Choi, Laibson, Madrian & Metrick
2. SAVINGS ADEQUACY In January 2001, we administered a savings adequacy survey to a random sample of employees at a large U.S. food corporation (Company A) with approximately 10,000 employees. Of these employees, 1,202 were sent an e-mail soliciting their participation in a Web-based survey on satisfaction with various aspects of the company-sponsored 401(k) plan.3 Because participation in the survey was solicited by e-mail and the survey itself was conducted on the Web, the universe of potential respondents is restricted to those with Internet access at work.4 Our survey had two different versions. In this section, we discuss the savings adequacy version that was sent to 590 of the employees with computers. From this sample we received 195 usable responses. A copy of the complete survey is reproduced in Appendix B; we discuss only a subset of the questions in the analysis below. In addition to the survey
responses, we also have administrative data on the 401(k) savings choices of survey respondents both before and after the survey. This includes participation decisions, contribution rates, and asset allocation choices from January 1996 through April 2001. We first asked respondents to report how much they should ideally be saving for retirement.5 The average response is 13.9 percent of income. We than asked respondents to evaluate their actual saving rate. Twothirds (67.7 percent) of the respondents report that their current saving rate is "too low" relative to their ideal saving rate.6 One-third (30.8
percent) of the respondents report that their current saving rate is "about right." Only 1 out of 195 respondents (0.5 percent) reports that his or her current saving rate is "too high." To evaluate how well individual perceptions of savings adequacy correlate with actual savings behavior, we report in Table 2 the distribution
The solicitation included an inducement to actually complete the survey: two respondents were randomly selected to receive gift checks of $250, and one respondent was selected to receive a gift check of $500.
" Naturally, restricting our sample to Internet users biases our sample toward employees with greater financial sophistication. Our survey reveals that an employee's level of Internet experience correlates with his self-reported financial knowledge. Likewise, home Internet access also correlates with financial knowledge. See question 10 from the survey (Appendix B).
See question 11 from the survey (Appendix B). For our empirical analysis we aggregate the categories "far too low" and "a little too low" into one category ("too low"). Likewise, we aggregate the categories "far too high" and "a little too high" into one category ("too high"). 6
Defined Contribution Pensions
73
TABLE 2.
Self-reported Retirement Savings Adequacy and the Distribution of Actual 401(k) Contribution Rates (Company A) Distribution of 401(k) Contribution Rates as a Fraction of Income
Respondents who describe their savings rate as "too low" Respondents who describe their savings rate as "about right"
5%-8%
9%-12%
36%
36%
27%
12%
15%
73%
See question 11 from the survey in Appendix B. We aggregate the categories "far too low" and "a little too low" into one category ("too low").
of actual pre-tax 401(k) saving rates conditional on respondents' answers
to the savings adequacy questions discussed above. Since we use the plan's administrative records, our analysis of actual 401(k) saving rates does not suffer from reporting biases. We divide the actual pre-tax 401(k) saving rates into three categories: 0 to 4 percent of income, 5 to 8 percent of income, and 9 to 12 percent of income. Our scale tops out at 12 percent
because this is the maximum pre-tax 401(k) contribution rate in Company A. Among the respondents who said that their current saving rate is "too low," 36 percent had an actual 401(k) saving rate of 0 to 4 percent,
another 36 percent had a saving rate of 5 to 8 percent, and 27 percent had a saving rate of 9 to 12 percent. In contrast, among those who said that their current savings rate is "about right," 12 percent had a 401(k)
saving rate of 0 to 4 percent, 15 percent had a saving rate of 5 to 8 percent, and 73 percent had a saving rate of 9 to 12 percent. These comparisons reveal that respondents who report that their saving rate is too low do have lower actual saving rates than respondents who report that their saving rate is about right. In the former group the average pre-tax 401(k) contribution rate is 5.8 percent of income, in contrast to 9.0 percent in the latter group. We also asked respondents to describe their plans for the future. None of our respondents expressed an intention to lower their contribution rate. But 35 percent of the respondents who said that their saving rate was too low intended to increase their contribution rate over the next few months. By contrast, 11 percent of respondents who said their saving rate was about right intended to increase their contribution rate over the next few months. Among those who planned to raise their contribution rate, over half (53 percent) said that they planned to do so in the
74
Choi, Laibson, Madrian & Metrick
next month. Another quarter (23 percent) planned to make the change within two months. So far our data shows a familiar pattern. Respondents report that they save too little and that they intend to raise their saving rate in the future. Other savings adequacy surveys reach similar conclusions (Bernheim, 1995; Farkas and Johnson, 1997). Our survey is distinguished by our ability to cross-check responses against 401(k) records. We have shown that respondents who say that their saving rate is too low actually do have substantially lower pretax 401(k) contribution rates. So their retrospective reports are accurate. We have also checked to see whether their forward-looking plans are consistent with their actual subsequent behavior. Of those respondents who report that their saving rate is too low and that they plan to increase their contribution rate in the next few months, only 14 percent actually do
increase their contribution rate in the four months after the survey. Hence, we find that respondents overwhelmingly do not follow through on their good intentions. In summary, out of every 100 respondents, 68 report that their saving rate is too low; 24 of those 68 plan to increase their 401(k) contribution rate in the next few months; but only 3 of those 24 actually do so. Hence, even though most employees describe themselves as undersavers and many report that they plan to rectify this situation in the next few months, few follow through on this plan.
Needless to say, these data are hard to interpret. It's not clear what subjects mean when they say that they save too little. It's also not clear what subjects mean when they say that they intend to raise their contribution rate in the next few months. However, this evidence is at least consistent with the idea that employees have a hard time carrying out the actions that they themselves say they wish to take. Employers seem to be concerned about such failures. Many of the institutional changes discussed below in Section 3 were initiated by plan administrators in an effort to raise employee saving rates.
3. Seven Institutional Features of 401(k) Plans In this section, we turn to an analysis of how several features of 401(k) plans affect employee 401(k) saving behavior.
3.1 Automatic Enrollment The typical 401(k) plan requires an active election on the part of employees to initiate participation. A growing number of companies, however, have started automatically enrolling employees into the 401(k) plan unless the employee actively opts out. While automatic enrollment is still
Defined Contribution Pensions
75
uncommon, a recent survey indicates that its adoption has increased quite rapidly over the past few years.7 The interest of many companies in automatic enrollment has stemmed from their persistent failure to pass the IRS nondiscrimination rules that apply to pension-plan provision. As a result of failing these tests, many companies have had to make either ex post 401(k) contribution refunds to highly compensated employees or retroactive company contributions on behalf of non-highly compensated employees in order to come into compliance. In addition, many companies have tried to reduce the possibility of non-discrimination testing problems by ex ante limiting the contributions that highly compensated employees can make. The hope of many
companies adopting automatic enrollment has been that participation among the non-highly compensated employees at the firm will increase sufficiently that non-discrimination testing is no longer a concern. While some companies have been concerned about the potential legal repercussions of automatically enrolling employees in the 401(k) plan, the U.S. Treasury Department has issued several opinions that support employer use of automatic enrollment. The first Treasury Department opinion on this subject, issued in 1998, sanctioned the use of automatic enrollment for newly hired employees.8 A second ruling, issued in 2000, further validated the use of automatic enrollment for previously hired employees not yet participating in their employer's 401(k) plan.9 In addition, during his tenure as Treasury Secretary, Lawrence H. Summers publicly advocated employer adoption of automatic enrollment.'0
A growing body of evidence suggests that automatic enrollmenta simple change from a default of non-participation to a default of participationsubstantially increases 401(k) participation rates." To assess the impact of automatic enrollment on savings behavior, we examine the experience of three large companies analyzed in Choi, Laibson, Madrian, and Metrick (2001) that implemented automatic enrollment between January 1997 and April 1998. Companies B and C implemented In a recent survey, Hewitt Associates (2001) reports that 14 percent of companies utilized automatic enrollment in 2001, up from 7 percent in 1999. 8 See IRS Revenue Ruling 98-30 (Internal Revenue Service, 1998).
See IRS Revenue Ruling 2000-8 (Internal Revenue Service, 2000a). See also Revenue Rulings 2000-33 and 2000-35 (Internal Revenue Service, 2000b).
See "Remarks of Treasury Secretary Lawrence H. Summers at the Department of Labor Retirement Savings Education Campaign Fifth Anniversary Event" at http://www.ustreas .gov/press/releases/ps785.htm along with related supporting documents. " See Madrian and Shea (2001a), Choi, Laibson, Madrian, and Metrick (2001), Fidelity (2001), and Vanguard (2001).
76
Choi, Laibson, Madrian & Metrick
automatic enrollment for new hires. Company D also implemented automatic enrollment for new hires, but in addition subsequently applied it to non-participating employees who were 401(k)-eligible at the time when automatic enrollment was initially adopted.12 Table 3 illustrates the difference in 401(k) participation rates by tenure before and after automatic enrollment. For each company, we report three columns of figures. The first and second columns contain the fraction of employees hired before and after automatic enrollment was implemented who are 401(k) plan participants at six-month increments of tenure.'3 The third colunm differences these participation rates, yielding the incremental impact of automatic enrollment on plan participation. In all three companies, 401(k) participation for employees hired before automatic enrollment starts out low and increases quite substantially
with tenure. At six months of tenure, 401(k) participation rates range from 26 to 43 percent at these three companies. Participation rates increase to 50 to 62 percent at 24 months of tenure, and to 65 to 69 percent at 36 months of tenure. The profile of 401(k) participation for employees hired under automatic enrollment is quite different. For these employees, the 401(k) participation rate starts out high and remains high. At six months of tenure, 401(k) participation ranges from 86 to 96 percent at these three companies, an increase of 50 to 67 percentage points relative to 401(k) participation rates prior to automatic enrollment. Because 401(k) participation increases with tenure in the absence of automatic
enrollment, the incremental effect of automatic enrollment on 401(k) participation declines over time. Nonetheless, at 36 months of tenure, 401(k) participation is still a sizable 31 to 34 percentage points higher under automatic enrollment. While most companies that implement automatic enrollment do so only for newly hired employees, some companies have applied automatic enrollment to previously hired employees who have not yet initiated participation in the 401(k) plan. Choi, Laibson, Madrian, and Metrick (2001) show that for previously hired employees at Company D, automatic enrollment also substantially increases the 401(k) participation rate, Because of concurrent changes in eligibility for employees under the age of 40 at Company D, we restrict the sample of employees in the analysis at the company to those aged 40 or over at the time of hire. These employees were immediately eligible to participate in the 401(k) plan, both before and after the switch to automatic enrollment. 13 Because of differences in the available data from these companies, the numbers are not directly comparable across companies. For Company C, we have data on 401(k) participation on the data collection dates, and thus the numbers in columns 1 and 2 for Company C represent contemporaneous 401(k) participation rates. For Companies B and D, we have data on the date of initial 401(k) participation, and thus the numbers in columns 1 and 2 for Companies B and D represent the fraction of employees who have ever participated in the 12
401(k) plan.
26.4 37.8 47.7 54.1 60.0 64.7
Before AE
93.4 95.7 97.0 97.6 97.7 98.8
After AE
Hire date
Company B
67.0 57.9 49.3 43.5 37.7 34.1
before
After-
--
35.7 40.2 44.3 49.8
-
85.9 85.3 86.0 85.7
-
45.1 41.7 35.9
50.2
Participation (%) Company C Hire date AfterBefore AE After AE before 42.5 49.6 56.6 61.7 65.6 69.0
100.0
96.0 96.6 97.2 99.1 98.8
Hire date Before AE After AE
Company D
53.5 47.0 40.6 37.4 33.3 31.0
before
After-
The sample for Companies B and C is all 401(k)-eligible employees. The sample for Company D is 401(k)-eligible employees aged 40+ at the time of hire. For Company D, the data in the "Before AE" column include only employees not yet subject to automatic enrollment when it was applied to previously hired non-participants. For Companies B and D, the first two columns of numbers give the fraction of employees who have ever participated in the 401(k) plan. For Company C, the first two columns give the fraction of employees contemporaneously participating in the 401(k) plan.
24 30 36
12 18
6
Tenure (months)
TABLE 3.
401(k) Participation by Tenure Before and After Automatic Enrollment
78
Choi, Laibson, Madrian & Metrick
although the increase in participation is slightly smaller than that seen for newly hired employees. Madrian and Shea (2001a) and Choi, Laibson, Madrian, and Metrick (2001) also discuss how the effects of automatic enrollment vary across various demographic groups. While automatic en-
rollment increases 401(k) participation for virtually all demographic groups, its effects are largest for those individuals least likely to participate in the first place: younger employees, lower-paid employees, and Blacks and Hispanics. One might conclude that since 401(k) participation under automatic
enrollment is so much higher than when employees must choose to initiate plan participation, automatic enrollment "coerces" employees into participating in the 401(k) plan. However, if this were the case, we should expect to see participation rates under automatic enrollment declining with tenure as employees veto their "coerced" participation and opt out. But remarkably few 401(k) participants at these companies, whether hired before automatic enrollment or hired after, reverse their participation status and opt out of the plan. In our three companies, the fraction of 401(k) participants hired before automatic enrollment who drop out in a 12-month period ranges from 1.9 to 2.6 percent, while the fraction of participants subject to automatic enrollment who drop out is only 0.3 to 0.6 percentage points higher. To us, this evidence suggests that most employees do not object to saving for retirement. In the absence of automatic enrollment, however, many employees tend to delay taking action. Thus, automatic enrollment appears to be a very effective tool for helping employees begin to save for their retirement. While automatic enrollment is effective in getting employees to participate in their company-sponsored 401(k) plan, it is less effective at motivating them to make well-planned decisions about how much to save for retirement or how to invest their retirement savings. Because companies cannot ensure that employees will cho9se a contribution rate or an asset allocation before the automatic enrollment deadline, the company must establish a default contribution rate and a default asset allocation. Most employees follow the path of least resistance and passively accept these defaults.
Figure 1 shows the distribution of 401(k) contribution rates at our three companies for employees hired before and after automatic enrollment. Because contribution rates may change with tenure, for all three companies we have restricted the sample to employees hired before and after automatic enrollment with equivalent levels of tenure.14 All three 14 In Company B, the sample is restricted to employees with 24-35 months of tenure; in Company D to those with 0-23 months of tenure; and in Company D to those with 12-35 months of tenure.
Defined Contribution Pensions
79
FIGURE 1. The Distribution of Contribution Rates of 401(k) Participants Hired Before and After Automatic Enrollment 80 70
J
5
6050
.2
71
64
42
40 30 20
1
U-
Iii 1214
C)
0 Company B Company B Company C Pre-AE
Post-AE
Pre-AE
AE default and <Match Threshold
42
3 27
22.-
12
Company C Company D Company D Post-AE Pre-AE Post-AE
AE Default Match Threshold
D>Match Threshold
companies match employee contributions up to 6 percent of compensation, the match threshold in Figure 1. But the default contribution under automatic enrollment is much lower than this-2 percent in company B and 3 percent in companies C and D. Before automatic enrollment, 63 to 79 percent of plan participants at these companies contribute at or above the match threshold. Only 11 to 20 percent voluntarily choose the contribution rate specified by their employers as the default under automatic
enrollment. In contrast, 42 to 71 percent of participants hired under automatic enrollment contribute at the default rate, while only 26 to 49 percent contribute at or above the match threshold. Automatic enrollment has similar effects on the asset allocation of plan participants. Figure 2 shows the allocation of 401(k) balances between stocks, bonds and the combination of stable value and money market funds. Once again, because asset allocation may change with tenure, we have restricted the sample to employees with equivalent levels of tenure.15 In two of the three companies, the default fund under automatic enrollment is a stable-value fund, while in the third it is a money market fund. As Figure 2 shows, employees hired before automatic enrollment have the majority of their plan assets (53 to 81 percent) allocated to the stock market, and only a small fraction of their assets (10 to 18 percent) allocated to stable-value or money market funds. These percentages are 15
See footnote 14.
80
Choi, Laibson, Madrian & Metrick
FIGURE 2. Asset Allocation of 401(k) Participants Hired Before and After Automatic Enrollment 100%