Advances in Taxation, Volume 13

LIST OF CONTRIBUTORS Dale Bandy School of Accounting University of Central Florida Donna Bobek School of Accounting U...

Author: Thomas M Porcano

101 downloads 914 Views 6MB Size Report

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

DOWNLOAD PDF

LIST OF CONTRIBUTORS Dale Bandy

School of Accounting University of Central Florida

Donna Bobek


Bradley D . Childs

Department of Accounting Belmont University

Anthony P . Curatola

Department of Accounting Drexel University

Richard C . Hatfield


Jodie Houston

Department of Commerce The Australian National University

Andrew J. Judd


Charles F. Kelliher


James R. Lackritz

College of Business Administration San Diego State University

Linda Garrett Levy

Department of Accounting University of Colorado-Denver

J. David Mason

Department of Accounting Clemson University vii

William A . Raabe

School of Business Samford University

William Shafer

Graziadio School of Business and Management, Pepperdine University

Alfred Tran

Department of Commerce The Australian National University

Janet Trewin


L . Melissa Walters-York

R .O Anderson School of Management University of New Mexico

Peter J. Westort

Department of Accounting University of Massachusetts-Boston

G. E. Whittenberg

College of Business Administration San Diego State University

EDITORIAL BOARD EDITOR

Thomas M . Porcano Miami University Kenneth Anderson University of Tennessee

Gary A . McGill University of Florida

Caroline K . Craig Illinois State University

Daniel P . Murphy University of Tennessee

Anthony P. Curatola Drexel University

Charles E . Price Auburn University

Ted D . Englebrecht Old Dominion University

William A . Raabe Samford University

Philip J . Harmelink University of New Orleans

Michael L . Roberts University of Alabama

D . John Hasseldine University of Nottingham

David Ryan Temple University

Peggy A. Hite Indiana University-Bloomington

Dan L . Schisler East Carolina University

Beth B . Kern Indiana University-South Bend

Toby Stock University of Colorado-Bould

Suzanne M . Luttman Santa Clara University

Is

AD HOC REVIEWERS Janet A . Meade University of Houston John R . Robinson University of Texas-Austin Patrick J . Wilkie George Mason University

ADVANCES IN TAXATION EDITORIAL POLICY AND CALL FOR PAPERS Advances in Taxation (AIT) is a refereed academic tax research annual. Academic articles on any aspect of federal, state, local, or international taxation will be considered. These include, but are not limited to, compliance, computer usage, education, law, planning, and policy. Interdisciplinary research involving economics, finance, or other areas also is encouraged. Acceptable research methods include any analytical, behavioral, descriptive, legal, quantitative, survey, or theoretical approach appropriate for the project. Manuscripts should be readable, relevant, and reliable. To be readable, manuscripts must be understandable and concise. To be relevant, manuscripts must be directly related to problems inherent in the system of taxation. To be reliable, conclusions must follow logically from the evidence and arguments presented. Sound research design and execution are critical for empirical studies. Reasonable assumptions and logical development are essential for theoretical manuscripts. AIT welcomes comments li'om readers.

Editorial correspondence pertaining to manuscripts should be forwarded to: Professor Thomas M. Porcano Department of Accountancy Richard T. Farmer School of Business Administration Miami University Oxford, Ohio 45056 Phone: 513.529.6221 Fax: 513.529.4740 E-mail: [email protected] See http://www.jaipress.com/ for Guidelines regarding manuscript submissions.

xiii

and additional

information

THE EFFECT OF POLICY OBJECTIVES, COMPLEXITY, AND SELF-INTEREST ON INDIVIDUALS' COMPARATIVE FAIRNESS JUDGMENTS OF A FLAT TAX Donna Bobek and Richard C . Hatfield

ABSTRACT The purpose of this study was to identify how individuals form fairness judgments about a tax system and what factors they attend to when comparing the fairness of a flat tax system with the current tax system. Based on equity theory and prior tax research the complexity of the tax system, the policy objectives achieved by the tax system and the financial effect of the tax system were hypothesized to influence individuals' fairness judgments. Further, the function served by an individual's attitude toward the tax system was expected to further explain when self-interest would be particularly salient. The study's hypotheses were tested using the responses to a questionnaire sent to a cross-section of U .S. citizens. The study participants compared the current federal tax system with a flat tax system that differed from the current system on three dimensions . The alternative system was less complex than the current tax system, it achieved different policy objectives

Advances in Taxation, Volume 13, pages 1-25 . Copyright O 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0.7623-0774-9

2

DONNA BOBEK AND RICHARD C . HATFIELD than the current system and the personal tax liability of the participants differed between the two systems . Regression analysis was used to assess the relative influences on the respondents' fairness judgments. Respondents' comparative fairness judgments were influenced by their judgments regarding economic goals achieved by each tax system, unjustified complexity, and, especially, self-interest .

INTRODUCTION Concern about the fairness of the federal income tax system is pervasive and persistent. One reason for the interest in tax fairness is its link to tax compliance (Spicer & Lundstedt, 1976 ; Spicer & Becker, 1980 ; Etzioni, 1986 ; Hite & Roberts, 1992 ; Moser et al ., 1995) . Graetz (1997) notes that there is much evidence about the declining trend in tax compliance, which is partly due to the belief by many taxpayers that they are not treated fairly by the tax system . A second reason for the interest in tax fairness is the role that it plays in the debate over tax reform . The last major overhaul of the tax system occurred in 1986 . A driving force of that legislation was a concern for horizontal equity (Weiss, 1996) . Since 1986, there have been a number of incremental changes to the tax system (1991, 1993 and 1997) . These incremental changes have renewed the perceived need for "fundamental" tax reform .' If tax reform is to be politically successful, as well as have a favorable effect on tax compliance, it must be perceived as improving fairness . This study investigates taxpayers' fairness judgments of a flat tax proposal relative to the current tax system. Although prior studies have investigated the fairness of changes made by the Tax Reform Act of 1986 (Wartick, 1994 and Hite & Roberts, 1992), it is not clear how fairness judgments would be affected by a fundamental change in the tax system, such as with a "flat" tax . The motivation behind TRAM was primarily one of horizontal equity (distributive fairness) while the primary motivation behind the flat tax proposals (and more recently, the retail sales tax proposals) is simplicity (procedural fairness) . In addition, the institution of a flat tax would be a replacement of the entire tax system . To determine the viability of the flat tax as a replacement of the current system, empirical research is necessary to consider numerous ramifications of the change, including the perceived fairness of the flat tax relative to the current system . In the present study, equity theories (e .g . Leventhal, 1980) are utilized to identify factors that taxpayers should consider when assessing fairness . This theory is tested by having actual taxpayers compare the current federal income tax system to an alternative system that is similar to some of the recent flat tax

The Effect of Policy Objectives, Complexity and Self-Interest

3

proposals . We extend prior research (Hite & Roberts, 1992 ; Wartick, 1994) regarding the fairness of tax reform by including two dimensions of fairness, distributive and procedural justice (Leventhal, 1980) in addition to self-interest . The results of our analyses suggest that, while taxpayers appear to be somewhat concerned with economic and complexity issues, financial selfinterest is a highly salient influence on their fairness judgments . The respondents did associate simplicity (both the amount of effort required to file taxes and a procedural justice issue dealing with the number of loopholes) of the flat tax with fairness . However, the differences in the policies achieved by the two tax systems did not strongly affect their fairness judgments . Only the objective of achieving economic goals was related to fairness judgments . The results of this study suggest that achieving an overall improvement in fairness ratings may be difficult as long as revenue neutrality also is a goal . Leonard (1996) identified a number of lessons from the 1986 tax reform process . One lesson was that tax reform is about trade-offs that create winners and losers . The results of this study suggest that the winners will be much more likely to consider the new system as fair than will the losers . Further, this study also finds evidence that self-interest affects the respondents' judgments about which system better achieves policy objectives and simplicity objectives . There is, however, some prior research that has found that education (Christensen & Weirich, 1996 ; Ericksen & Fallan, 1996 ; White et al ., 1990 ; Wartick, 1994) and communication (Roberts, 1994 ; Maroney et al ., 1998) may mitigate the effect of self-interest . Taken together, this suggests that care must be taken to effectively communicate the benefits and justifications for tax changes if the desired improvement in fairness perceptions is to be achieved . The remainder of this paper is organized as follows : in the next section theory and prior research are discussed and hypotheses are proposed . The following section describes the research method used to test the study's hypotheses and reports the results . In the last section, conclusions and future research are discussed .

THEORY AND DEVELOPMENT OF HYPOTHESES Prior Tax Research

While tax fairness perceptions and their relationship to compliance have been studied extensively (e .g. Maroney et al ., 1998 ; Worsham, 1996 ; Sheffrin, 1994 : Hite & Roberts, 1992 ; Kaplan & Reckers, 1985 ; and Scott & Grasmick, 1981), less attention has been paid to the perceived fairness of explicit changes to the tax system . Since the 1996 and 2000 presidential election campaigns have

4

DONNA BOBEK AND RICHARD C . HATFIELD

included significant discussion of "radical" tax reform, a better understanding of how fairness perceptions are affected by tax changes is needed . Two studies that investigated the perceived fairness of changes made by the Tax Reform Act of 1986 (TRA86) are Hite and Roberts (1992) and Warrick (1994) . Hite and Roberts (1992) investigated the effectiveness of TRA86 at achieving its goals of improving fairness, simplicity and economic growth . They found that fairness and simplicity were both related to subjects' judgments of the tax system . In addition, they found little correlation between self-interest (i .e . whether participants thought their tax liability would increase or decrease) and fairness ratings of ten specific provisions . However, the fairness rating of the overall system was highly correlated with their self-interest measure . Finally, Hite and Roberts found that even though most "experts" judged TRA86 as improving fairness (Sheffrin, 1994), their respondents found it to be less fair than the prior law . Wartick (1994) studied the perceived fairness of two specific changes made to the tax law by TRA86 . She hypothesized and found a self-interest bias (i .e . those made worse off by the change considered it less fair) . She also hypothesized that the self-interest bias could be mitigated by providing subjects with justification for the change . The presence of a justification was successful at mitigating, but not eliminating, the self-interest bias for only one of the tax changes. She concluded from this that taxpayers must consider the justification to be compelling if it is to mitigate the self-interest bias . Taken together, these two studies do not provide a consistent understanding of how self-interest influences taxpayers' fairness judgments of tax reform . Additionally, prior research has not addressed what characteristics of a tax law change, beyond self-interest, influence taxpayers' perceptions of the fairness . We extend this prior research in at least three ways . First, we consider fundamental tax reform .' Second, we draw on equity theory to hypothesize what characteristics of tax reform will affect taxpayers' fairness judgments . Third, we attempt to clarify the role of self-interest . Understanding how fairness judgments are made with respect to such a large tax law change is necessary as such proposals continue to be considered by Congress (Hamilton, 2000) . The current study draws on equity theory to form a theoretical framework for addressing these research questions . Equity Theory Equity theory historically began with a discussion of what is now referred to as distributive justice. Adams (1963, 1965) claimed that people compare the ratio of their work inputs (effort) to their work outputs (reward) to the same ratios of others (e.g . coworkers) . If these ratios are not equal, tensions can occur causing


5

those receiving relatively more outputs to feel guilty, while causing those receiving relatively less outputs to feel angry. Adams' work was followed by several empirical studies (e .g . Andrews, 1967 ; Garland, 1973 ; Pritchard et al ., 1972) that provided strong support for the notion that distributive justice concerns influence inputs (primarily effort in work-related studies) . Thibaut and Walker (1975) put forward a complementary theory, which was later refined by Leventhal (1980) and Leventhal et al . (1980) . These studies expand equity theories to include procedural justice as well as distributive justice . Procedural justice refers to the notion that consideration of the fairness of the underlying process by which decisions are made and outcomes determined may influence attitudes and actions beyond consideration of the fairness of the distribution itself . Leventhal (1980) offers a comprehensive theory of fairness : justice judgment theory . He identified distributive justice and procedural justice as characteristics that individuals consider when making judgments about fairness . Distributive Justice in a Tax Compliance Setting The obvious factor considered in tax studies of distributive justice is the tax rate structure . Porcano (1984) and Hite and Roberts (1991) found that respondents preferred a mildly progressive tax rate structure . Porcano (1984) also found evidence that taxpayers preferred a tax system that differentiated on the basis of taxpayers' needs . Porcano and Price (1992) found that a construct they termed "equality" (all taxpayers being treated the same) was associated with respondents' fairness ratings of various tax provisions . All of these findings are consistent with distributive justice theory . Kinsey (1989) argued that in terms of understanding tax fairness judgments, a consideration of distributive justice without a consideration of the policies that create the distributions might be missing the point . She identified the concept of policy justice as the fairness of the content of the tax laws . It is the tax policies, along with the tax rates, which determine the distribution rules of a particular tax system . Therefore, it is the degree of support for the tax policies that should determine whether an individual believes the particular tax system is fair. It is this aspect of distributive justice considered in the current study . Porcano and Price (1992), Hite and Roberts (1992) and Wartick (1994) all studied the fairness of specific tax provisions . However, Kinsey et al . (1991), Porcano and Price (1992) and Maroney et al . (1998) all find that the fairness ratings of specific provisions are not sufficient to explain taxpayers' fairness judgments of the overall tax system . Further, Sheffrin (1993) argues that it is not an economical use of the average citizen's time to be conversant on the full range of tax policies available . Therefore, we operationalize Kinsey's (1989) policy fairness concept using broad objectives . This seems particularly justified

6


considering that many of the current tax reform proposals are fundamentally different than the current system . There are three general policies that the federal income tax system addresses . First, the primary purpose of any tax system is to raise money to run the government. Second, the tax system can be used to achieve economic goals (e .g . encouraging savings and investment, aiding job creation, etc) . Third, the tax system can be used to achieve social goals (e .g . encourage charitable giving, help the needy, encourage home ownership, etc .) . According to Kinsey's (1989) theory of distributive tax fairness, taxpayers' fairness judgments will be related to how well a particular tax system achieves the policies they believe are important. Therefore, in a comparison of the current tax system with a flat tax system, taxpayers' judgments of the comparative fairness will be related to how well each system achieves the policies believed to be important . Therefore, Hypothesis 1, stated in the alternative is : Hl : In comparing two tax systems, taxpayers will judge the system that is

deemed better at achieving important (tax, economic or social) policies, as fairer . Procedural Justice in a Tax Compliance Setting

The most common procedural fairness variable studied by prior tax research is complexity (e.g . Carnes & Cuccia, 1996 ; Hite & Roberts, 1992 ; Christensen et al ., 1994 ; Porcano & Price, 1992 ; Milliron, 1985) . The tax system has long been criticized for being overly complex, and recent changes (e .g. Taxpayer Relief Act of 1997) have only exacerbated the problem . However, the results of prior research regarding the relationship between complexity and fairness have not provided definitive answers . Carnes and Cuccia (1996) sought to reconcile previous conflicting results and came to several conclusions . Complexity is negatively related to fairness perceptions, however this relationship is moderated by the justification for the complexity . Specifically, justified complexity did not negatively affect fairness judgments to the same degree as unjustified complexity . In their study, the complexity items that were considered the most unjustified were : choice of tax form, taxation of capital gains, determination of itemized deductions, and the determination of tax credits? Note that most, if not all, of these items would be eliminated by a flat tax? Following the findings of Carnes and Cuccia (1996), we focus on unjustified complexity as our procedural justice issue . There are at least two specific areas where complexity has been found to be unfair . First, complexity has been deemed unfair when it creates excessive computations (Carnes & Cuccia, 1996) .


7

Second, complexity has been perceived as unfair when taxpayers believe that the complexity results from special provisions that are only helpful to some taxpayers - i .e. loopholes exist (Milliron, 1985 ; Christensen et al ., 1994) . Hypothesis 2 is a two-part hypothesis, stated in the alternative : In comparing two tax systems, taxpayers will judge the system that requires the least effort to compute tax due as fairer . 1-12A :

H2B : In comparing two tax systems, taxpayers will judge the system they

believe is easiest for other taxpayers to exploit as less fair .

Self-Interest

Wartick (1994) identified a strong influence on fairness judgments depending upon whether respondents were made better off or worse off by the particular provision of interest. Porcano (1984) and Hite and Roberts (1991) also identified a self-interest bias . Hite and Roberts (1992) however found little evidence of a self-interest bias . Leventhal (1980) considers the possibility of a self-interest bias within the framework of his theory . He discusses the "rule of justified self-interest" where people feel it is fair to take as much for themselves as possible in certain situations . Further, Messick and Sentis (1983) state that the traditional determination of fairness requires more information than most people have . People generally know much more about their own cost/benefit ratio than they do of others' . This asymmetry of information leaves ample room for one's own preferences (i .e . costs and benefits) to influence ones' perceptions of fairness . This leads to the primary expectation regarding self-interest stated in Hypothesis 3 . 113 : Taxpayers' fairness judgments will be positively related to how they perceive they will be financially affected by tax reform . Although Hypothesis 3 predicts a main effect for self-interest, it is likely that this effect will be different for different people . Herek's (1986) functional theory of attitudes 5 suggests that a person's attitude toward the tax system will influence how self-interest affects fairness perceptions . Herek (1986) divides attitudes into two categories that emphasize the source of benefit derived from holding an attitude . The benefit of holding an attitude may come from the object (e .g . reducing taxes) or it may simply come from an expression of that attitude (a psychological benefit) . If the source of benefit is the attitude object itself, then the attitude is said to serve a utilitarian function . If the source of the benefit is

8


expression of the attitude then the attitude is said to serve an expressive function . This suggests that the more utilitarian a taxpayer's attitude toward the tax system is, the more likely it is that his/her fairness judgments will be affected by a self-interest bias . This leads to Hypothesis 4 . H4 : The effect of self-interest will be moderated by the function served by the taxpayer's attitude toward the tax system . The more utilitarian the function, the larger the influence of self-interest. To summarize the predictions of the hypotheses, a model of the comparative fairness ratings of an alternative tax system (A) and the current tax system (C), can be expressed as follows : F = a,(DJ A c) + a2(PJ A c) + 0'3(Sl) + a4 (AF) + a s(Sl X AF)

(1)

where : F = comparative fairness judgment of the current system vs . an alternative tax system . DJA _ c = judgment of which system better achieves important policies (distributive justice) . PJ A _ c = judgment of which system minimizes unjustified complexity (procedural justice) . Sl = whether subjects believe they would be better off or worse off with the alternative tax system. AF = attitude function .

RESEARCH METHOD Subjects Experienced taxpayers completed an experimental instrument consisting of a questionnaire sent out to a cross-section of the U .S . population .' The total design method of maximizing response rates was used in an effort to obtain as many responses as possible (Dillman, 1978) . This initial mailing was followed-up with a postcard reminder one week later . A final follow-up, which included a duplicate questionnaire, was mailed to 25% of the non-respondents four weeks after the initial mailing . Of the 1,000 questionnaires originally mailed, 600 were sent to households that were below the median household income level and 400 were sent to households that were above the median household income level .7 Table 1 details the response rate for each group . The usable response rates for each group were 16 .1 % and 25 .2%, respectively, and the overall response rate was 20% .R To test the possibility of non-response


Table 1 .

Response Rates by Income Group . Below Median Household Income

Above Median Household Income

Total

600

400

1000

(86)

(23)

(109)

514

377

891

93

98

201

(10) 83

(3) 95

(13) 178

Initial Sample Less : Returned as Undeliverable Adjusted Sample Size Returned Questionnaires Less : Unusable Responses Net Usable Responses Total Returned/Adjusted Sample Size Usable Responses/Adjusted Sample Size

18.0%

26 .0%

22.5%

16.1%

25 .2%

20.0%

bias, responses were compared between those returning the completed instrument early and those returning the completed instrument later . No significant differences were discovered . Table 2 summarizes the demographics of the questionnaire respondents . The average age of the respondents was 50 .1 years . Seventy-four percent of the respondents were men and 26% were women . The $20,001-$50,000 income group had the most respondents (71) ; while the over $75,000 income group had the least (25) . Most of the respondents had completed at least some college and considered themselves moderate to conservative on economic issues . Compared to the national population, the respondent group was older, more educated and mostly male . 9 Questionnaire and Variable Measurement The questionnaire was designed to obtain measures of the variables identified in the model . The alternative tax system described in the instrument is presented in the appendix . The following is a summary of the questionnaire contents.

• Respondents' beliefs about what goals are important for a tax system to achieve were measured first.

10


Table 2. Sample Demographic Comparisons . Sample'

National Population'

Age Under 25 25-44 45-65 65+

= = = =

6% 36% 33% 25%

Age 20-24 25-44 45-65 65+

= = = =

9% 44% 29% 18%

Family Income $20,000 or less $20,001 - $50,000 $50,001-$75,000 $75,000+

= = = =

18% 42% 26% 15%

Family Income Less than $15,000 $15,000 - $49,999 $50,000 - $74,999 $75,000+

= = = =

14% 47% 20% 19%

Sex Male Female

= 74% = 26%

Sex Male Female

= 48% = 52%

= = = = =

Education Did not Complete High School High School Graduate Some College College Graduate Post-Graduate Study

= 18% = 34% = 24% = 16% = 8%

Education Did not Complete High School High School Graduate Some College College Graduate Post-Graduate Study

4% 17% 43% 22% 14%

Who Prepares Tax Return? Me and/or Spouse = 44% Paid Preparer = 50% Other = 6% 'Numbers are expressed as a percentage of the sample of respondents . 'Age and Sex Estimates : U .S . Bureau of the Census estimates for 7/1/96 . Numbers represent percentage of adults over age 19 in each category . Family Income : U .S . Bureau of the Census . Annual Demographic Survey . Total Money Income in 1995 . Numbers represent the percentage of families in each category . Education : U .S . Bureau of the Census, March 1996 Current Population Survey . Number represents percentage of persons over 25 in each category .

• After reading about the alternative tax system, respondents indicated : • which of the two tax systems they thought best achieved each of the goals identified in Section I of the questionnaire (this combined with the importance ratings resulted in the DJA C terms) ; • and which system best achieved procedural fairness in terms of the complexity in the tax system (this provided the PJ A_c terms) .

The Effect of Policy Objectives, Complexity and Self-interest

II

• Respondents were asked whether they thought they would pay less tax under the current system or the alternative tax system (this provided the SI measure) .

• The function served by respondents' tax attitudes also was measured (AF) . • The dependent measure collected was the respondents' assessment of the fairness of the current tax system versus the alternative tax system (F) .

• Demographic information was collected . One additional feature of the design is that each subject was exposed to one of two rate schedules as part of the alternative tax system (individuals were subject to a flat tax rate of either 20% or 25%) and one of two examples of how the new tax system worked . Two rate schedules were used in order to have a fairly even split between those who believed they would pay less with the alternative tax system and those who believed they would pay more . 1 ° Two examples were used so participants could better relate to the instrument . Individuals sampled from the income group lower than the median household income read an example that depicted a family with a household income of $22,000, while individuals sampled from the higher income group read an example that depicted a family with a household income of $59,000 . Both descriptions and examples are presented in the appendix . To insure that there were no unintended effects from instrument-version differences, income example and rate level were included in the main regression . Neither of these variables were significant at the 0 .05 level in explaining fairness judgments and, therefore, are not included in the results . Each of the variables included in the model are discussed below .

Dependent Variable (F) The dependent variable used in this study was a measurement of the respondents' answers to the question, "compared to each other, which tax system is the most fair?" This question was coded on a +3 (the alternative tax system is much fairer) to -3 (the current tax system is much fairer) .

Distributive Fairness Three policy objectives were considered in this study . These variables are represented in equation (1) as DJ A c and act as measures of distributive justice . The first policy objective measured in this study was how well the tax system achieved social goals (SG) . Participants were told that examples of social goals include helping low income families, encouraging home ownership, encouraging charitable giving, and redistributing wealth . The second policy objective measured how well respondents thought the tax system achieved economic goals

12

DONNA BOBEK AND RICHARD C. HATFIELD

(EG) . Participants were told that examples of economic goals include helping to create investment by business and encouraging savings and investment by individuals . The third policy objective was whether the system was able to just raise enough money to run the government, without trying to achieve social and economic goals (RM) . These variables were measured by asking respondents which tax system better achieved each policy objective . The scales were coded from -3 (current system better achieves this objective) to +3 (alternative system would better achieve this objective) . All three variables were then weighted by multiplying them by the subjects' response concerning the importance of that particular objective (measured from 1 (not important) to 7 (very important)) ." Using this coding scheme, Hypothesis 1 predicts a positive effect of SG, EG, and RM on the dependent variable. Procedural Fairness

As stated previously, this study concentrates on "unjustified complexity" as the procedural fairness issue . We measured two variables that address two different aspects of this type of complexity . In equation (1), these variables are represented by the PJ A c term . The first variable measured which tax system has the fewest loopholes that can be used to avoid paying taxes (LH) . This measure was coded from -3 (current system has fewer loopholes) to +3 (alternative system has fewer loopholes) . The second complexity variable dealt with the amount of effort necessary to comply with a tax system (EFF) . This variable was coded from -3 (current system requires the least amount of effort) to +3 (alternative system requires the least amount of effort) . Using this coding scheme, Hypothesis 2 predicts a positive effect of LH and EFF on the dependent variable . Since EFF is a measure of general complexity rather than of unjustified complexity, an additional variable was measured to obtain the unjustified portion of EFF . Complexity necessary (CN) is a measure of the respondents' judgment of how complex a tax system needed to be to achieve the appropriate objectives . Respondents were asked, "In general, in order for a tax system to achieve all of the things that are important to you, how complex do you think it needs to be?" The response scale was coded from I (very simple) to 7 (very complex) . This variable is expected to interact with EFF . The direction of this interaction is negative based on the manner in which the variables were coded. An increase in the respondent's estimation of the complexity necessary in a tax system should result in a decrease in the effect of effort required on the perceived fairness of a tax system.


13

Self-Interest The self-interest variable (Sl) was a class variable . Respondents indicated whether they thought they would pay more (coded -1), the same (coded 0), or less (coded 1), with the alternative tax system (compared to what they pay with the current system) . Using this coding scheme, Hypothesis 3 predicts a positive effect of Sl on the dependent variable . Using the two rate schedules discussed earlier, 45% of the respondents believed they would pay less, 22% believed they would pay about the same, and 33% believed they would pay more under the alternative tax system . Attitude Function Attitude function was measured using a "global" measure of attitude function obtained from the demographic section of the questionnaire . Respondents were asked, "Please circle a number to show what you mainly consider when deciding whether a tax system is fair or not ." The response scale was coded from -3 (Financial Effect on Me Personally) on one end, to +3 (Objectives Achieved by Tax System) on the other end . There is no main effect expected for this measure . However, given the coding scheme, Hypothesis 4 predicts a negative interaction with the self-interest variable .

RESULTS Descriptive Statistics Table 3 reports the means and standard deviations of the variables used to test the research hypotheses . As reported in Table 3, the average respondent rated the alternative tax system as fairer than the current system, believed he/she would pay less with the alternative system, believed the alternative system was better at achieving simplicity goals, and rated the alternative system better at achieving economic and social goals . The negative mean for Attitude Function indicates that the average respondent considered him/herself to be on the utilitarian side of the attitude function continuum . Hypotheses Tests Hypotheses I and 2 consider the policy objective variables and the complexity variables respectively . As stated above, these variables are expected to have a positive effect on the comparative fairness judgments . Table 4 reports the regression results . The adjusted R-squared for this model is 0 .52 . Since the

14

DONNA BOBEK AND RICHARD C . HATFIELD Table 3 .

Variable

Descriptive Statistics of Variables . Mean

Standard Deviation

Number of Observations

Range* (Min, Max)

1 .003

1 .56

178

-3,3

SG (Social Goals)

0 .79

9 .84

169

-21,21

RM (Raise Enough (Money)

5 .06

8 .67

172

-21,21

EG (Economic Goals)

5 .43

10 .84

171

-21,21

EFF (Effort Required)

2 .31

1 .35

176

-3,3

LH (Loop Holes)

1 .77

1 .71

175

-3,3

0.12

0 .88

174

-1, 1

-0.41

1 .84

174

-3,3

2.12

1 .42

177

L 7

F (Relative Fairness of Current System Compared to Alternative System)

Sl (Financial Effect of Alternative System Compared to Current System) AF (Attitude Function) CN (complexity deemed necessary) necessary)

* The scale for all variables other than AF and CN were designed such that a negative response indicates that the current tax system better achieves that goal or measure and a positive response indicates that the alternate tax system better achieves that goal or measure. The AF scale goes from -3 (participant considers only the personal financial effect of a tax system) to +3 (participant cares only about the objectives achieved by the tax system). The CN scale measures how complex the participant thinks the tax system needs to be from I (very simple) to 7 (very complex) .

hypotheses make directional predictions, the reported p-values are based on directional tests (equivalent to one tailed t-tests) . Hypothesis 1 predicts a significantly positive effect of SG, EG, and RM . However, only EG has a significant effect (p-value = 0 .002) . Neither SG nor RM have a significant effect on the dependent variable (p-values equal 0 .25 and 0 .18 respectively) . These results provide limited support for Hypothesis 1 .

The Effect of Policy Objectives, Complexity and Self-Interest Table 4.

15

General Linear Model Results .

F = a,,+ a 1 SG + a ,RM + a9EG + a9 EFF + a 5 LH + a6 SI + a y4F + a 8SIxAF + a 9CN + a i0 CNxEFF (n= 161) Source

F-value

Social Goals (SG) Raise Enough Money (RM) Economic Goals (EG) Effort Required (EFF) Loopholes (LH) Self Interest (5I) Attitude Function (AF) SIxAF Complexity deemed necessary (CN) CNxEFF Model

p-value*

Coefficient Estimate**

0.48 0 .86 8 .61 3 .80 15 .05

0 .25 0 .18 0.002 0.03 0.0001

-0.01 +0.01 +0.03 +0.24 +0.28

5 .47 0 .01 3 .40

0.003 0 .94 0.016

+0.33 -0.003 -0.13

1 .27 5 .30 15 .15

0.26 0.012 0.0001

-0.11 --0.08

Adjusted R 2 =0 .52 The dependent variable (F) represents the respondent's opinion regarding the relative fairness of current system compared to alternative system . SG, RM, EG, EFF, LH and CN represent perceived characteristics of the tax systems . SI and AF represent characteristics of the taxpayer (respondent) . * p-values are based on directional tests based on the hypothesized direction of the variable's effect on the dependent variable . ** To obtain coefficient estimates, the self-interest variable is treated as continuous . This assumption does not change the level of significance for any of the variables reported in the table (which assume that self-interest is a class variable) . The estimate for the intercept was 0 .289 and was not significant (p-value >0 .35) .

Hypothesis Table

4

2A

predicts a positive effect for the effort required variable

(EFF) .

reveals that this variable had a significant effect on the comparative

fairness of the alternative system over the current system (p-value

= 0 .03) .

The means suggest that respondents thought the alternative system would

(2 .3 on scale from -3 to +3) . In addition, the interEFF is significant and in the expected direction . This finding additional support for Hypothesis 2A . The main effect of the effort

be much simpler to use action of provides

CN

and

variable is significant and is moderated by the respondents' perception of the amount of complexity necessary in a tax system . Therefore, the simpler the respondents believed a tax system needed to be, the greater the effect the

EFF

variable .

of

16


Hypothesis 2B predicts a positive effect for LH . Table 4 shows a significant positive effect for this variable (p-value = 0 .0001) . Therefore, respondents' fairness judgments were influenced by concern over loophole opportunities created by complexity . This is consistent with the findings of Milliron (1985) and Christensen et al . (1994) . Hypothesis 3 predicts a positive effect for the self-interest variable (SI) . Table 4 shows that Sl has a positive and significant (p-value = 0 .003) effect on the comparative fairness judgment of the respondents . Table 4 also reveals the significantly negative effect of the interaction between SI and the respondents' attitude function (p-value = 0.015) . This finding is consistent with the prediction of Hypothesis 4 which states that the more utilitarian the taxpayer's attitude function the larger the influence of self-interest .

Additional Analysis Some additional tests were performed on the self-interest variable . The selfinterest variable was found to be correlated with all of the policy and complexity variables . A test for multicollinearity indicated that the correlation did not unduly influence the estimates . The highest variance inflation factor was 1 .7 (a VIF in excess of 10 would indicate a problem (Neter et al ., 1990)) . However, a closer look at this correlation provides some insight into the global effect of self-interest . Table 5 presents the means of the distributive and procedural justice variables for each level of Sl (alternative tax system would decrease/not change/increase amount paid) . The respondents' perception of which tax system better achieved the objectives was significantly influenced by the monetary effect on the respondent . For example, if respondents felt they would pay less tax under the current system, then the current system was deemed to better achieve social goals . However, if the alternative system resulted in less tax, it was deemed to better achieve social goals . With all of the other variables, the alternative system always was deemed to be better at achieving the objective (or of being less complex) . However, the self-interest variable affected the extent of this favor for the alternative system . For example, the alternative system was considered to better achieve economic goals with a measure of 0 .20 (zero is the midpoint) when the current system resulted in less tax . When the alternative system resulted in less tax this measure increased to 1 .44 in favor of the alternative system . This suggests that the effect of self-interest goes well beyond the perceptions of fairness .


Table 5 .

17

Means for each independent variable' at the different levels of self-interest (SI) . (n = 171) Social Run Economic Effort Loopholes Gains Government Goals

Financial Self-Interest

Current system results in less tax paid Tax paid is the same under both systems Alternative system results in less tax paid

-0 .88*

0.65*

0 .20*

2 .09

.25 1

-0 .11 .

0 .84

0 .84

1 .84

1 .43

1 .32*

1,44*

2 .68*

2.29*

1 .08*

' Only the actual measure of which system better achieves these measures is considered here (i .e . the policy objectives are not weighted by importance). Therefore, the range for all of these variables is from -3 to 3 . * The mean is significantly different from the other means (in same column) at the 0 .05 level .

Limitations

All of the common issues resulting from a survey method apply to the current study . For example, demand effects always are a concern with survey research . The order of the questions in our survey instrument was intended to reduce this effect to the extent possible . A concern in this study is a bias created by the difference between national demographic averages and the average demographics of the respondents in this study . Participants in this study tended to be older, more male and more educated than national averages . It should also be noted that data used in this study was collected in 1996 . Thus, all generalizations are made within these limitations .

CONCLUSIONS Equity as a desired characteristic of a tax system goes back at least as far as 1776, where it is described in Adam Smith's "Wealth of Nations" (1935) . Current debate over tax reform centers around the concept of fairness . Therefore, it is crucial that we understand how the perceptions of fairness are affected by other desired objectives of a tax system (e.g . desired social and economic goals or simplicity) . The purpose of this research was to determine the factors that individuals consider when making judgments about the comparative fairness of two alternative tax systems, the current federal income tax system and a "flat"

18


tax system . Respondents' perceptions of which tax system was fairer was the dependent variable of interest . This study started with variables derived from equity theories (Leventhal, 1980) that were expected to influence perceived fairness . Distributive justice variables were represented by policy objectives of the tax systems (Kinsey, 1989) and procedural justice variables were represented by perceptions of a tax system's unjustified complexity (Carnes & Cuccia, 1996) . While respondents overwhelmingly considered the alternative "flat" tax system as better at achieving the policy objectives, only the system's ability to achieve economic goals had a significant effect on the perceptions of the relative fairness of the two tax systems . This suggests that perhaps taxpayers do not consider policy fairness as Kinsey (1989) suggested . Following Sheffrin's (1994) argument, it may not be economical for taxpayers to spend the time to be able to adequately determine how well a tax system achieves various goals . The two examples of unjustified complexity we considered were the amount of effort required to comply with the tax system and the amount of loopholes in the tax system . Again, respondents favored the alternative flat tax system on both of these dimensions . Additionally, both variables were found to have a significant effect on the respondents' perceptions of which tax system was more fair . This finding is consistent with expectations derived from Carnes and Cuccia (1996) . Additionally, effort necessary was found to be more influential for respondents who believed the tax system need not be very complex . Consistent with Wartick (1994) and Hite and Roberts (1992), self-interest had a significant affect on the fairness judgments of the tax systems . Further, self-interest also appeared to affect other judgments as well . For example, those made better off financially by the alternative tax system thought it was easier to comply with than those who were not made better off . An objective judgment of the ease of compliance should be independent of financial effect . However, the results indicate that self-interest is a driving force, not just for fairness judgments, but for other tax system related judgments as well . This finding has implications for policy makers . No matter what the touted or designed advantage of a new system, the perceptions of that system will likely be highly influenced by the financial effect on winners and losers .

Future Research

There are important components of the tax system that were not considered in this study ; the IRS, the political system that creates the tax rules, and the level of government spending . It would be useful to know if and how taxpayers'


19

fairness judgments are influenced by these and other features of the tax system . Based on comments written by respondents (and anecdotal observation), the level of government spending is a "hot" topic . Most respondents believed that the government spends way too much ." Porcano and Price (1992) also suggested that dissatisfaction with the government and administration of the tax system may be affecting fairness judgments . Taxpayers may be taking a utilitarian approach to their taxes in an effort to balance their terms of trade . In this study, different revenue collection systems were compared, while the spending side of the exchange relationship was held constant . Government spending could fit into the theoretical model in two places . First, the political process that determines the collection and expenditures of the government is a procedural justice issue . Second, if there is a prevalent belief that government spending is too high, taxpayers may believe that any decrease in their tax burden also will mean a reduction in government spending . These issues need to be studied further.

NOTES I . Hamilton (2000) summarizes a tax reform summit held by Bill Archer, R-Texas in April . 2000 . A key conclusion was that tax code complexity is a "cross-cutting problem" affecting not only taxpayer compliance costs but also affecting issues related to IRS oversight . 2. A flat tax was selected for comparison for two reasons . First, several flat tax proposals (Armey-Shelby, Gephardt & Forbes) have received attention in both Congress and the press (e .g . Lyons, 1996 ; Calegari et al ., 1996 ; Glenn, 1996 ; Tax Notes July 8, 1996 ; Donmoyer & Lyons, 1996 ; Los Angeles Times, 9/24/95) . Second, it would be problematic to have people attempt to determine the financial effect of a national sales tax relative to an income tax . 3 . Carnes and Cuccia (1996) put all of these in the category of "excessive computations" . 4. Any tax reform that leaves the IRS intact would continue to be subject to the same administrative concerns as the current system . The operation of the IRS has come under scrutiny of late as evidenced by the IRS Restructuring and Reform Act of 1998 . Nevertheless, in the present study we hold the administrative process and the government itself constant . Therefore, the primary procedural justice issue is complexity . 5 . Herek's approach is consistent with earlier functional theory of attitude researchers such as Katz (1960) . 6. Subjects were selected based on a stratified sample . The sample was obtained from a commercial firm, Compilers Plus, which provided a stratified sample of households both above and below the median U .S . household income (approximately $35,000) . The National Consumer Household Database from Compilers Plus is built from "many sources" including, telephone directories, census data, drivers license records, real estate transactions, mail order information, questionnaires and warranty cards, change of address information, lifestyle data and other public information .

20


7 . Based on pilot test results, above-median income households were more likely to respond to the questionnaire than below-median income households . Increasing the number of questionnaires sent to below-median income households lowered the overall response rate, but provided a better split of respondents from below- and above-median income households . 8 . Response rates of some tax research which at least partially relied on Dillman's method include Hite and Roberts (1991) 66%, Collins et al . (1990) 37%, and Gerbing (1988) 56% . A study by Copeland and Harmelink (1995) had just a 5 .5% response rate, however they did not institute any follow-up procedures . The questionnaire used this study was somewhat effortful since it was ten pages long and required some reading and comprehension . Therefore, a very high reponse rate (like Hite & Roberts) was not expected, however with the use of Dillman's follow-up methods, a response rate of 20% was acheived . (See Table 2 for details .) 9 . Since the questionnaire was to be completed by the member of the household that is responsible for completing the tax return, these deviations from population demographics are not unexpected . 10 . Based on pilot test results, the majority of those exposed to the 20% tax rate believed they would pay less with the alternative system . Therefore, using an additional 25% rate provided a more-even distribution between those who believed they would pay less and those who believed they would pay more with the alternative tax system . 11 . Results of the analysis are similar when these variables are not weighted . 12 . Respondents were asked to rate the level of federal government spending on a scale of 1-7, with 1 = way too little and 7 = way too much . The mean response was 6 .45 .

ACKNOWLEDGMENTS We would like to thank the members of the first author's dissertation committee, Sandra Kramer, Jack Kramer, Bill Messier and Steven Shugart . In addition to the editor and two anonymous reviewers, we would like to thank Joel Demski, Pierre (Jinghong) Liang, Anne Christensen, Robin Roberts and participants at a University of Florida workshop and the 1999 AAA Western Regional Meeting for their helpful comments .

REFERENCES Adams, J. S . (1963) . Toward an understanding of inequity . Journal of Abnormal and Social Psychology, 67, 422-436 . Adams, J . S . (1965) . Inequity in social exchange . In : L. Berkowitz (Ed.), Advances in Experimental Social Psychology, Vol . 2, 267-299 . Andrews, I. R . (1967) . Wage inequity and job performance : An experimental study . Journal of Applied Psychology, 51, 39-45 . Calegari, M ., Key, K ., & Smith, J . (1996). Flat tax ramifications for self-employed taxpayers . Tax Notes. July 29, 641-646 .


21

Carnes, G . A., & Cuccia, A . D . (1996) . An analysis of the effect of tax complexity and its perceived justification on equity judgments . The Journal of the American Taxation Assiciation, 18 (Fall), 40-56 . Christensen, A . L ., & Weihrieh, S . G. (1996) . Tax fairness : different roles, different perspectives . Advances in Taxation, 8, 27-61 . Christensen, A . L ., Weihrich, S ., & Newman, M . (1994) . The impact of education on perceptions of tax fairness . Advances in Taxation, 6, 63-94 . Collins, J . H ., Millron, V . C ., & Toy, D . R . (1990) . Determinants of tax compliance : A contingency approach . The Journal of the American Taxation Association, 12 (Fall), 1-29 . Copeland, P . V ., & Harmelink . P . J . (1995) . Using taxpayers perceptions of fairness to redesign the federal income tax structure . Advances in Taxation . 7, 43-72. Dillman, D . A . (1978) . Mail and Telephone Surveys : The Total Design Method . New York : John Wiley and Sons, Inc . Donmoyer, R ., & Lyons, L . (1996) . After a lull, tax reform debate heats up again in Washington . Tax Notes . June 17, 1996, 1588-1591 . Eriksen, K ., & Fallan, L. (1996) . Tax knowledge and attitudes toward taxation : A report on a quasiexperiment . Journal of Economic Psychology, 17, 387-402 . Etzioni, A . (1986) . Tax evasion and perceptions of tax fairness : A research note. The Journal ol Applied Behavioral Science, 22, 177-185 . Garland, H . (1973) . The effects of piece-rate underpayment and overpayment on job performance : A test of equity theory with a new induction procedure . Journal of Applied Social Psvchology, 3, 325--334 . Gerbing, M . D . (1988). An empirical study of taxpayer perceptions of fairness . Working Paper. Lewis and Clark College, Portland, Oregon . Glenn, H . (1996) . Flat tax would sting low-income families, study finds . Tax Notes, July 15, 1996, 275-276 . Graetz, M . J . (1997) . The Decline [and Fall?] of the Income Tax System . New York, NY : W .W . Norton & Company . Hamilton, A . (2000) . Tax reform summit: Tax simplification and reform commission takes shape . Tax Notes, April 17, 2000, 314, Herek, G . M . (1986), The instrumentality of attitudes : Toward a neofunctional theory . Journal of Social Issues, 42, 99-114 . Hite, P . A ., & Roberts. M . L . (1991) . An experimental investigation of taxpayer judgments on rate structure in the individual income tax system. The Journal of the American Taxation Association, 13 (Fall), 47-63 . Hite, P . A ., & Roberts . M . L . (1992) . An analysis of tax reform based on taxpayers' perceptions of fairness and self-interest . Advances in Taxation, 4, 115-138 . Kaplan, S ., & Rockers, P. (1985) . A study of tax evasion judgments . National Tax Journal, 38, 97-102 . Katz, D . (1960) . The functional approach to the study of attitudes . Public Opinion Quarterly. 24. 163-204 . Kinsey, K . A . (1989) . The meaning of income tax fairness . Working Paper . The American Bar Foundation, Kinsey . K . A ., Grasmick H . G., & Smith . K . W . (1991) . Framing justice: taxpayer evaluations of personal tax burdens . Law and Society Review, 25, 845-873 . Leonard, R . J . (1996) . Lessons from 1986 : Hot buttons and third rails . National Tax Journal, 49, 437-445 .

22


Leventhal, G. S . (1980) . What should be done with equity theory? New approaches to the study of fairness in social relationships . In : Gergen, Greenberg & Hills (Eds), Social Exchange: Advances in Theory and Research (pp. 27-55) . Plenum Press. Leventhal, G . S ., Karuza J., & Fry, W . R . (1980) . Beyond fairness: A theory of allocation preferences. Justice and Social Interaction (pp . 167-218) . New York : Springer-Verlag . Los Angeles Times . (September 24, 1995) . Flat tax is gaining respect. Lyons, L . (1996) . Economists agree tax reform within reach, differ on how to get there . Tax Notes, September 23, 1996, 1591-1594 . Maroney, J. J., Rupert, T . J ., & Anderson . B . H . (1998) . Taxpayer reaction to perceived inequity : An investigation of indirect effects and the equity-control model . The Journal of the American Taxation Association, 20 (Spring), 60-77 . Messick, D ., & Sentis . K. (1983) . Fairness, preference, and fairness biases . Equity Theory . New York . Praeger Publishers . Milliron, V. C. (1985). An analysis of the relationship between tax equity and tax complexity . The Journal of the American Taxation Association . 7 (Fall), 19-33 . Moser, D. V ., Evans IE, J . H ., & Kim, C. K. (1995) . The effects of horizontal and exchange inequity on tax reporting decisions . The Accounting Review, 70(4), 619-634 . Neter, J ., Wasserman, W ., & Kutner, M . H . (1990) . Applied Linear Models . (3rd ed .) . Homewood, IL : Irwin . Porcano, T . M . (1984) . Distributive justice and tax policy . The Accounting Review (October), 619-636. Porcano, T. M ., & Price, C. E . (1992). Some evidence on the assocation between judgment criteria and fairness perceptions . Advances in Taxation, 4, 183-210 . Pritchard, R . D ., Dunnette, M . D ., & Jorgenson. D . O . (1972) . Effects of perceptions of equity and inequity on worker performance and satisfaction . Journal of Applied Psychology, 56, 75-94 . Roberts, M . (1994). An experimental approach to changing taxpayers' attitudes towards fairness and compliance via television . The Journal of the American Taxation Association, 16 (Spring), 67-86 . Scott, W . J ., & Grasmick . H . G . (1981) . Deterrence and income tax cheating : testing interaction hypotheses in utilitarian theories . The Journal of Applied Behavioral Science, 17, 395-408 . Sheffrin, S . (1993) . What does the public believe about tax fairness? National Tax Journal, Vol . 46, 301-308 . Sheffrin, S . (1994) . Perceptions of fairness in the crucible of tax policy . In : J . Slemrod (Ed .), Tax Progressivity and Income Inequality (pp . 309-340). Cambridge University Press . New York . Smith, A . (1937) . The Wealth of Nations . New York : Random House, Modern Library, 777-779 . Spicer, M . W ., & Becker . L . A . (1980) . Fiscal inequity and tax evasion : An experimental approach . National Tax Journal, 33, 171-175 . Spicer, M . W ., & Lundstedt. S . B . (1976) . Understanding tax evasion . Public Finance, 31, 295-305 . Thibaut, J ., & Walker, L . (1975) . Procedural Justice: A Psychological Analysis. Hillsdale, NJ: Lawrence Erlbaum Associates . Wartick, M. L. (1994) . Legislative justification and the perceived fairness of tax law changes : A referent cognitions theory approach. The Journal of the American Taxation Association, 16 (Fall), 106-123 . Weiss, R . D . (1996) . The tax reform act of 1986 : Did congress love it or leave it? The National Tax Journal, 49, 447-459 . Tax Notes . (1996) . An interview with Steve Forbes . Tax Notes, July 8, 1996, 165-173 .

The

Effect of

Policy Objectives, Complexity and Self-Interest

23

White, R., Curatola, A ., & Samson, W . (1990) . A behavioral study investigating the effect of knowledge of income tax laws and tax policy on individual perceptions of federal income tax fairness . Advances in Taxation, 3, 165-185 . Worsham, R . (1996) . The effect of tax authority behavior on taxpayer compliance : a procedural justice approach . Journal of the American Taxation Association, 18 (Fall), 19-39 .

APPENDIX Description

of Alternative

Tax System

Some people think a simpler tax system might be better . One simpler system is described below . We'll call this system, Tax System A . Tax System A would collect about the same amount of money as the current income tax system . However, in general, with Tax System A most individuals would pay less in federal income tax than with the current system, but most businesses would pay more . With Tax system A individuals will pay 20% tax on their salary, wages and money received from pensions . Businesses (both corporations and small businesses) will also pay 20% tax on net business income . Net business income is total receipts less total expenses . Other sources of income including interest income, dividend income and capital gains are not taxed no matter if received by an individual or business . All itemized deductions (home mortgage interest, state and local taxes, charitable deductions, medical expenses, etc .), and tax credits (childcare credit, earned income credit, etc .), will be eliminated . Instead there will be a personal allowance . An example of how one family's tax return would look using tax system A follows Example of Tax System Tax return

Jim and Mary Johnson are both employed full-time . In 1995, Jim received wages of $33,000, and Mary received wages of $26,000 . They had $2,000 of interest income . They have 2 children and during 1995, they paid $8,000 in interest on their home mortgage and $4,000 in state and local taxes . Using Tax System A their income would be computed as follows and only this form would be filed:

24


Tax System A Tax Return I . Wages, salary and distributions from pensions 2 . Personal Allowance (choose the same one you claim with the current system) (a) $20,000 for married filing joint (b) $14,000 for head of household (c) $10,000 for single 3 . Personal Allowance for Dependents Number of Dependents, not including you or your spouse, two times $5,000 4 . Total personal allowance (line 2 plus line 3) 5 . Taxable compensation (line I less line 4) 6 . Net Business Income (from a business you own, this can be less than 0 if you had a loss) 7 . Total Taxable Income (line 5 plus line 6 but not less than 0) 8 . Tax (20% of line 7)**

$59,000-

$20,000

$10,000 $30,000 $29,000 $0 $29,000 $5,800

To help you compare Tax System A with our current system : with our current system, the Johnsons would have completed three tax forms : (1) Form 1040; (2) Schedule A, for itemized deductions ; and (3) Schedule B, for interest income ; and they would have paid $5,850 in federal income taxes for 1995 . With Tax System A, however, they would have filed only the tax form shown above and would have paid $5,800 in federal income tax . * Note : wage income was manipulated betwen respondents so that the example viewed by respondents would be more similar to their own income levels. Individuals sampled from the income group higher than the median household income read the preceding example . Individuals sampled from the income group below the median household income read the following example .

Example of Tax System A Tax Return

Mary Johnson, a widow is employed full-time . In 1995, Mary received wages of $22,000 . She and her teenage son live in an apartment she rents . Under the current tax system, Mary qualifies for the earned income credit . Using Tax System A their income tax would be computed as follows and only this form would be filed:

Tax System A Tax Return I . Wages, salary and distributions from pensions 2 . Personal Allowance (choose the same one you claim with the current system) (a) $20,000 for married filing joint (b) $14,000 for head of household (c) $10,000 for single 3 . Personal Allowance for Dependents Number of Dependents, not including you or your spouse, one times $5,000

$22,000

$14,000

$5,000

The Effect of Policy Objectives, Complexity and Self-Interest 4 . Total personal allowance (line 2 plus line 3) 5 . Taxable compensation (line I less line 4) 6 . Net Business Income (from a business you own, this can be less than 0 if you had a loss) 7 . Total Taxable Income (line 5 plus line 6 but not less than 0) 8 . Tax (20% of line 7)"

25 $19,000 $3,000 $0 $3,000 $750

To help you compare Tax System A with our current system : with our current system Ms Johnson would have completed two tax forms : (1) Form 1040; (2) Schedule EIC, for the earned income credit : and she would have paid $1,310 in federal income taxes for 1995 . With Tax System A, however, she would have filed only the tax form shown above and would have paid $750 in federal income tax . *" Note : Tax rate was manipulated between respondents to increase the variability of the financial effect of the alternative tax system . Some respondents saw a tax rate of 25% .

TAX SUBSIDY INCREASES IN ANTICIPATION OF TAX RATE CHANGES Bradley D . Childs

ABSTRACT Corporate tax rate changes are an incentive for high marginal tax (HMT) firms to shift taxable income into a period of lower tax rates . The Tax Reform Act of 1986 (TRA 86) reduced the top corporate statutory rate for ordinary income from 46% to 34%, and increased the corporate rate on net capital gains (NCGs) from 28% to 34% . Given the effective dates of TRA 86, HMT firms had an incentive to delay recognition of ordinary taxable income from 1986, and these firms had an incentive to accelerate NCGs into 1986. Prior research has shown that HMT firms shifted book-tax conforming income away from 1986 . The present study finds that HMT firms shifted non-conforming income (i .e . the difference between pre-tax income and taxable income) into 1986 . It also finds that HMT firms shifted NCGs (a tax subsidy due to preferential tax rates) into 1986. This evidence of shifting behavior suggests that implicit tax costs, as well as other cost disincentives, were insufficient to offset the tax rate change incentives .

Advances in Taxation, Volume 13, pages 27-47 . Copyright © 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0-7623-0774-9 27

28

BRADLEY D. CHILDS

1. INTRODUCTION The Tax Reform Act of 1986 (TRA 86) changed corporate tax rates, reducing the top ordinary statutory rate from 46% to 34% over two years, and increasing the rate on net capital gains (NCGs) from 28% to 34% . Both rate changes took effect in 1987 . High marginal tax (HMT) firms, the natural tax clientele, had an incentive to delay the recognition of ordinary taxable income from 1986 . For NCGs, HMT firms had an incentive to accelerate the recognition of this income into 1986 . 2 Prior research (Scholes, Wolfson & Wilson, 1992 ; Guenther, 1994) has shown that HMT firms decreased book-tax conforming income in 1986 . Conforming income is income that is the same for book and tax purposes . A primary cost of decreasing conforming income for tax savings is the cost of reporting reduced financial income . The prior evidence suggests that the cost of reporting reduced financial income is less than the benefit of the tax savings generated by this tax strategy . The present study extends previous work by investigating the tax strategy of increasing non-conforming income in 1986 in anticipation of the tax rate changes enacted by TRA 86 . Non-conforming income (NCI) is computed by subtracting estimated taxable equivalent income from book pre-tax income (PTI) . Taxable equivalent income is estimated by dividing current tax expense by one minus the top statutory rate . Increasing NCI at a time when PTI is stationary or declining reduces taxable income . NCI is the pre-tax equivalent of a tax subsidy . This is so because NCI is an accumulation of tax preferences, and most tax subsidies are calculated as tax preferences times the tax rate . One tax subsidy that is computed differently is the capital gain differential (CGD) . This is computed by multiplying the NCG realization, which is often conforming income, by the difference in ordinary and capital gain tax rates . A tax strategy of increasing NCI (i .e . tax subsidies) incurs a major potential cost in the form of implicit taxes (e .g . a higher bidding price for a tax-favored asset) . Therefore, the present study explores the joint hypothesis that HMT firms will use the strategy of increasing NCI in 1986, and that the tax savings achieved outweigh the associated cost of implicit taxes . This study finds that HMT firms increased their tax subsidies, including the CGD, in 1986 in anticipation of the tax rate changes . This study can be distinguished from other studies not by the effect (reducing taxable income in 1986), but by the strategy employed (shifting NCI vs . conforming income) and by examining the cost disincentives toward executing the tax strategy .

Tax Subsidy Increases in Anticipation of Tax Rate Changes

29

The remainder of this paper is organized as follows . Section 2 reviews the literature on corporate taxable income shifting, implicit taxes and corporate capital gain realizations . Section 3 develops the alternative hypotheses and discusses factors in favor of the null hypotheses . Section 4 explains the research design . Section 5 presents descriptive statistics and test results . Section 6 concludes the paper, and offers suggestions for further research .

2. LITERATURE REVIEW Corporate Taxable Income Shifting

In the intertemporal tax literature, the sample firms are typically assumed to be HMT firms . It also is assumed that the shifting of PTI is accompanied by substantial taxable income shifting, inferring that a substantial level of booktax conformity is present . The HMT assumption is considered reasonable because the inclusion of low marginal tax (LMT) firms would tend to bias the test results in favor of the null hypotheses .' The conformity assumption also is considered reasonable because there is no obvious reason why HMT firms would systematically want to decrease PTI without a corresponding decrease in taxable income . Scholes et al . (1992) investigated intertemporal income shifting in anticipation of rate changes by observing whether firms decreased gross margins (normally, conforming income) in the fourth quarter of a fiscal year prior to a rate decrease . Since TRA 86 phased in the rate reductions over two years, the majority of firms had two fiscal quarters that were tested . Their study found that gross margins were decreased as expected in quarters prior to a rate reduction . However, it found that the smaller firms in the study's sample did not demonstrate this behavior, which suggests that some firms did not aggressively minimize taxes . Guenther (1994) investigated whether HMT firms made income-decreasing accruals prior to a rate decrease . His study based on TRA 86 found evidence that such accruals were made by calendar year firms, but not by the fiscal yearend (May-July) firms . The fiscal year-end finding was anomalous because these firms had a greater tax rate incentive for making the accruals than the calendar year firms . 4 Lopez et al- (1998) offered an explanation for this anomaly by introducing a tax subsidy measure that partitioned Guenther's (1994) sample based on prior taxpayer aggressiveness . They found that firms with a history of high tax subsidy

30

BRADLEY D . CHILDS

use in the 1983-1985 period made greater income-decreasing accruals than did other firms . Maydew (1997) examined whether firms that had net operating loss (NOL) refund potential in the years following TRA 86 increased their book losses in order to increase NOL carrybacks into years with higher tax rates . He found that the natural tax clientele for this tax strategy decreased gross margins and increased selling and administrative expenses in the year of the NOL . Implicit Taxes

Under implicit tax theory, shifting tax subsidies implies shifting implicit taxes . In the implicit tax model (Scholes and Wolfson 1992), competitive bidding by the appropriate tax clientele drives up the prices of tax-favored investments. For example, HMT insurers are a natural tax clientele for municipal bonds. Their bidding is exected to reduce the pre-tax return realized on these taxfavored investments . The tax subsidy computed on this investment will be the amount of tax-exempt municipal interest multiplied by the top statutory rate. The implicit tax for this investment will be the difference between tax-neutral interest from an equivalent investment and the tax-exempt interest . When the assumptions of the model are fulfilled, the after-tax return from owning municipal bonds will be equal to the after-tax return from owning tax-neutral bonds .' Under these circumstances, implicit taxes will equal the pre-tax equivalent of the tax subsidy (PTTS) . Wilkie (1992) and Callihan and White (1999) have used this measure to estimate implicit tax burdens .' When a firm has implicit taxes equal to PTTS, that firm will not have an incentive to shift tax subsidies, because the tax savings generated by the tax subsidy will be offset by implicit tax costs . Both Wilkie (1992) and Callihan and White (1999) have found the PTTS variable to be negatively related to pre-tax returns . While this relation is consistent with implicit tax theory, nevertheless, each paper shows an area where implicit taxes may not be present . For instance, Wilkie (1992) did not always find this relation for implicit taxes associated with timing differences . He attributed this to measurement errors, such as the lack of present value accounting for deferred taxes . Callihan and White (1999) found lower implicit tax costs for market leaders and concentrated industries . They suggested that implicit taxes incurred by firms are lower in less competitive market structures . Capital Gains

Only NCGs, the excess of long-term capital gains over short-term capital losses, are eligible for preferential capital gain tax rates . Many articles in this area


31

(Slemrod & Feldstein, 1978 ; Geiger & Hunt, 1989 ; Repetti, 1990; Johnson, 1992) note that capital gains are not taxed as they accrue . Investors that hold on to capital gain assets experience declining real rates of taxation because capital gain taxation is transaction-based . The incentive to hold appreciated capital assets when other investments provide higher pre-tax yields is known as the "lock-in effect ." When capital gain tax rates decrease the "lock-in effect" is diminished, and when rates increase the "lock-in effect" is enhanced . Porcano and Schull (1997) have investigated the taking of NCGs by different taxable entities over time . From an analysis of Statistics of Income (SOI) data published by the Internal Revenue Service, they find that corporations realize more (less) NCGs when capital gain tax rates decrease (increase) . Porcano (1997) also examined the ways that corporations can realize NCGs : (1) investment dispositions ; (2) property, plant and equipment disposition ; (3) line-of-business dispositions ; and (4) extraordinary transactions . He finds the most significant responses occurred in reaction to the capital gain tax rate increases enacted by TRA 86 . Other tax legislation produced mixed results . The NCG testing in this study can be distinguished from the testing in these studies in several ways . In Porcano and Schull (1997) the data are aggregated and they are not related to measures of profitability . In this study the data are not aggregated and some of the prediction models are based upon expected profitability . In Porcano (1997) the tested variables may not be proportionately related to NCGs over time and across firms . In addition, both studies examined corporate behavior after the tax rate change, whereas this study examines corporate behavior in anticipation of the tax rate change .

3 . HYPOTHESES DEVELOPMENT Non-Conforming Income

The anticipated ordinary tax rate reduction in TRA 86 was an incentive for HMT firms to delay recognition of ordinary taxable income in 1986 . One strategy for accomplishing this objective would be to maximize tax subsidies in 1986 . The following hypothesis tests this strategy (all hypotheses are stated in the null form) : HI : Tax subsidies of HMT firms in 1986 (after adjustment for bias and distortion) will be no different than as predicted by tax subsidy models using data from 1983-1985 . Nor will the prediction errors of HMT firms be different from those of LMT firms .

32

BRADLEY D. CHILDS

Finding evidence of this corporate behavior may be problematic because of tax subsidy exhaustion, implicit tax theory and tax audits of non-conforming income . Each of these factors favors the null hypothesis and is discussed in turn . Lopez et al . (1998) noted that several firms may have reached their maximum ability to utilize existing tax subsidies, which is why their tax subsidy measure was a dummy variable . All else equal, and without consideration of the other factors that inhibit tax subsidy utilization, an HMT firm presumably prefers more tax subsidies than less . Guenther et al . (1997) find evidence that firms prefer high NCI (i .e . tax subsidies) levels' Thus, HMT firms already maximizing tax subsidies may not have any further ability to increase current subsidy levels . Under certain assumptions, implicit taxes may be great enough to make an HMT investor indifferent to tax subsidy utilization .' Asset prices of tax-favored investments under implicit tax theory are related to tax rates . Lower tax rates decrease asset prices because tax subsidies are less valuable in a low tax rate regime . The effect of changing asset prices depends upon the specific tax subsidy . For example, a long-term investment in municipal bonds in 1986 (assuming knowledge of TRA 86) might generate a higher tax subsidy in 1986, because the tax-exempt interest income will be less valuable after 1986 . Conversely, the tax subsidy from purchased depreciable assets in 1986 may decline in 1986, because the accelerated depreciation in the tax code will be less valuable after 1986 . Non-conforming income (NCI) for a year is potentially subject to that year's tax rates . Mills (1994) reported that NCI was associated with higher tax audit adjustments . Cloyd et al . (1996) found that firms taking aggressive tax positions changed financial accounting positions to conform to their tax positions, but these findings do not apply uniformly to NCI . For example, firms do not need to defend accelerated depreciation deductions by converting financial straightline depreciation to accelerated depreciation . Net Capital Gains The anticipated capital gains tax rate increase was an incentive for HMT firms with capital gain appreciation to increase their CGD tax subsidy in 1986 . The following hypothesis tests the shifting of the CGD tax subsidy into 1986 . H2 : The CGD tax subsidies of HMT firms in 1986 will be no different than as predicted by CGD prediction models using data from 1983-1985 . Nor will the prediction errors of HMT firms be different from those of LMT firms . Again, there are several reasons why finding this increase may not occur . These reasons are implicit tax costs, mixed income assets and discontinued operations . These factors favor the null hypothesis and each is discussed in turn .


33

The increased capital gain tax rates made capital assets less tax-favored . Under implicit tax theory, less tax-favored investments have lower prices, which in turn decrease the recognized gains . These decreased gains may work against an increase in the number of realizations . This conflict between an increase in the number of realizations and a decrease in the average gain makes uncertain the overall NCG effect (Ricketts & White, 1997) . However, lower prices will not affect all capital assets . For instance, capital assets that are extinguished by recognition, or those assets that are not subject to capital gain taxation by their new owners, may not experience any price declines due to implicit taxes .9 Ricketts and White (1997) summarize studies dealing with capital gain tax rate changes and revenue realizations for individual investors . They conclude that micro-level studies generally find an increase(decrease) in the amount of gain realizations following a decrease (increase) in the capital gains tax rate . Some assets, such as depreciable property used in a business, have their taxable gain characterized as part ordinary and part capital ." The incentives affecting the timing of these asset dispositions are affected by both TRA 86 rate change incentives . The ordinary tax rate reduction and the capital gain tax rate increase create opposite incentives so that a firm's overall incentive will depend on the relative mix of the gains . Thus, some NCGs that would otherwise have been realized in 1986 may be deferred until later years as a result of TRA 86 . The tax rate reconciliation, which disclosed CGD data, is reconciled to pretax income from continuing operations . If a firm's capital gain arises from discontinued operations, this will not be reflected in the tax rate reconciliation .

4. RESEARCH DESIGN Due to on-going tax legislation, total tax subsidy levels are rarely comparable from year to year . The research design of this study accounts for this lack of comparability by removing individual tax subsidies that may be biased or distorted between the control period, 1983-1985, and the test period, 1986 . Removing biased/distorted tax subsidies requires sufficient disclosure that can only be found in the tax footnotes of annual reports . This data requirement is a major constraint on sample size . In addition, tax subsidies usually are not distributed normally . To counter this, the research design uses nonparametric tests to compare actual adjusted tax subsidies with predicted adjusted tax subsidies ."

34

BRADLEY D . CHILDS

Variables Major tax subsidies remaining in the adjusted tax subsidy variable include accelerated depreciation, possessions tax credits, research tax credits, dividend income deductions, percentage depletion, goodwill amortization and tax exempt interest . Table 1 lists tax subsidies that were not included and the reasons for their exclusion . Panel A lists excluded tax subsidies that, without tax rate change incentives, would have been biased toward decreasing NCI . Panel B lists additional excluded tax subsidies because they are non-federal in nature or they distort the computation . Non-federal tax subsidies are the foreign tax rate differential and the net effect of state income taxes . Distorting tax subsidies are net operating loss carryforwards and tax audit adjustments ." The adjusted tax subsidy variable does not include the CGD tax subsidy because it is tested separately .

Sample Screening out the excluded tax subsidies and identifying the CGD tax subsidy required a hand-collected sample derived from the following conditions/ constraints : (1) calendar year firms on Compustat (n = 3583) ; (2) extraordinary items and discontinued operations are less than current tax expense for each year, 1983-1986 (n = 908) ; (3) annual report on 10-K on NAARS for 1983-1986 (n = 662) ; (4) full disclosure of the "deferred tax detail" and the "reconciliation of effective tax rates" for 1983-1986 (n = 322) ; and, (5) positive PTI, positive domestic ordinary PTI and non-zero federal current taxes for 1983-1985 and 1986 (n = 216) . The first condition holds the tax rate in the test period, 1986, equal to the tax rate in the base period, 1983-1985 . 13 The next condition minimizes the impact of items reported net of tax . The third and fourth constraints permit the data collection of the tax subsidies . The final condition defines HMT firms . 14 The positive domestic ordinary PTI screen overcomes the problem of whether net operating loss carryforwards are foreign or domestic . Thus, 322 are used in this study : 216 firms in the HMT sample and 106 firms in the LMT sample .

Tax Subsidy increases in Anticipation of Tax Rate Changes Table 1 .

35

Excluded Tax Subsidies .

Panel A Bias Direction in 1986

Reason for Bias

Investment Tax Credit [ETR Recon]

Reduced

Repeal

1 .1 .85

TRA 86

Safe Harbor Leasing [DT Detail]

Reduced

Repeal

7 .1 .82

TEFRA 82

International Sales Income [Both]

Reduced

Conversion

1 .1 .85

DEFRA 84

Restructuring (Other Long Term Accruals) [DT Detail]

Reduced

Economic Performance Rules

7 .1 .84

DEFRA 84

Completed Contract IDT Detail]

Reduced

Restricted

9 .26 .86

TRA 86

Installment

Reduced

Restricted

9 .26 .86

TRA 86

Tax Subsidy [Location]

Effective Date

Tax Legislation

Sales [DT Detail]

Panel B Tax Subsidy

Distortion Reason

State income taxes (net) [ETR Recon .]

Non-federal tax

Foreign tax rate differential IETR Recon .]

Non-federal tax

Net operating loss carryforwards [Both]

Tax subsidy status is disputed

Audit adjustments [ETR Recon .]

Adjustments concern prior tax years

HTR Recon . - effective tax rate reconciliation in tax footnotes DT Detail - itemized detail of the deferred tax expense in tax footnotes

36

BRADLEY D. CHILDS

Prediction Models for Adjusted Tax Subsidies (ATS) Random Walk Model The definitive behavior of tax subsidies is unknown, but prior research does reflect on this matter. For example, Wilkie (1988), in his analysis of pre-tax income changes from 1980-1984, reports that 47% of paired pre-tax income changes and tax preference changes have opposite signs. This finding is not incompatible with a random walk model as shown in equation (1). ,~TSi86 = ATSiB,~e

( 1)

where, ATSi86

= predicted adjusted tax subsidies for firm i for 1986; and,

ATSiBa~¢ = adjusted tax subsidies for firm i for 1983-1985 (adjusted tax subsidies are total tax subsidies less excluded tax subsidies).

Portfolio regression model Shevlin and Porter (1992) find that NCI is positively related to PT1. They also find a positive intercept for a regression of NCI on PTI. Partitioning PTI leads to the second prediction model. ,~TSi86 = ap+bp DOPTIis6+C p FPWli86

(2)

where, DOPTII86 = PTIis6-CGDI86/(0.46-0.28 )-FPTIIs6; CGDi86 = capital gain differential; FPTIi86 = foreign PTI; and, ap, bp, cp -- portfolio coefficients from estimating a regression equation that treats domestic ordinary PTI and foreign PTI as regressors and adjusted tax subsidies from 1983-1985 as the regressand, t5

Proportionate investment In theory, many tax subsidies are directly unrelated to PTI. The firm's expenditures rather than the firm's eventual PTI determine the tax subsidies. A basis for estimating a firm's potential for expenditures is the firm's stockholders' equity as shown in the third prediction model. fi~TSi86]SEi86= ATSiB,s¢/SEiB,~e

(3)


37

where,

SEi86 = stockholders' equity for firm i in 1986 ; and, SEiBaae =stockholders' equity for firm i for 1983-1985 . 1 n Implicit tax theory model Under certain market conditions implicit tax theory holds that the after-tax return of a tax-favored investment will equal the after-tax return of a tax-neutral investment. The resulting prediction model is shown in equation (4) .

ATROE i86 = ATROE 1B8te

(4)

where,

ATROE = predicted after-tax return on equity for firm i for 1986 ; ATROE, Base = [PTli B86eCTE iBafle ]/SE jBase, and, CTE. Baae = current tax expense from 1983-1985 . Prediction models for the capital gain differential - CGD The prediction models for the CGD tax subsidy are the same as the adjusted tax subsidy prediction models except the portfolio regression model is dropped due to reduced observations . The remaining models are shown in equations (5) and (6) . 8GD i86 = CGD. Ba, e

(5)

cGD i86/SE, 86 = CGD. BaYe/SE. Ba4C

(6)

5. DESCRIPTIVE STATISTICS AND TEST RESULTS Descriptive Statistics Table 2 lists descriptive statistics for the base period (1983-1985) and the test period (1986) for the HMT sample and the LMT sample .' The HMT sample shows growth for several median measures . Median investment (stockholders' equity plus deferred taxes) increases 34% ($334 million to $447 million) from the base period to the test period . Pre-tax income increases 31% ($54 .3 million to $71 .1 million) . Overall tax subsidies increase 58% ($6 .13 million to $9 .67 million) and adjusted tax subsidies increase 64% ($4 .02 million to $6 .59 million) .

BRADLEY D . CHILDS

38 Table 2 .

Descriptive Statistics (in millions) .

Panel A

HMT Sample

(n = 216)

Mean

Median

Std .Dev .

Max .

Min .

SE,_ SE,

1443 1591

334 447

3711 4146

32917 35799

5.23 7.12

PT1 aix Fri.

275 280

54 .3 71 .1

884 758

9890 8457

0 .645 0 .157

ATROEBATROE86

0 .133 0 .150

0 .126 0 .133

0.064 0.109

0 .388 0 .646

(0 .006) (0 .046)

TSBTS,,

35 .7 68 .8

6 .13 9 .67

119 258

1010 2514

(506) (417)

ATS,_ ATS,

36.4 51 .5

4 .02 6 .59

145 195

1742 2181

(145) (407)

PATS._ PATS,

5 .31 9.56

0 .455 0 .669 .

31 .9 60 .0

195 569

(238) (453)

TATS a' .' TATS,

31 .1 41 .9

2 .51 4 .56

143 161

1685 1612

(163) (78 .2)

(0 .779) 17 .3

0 .434 0 .221

83 .5 90 .2

483 759

(731) (263)

3 .46 8.58

0 .385 .191 1

10 .9 19 .2

52 .9 95 .0

(10.3) (45 .0)

ETS Ben ET S16 CGDa ., (n = 87) CGD16 (n=87) Panel B (n = 106)

LMT Sample Mean

Median

Std.Dev.

Max .

Min .

894 773

250 163

1962 1999

15717 17068

(262) (1171)

110 21 .9

1 .124 (0.361)

520 356

4630 2993

(418) (805)

ATROE,- (n=94) ATROE ss (n=94)

0.011 (0.069)

0.004 (0.028)

0 .109 0 .218

0.251 0.435

(0 .320) (1 .08)

TS B-

(21 .7) (26 .5)

(0.383) (2.50)

109 91 .9

326 229

(799) (390)

14 .8 1 .31

1 .96 0.102

61 .6 49.3

321 226

(88) (195)

PATS B PATS 86

1 .38 (1 .70)

0.014 (0.006)

25 .1 27.2

167 66 .4

(104) (217)

TATSBen TATS,,

13 .1 3 .01

0.933 0.000

60.1 37.5

330 219

(93) (139)

ETS BETS 86

(36 .2) (27 .8)

(1 .03) (1 .87)

134 91 .7

58 .4 125

(1025) (616)

CGDBase (n=27) CGD86 (n=27)

2 .21 1 .43

0.246 0.159

14.2 8.91

45 .1 21 .8

(26 .1) (25 .4)

SEBSEs6 PTl ag,= PTl s6

TS81 ATS 8.e ATS s6


Table 2 .

39

Continued .

Panel C Variable definition 86 refers to 1986 amount and base is the average of 1983, 1984 and 1985 . For all base ratios, the ratio is an average numerator divided by an average denominator SE - stockholders' equity plus deferred taxes : Compustat(#216+#74) PTI - pre-tax income: Compustat item (#18) CCE - total taxes less deferred tax expense : Compustat item (#16-#50) ATROE - after-tax return IPTI-CTEI/SE TS - tax subsidy [PTI*0 .46-CTE] ETS - excluded tax subsidy (listed in Table I) ATS - adjusted tax subsidy [TS-ETS] PATS - permanent adjusted tax subsidy : permanent tax subsidy [TS-total taxes (#16)] less excluded tax subsidies found in the ETR recon . TATS - timing adjusted tax subsidy: timing adjusted tax subsidy : timing tax subsidy [deferred tax expense (#50)] less excluded tax subsidies found in the DT detail CCD - capital gain differential tax subsidy : found in the ETR recon .

One median measure that does drop for HMT firms is excluded tax subsidies . The 51% decrease from $0.434 million to $0 .221 million is consistent with the predicted bias as described in Table 1 . Notwithstanding the decrease in the median, the excluded tax subsidies average goes from negative $0 .779 million in the base period to a positive $17 .3 million in the test period . This large swing is attributable to a reduction in the foreign tax rate differential experienced by a minority of the largest firms in the HMT sample . The CGD tax subsidy shows an increase in the median measure of more than 300% from the base period to the test period . The median NCG realizations grew from $2 .14 million in the base period to $6 .61 million in 1986 . 19 In contrast, the LMT sample shows reductions for most measures . These reductions reflect the sample's composition . Approximately one-third of the LMT sample had negative income in the base period, another third had negative income in the test period, and the remainder had negative income in both periods ."

Test Results Table 3 reports the results of three of the prediction models (random walk, portfolio regression and proportionate investment) for adjusted tax subsidies . In all three models, the predicted adjusted tax subsidies are less than the realized adjusted tax subsidies consistent with the HI, namely that HMT firms had an

40

BRADLEY D . CHILDS

Table 3 .

Non-parametric Results for Adjusted Tax Subsidies (ATS) .

Panel A

HMT sample n = 216

Prediction model: Random walk ATSits = ATS ieevc ATS, u/SE,a_

ATS, 86 /SE, y .

Mean

0.026

0 .040

Median

0.020

0 .030

Greater than count

84

132

Wilcoxon z-statistic

5 .36-

p-value

0.0000

Prediction model : Portfolio regression model ATSi86 = ap + by DOPT1 86 + cp FPTI. s6 ATS. S6/SE i,_

ATS16/SE,

Mean

0.026

0 .040

Median

0.020

0 .030

Greater than count

87

129


5.23-

p-value

0.0000

Prediction model: Proportionate investment model ATS,,,/SE,,, =ATS, a./SE,a. ATS,/SE Ia_

ATS s6 /SE, a _

Mean

0.026

0 .035

Median

0.020

0 .026

Greater than count

94

122


2 .54*

p-value

0 .0057

41


Table 3 . Continued . Panel B

Excluded firms sample n = 94

Prediction model : Random walk A

ATS,./SE..,,

ATSi86/SEIBeac Mean

0 .023

-0 .013

Median

0.020

0 .008

Greater than count

55

39

Prediction model : Portfolio regression model ATSift6 = ap + by DOPTIifl6 + cp PPTI, B6 ATS .,,/SE,,,,, Mean Median Greater than count

ATS .,,/SE, .,.

0.015

-0 .013

0.010

0 .008

49

45

Prediction model : Proportionate investment model ATS,,,/SE .,, = ATS,,,,,/SE,,,,, ., .,,/SE,, ATS

ATS ,,,/SE

Mean

0.023

-0.015

Median

0.020

0.009

Greater than count

55

39

* Significant at 0.05 level for one-tailed test .Wilcoxon z-statistic for signed-rank test Greater than counts number of times one value of matched pair is greater than the other value ATS (ATS) = predicted (actual)adjusted tax subsidy CGD - capital gain differential tax subsidy : found in the ETR recon . SE = stockholders' equity plus deferred taxes : Compustat (#216 + #74) DOPTI =domestic ordinary pre-tax income : Compustat (#272-CGD/0 .18) PPTI=foreign pre-tax income : Compustat (#273)

42

BRADLEY D . CHILDS Table 4 .

Non-Parametric Results for Capital Gains Differential (CGD) .

Panel A

HMT subsample n=87

Prediction model : Random walk 8GD~ss =CGD 1Ba . 2GD i86/SE,saw Mean Median Greater than count Wilcoxon z-statistic p-value

CGD ss/SE ie85e

0.005 0.004 36

0.012 0.007 51 3 .700.0011

Prediction model: Proportionate investment model WD, as/SEjs6 =CGD, B JSEiegu cGl). .ISE,agu Mean Median Greater than count Wilcoxon z-statistic p-value

CGD. ft6/SE. 86

0.005 0.004 36

0 .011 0 .007 51 3 .700.0011

Panel B

LMT subsample n=27

Prediction model : Random walk 2GD 186 = CGDIaax CGD 1B6/SEie ..a

"cGD~ss/SEie. .~ Mean Median Greater than count

0.002 0.001 20

0 .001 0 .000 7

Prediction model : Proportionate investment model r'GD,ss/SEs6 =CGD,B JSE, aie cGDiu/SEisga Mean Median Greater than count

0 .002 0 .001 20

CGDs6/SEis6 0 .001 0 .000 7

* Significant at 0.05 level for one-tailed test Wilcoxon z-statistic for signed-rank test Greater than counts number of times one value of matched pair is greater than the other value dGD = predicted capital gain differential CGD=capital gain differential : found in the ETR recon . SE = stockholders' equity plus deferred taxes: Compustat (#216+#74)

43

Tar Subsidy Increases in Anticipation of Tax Rate Changes

incentive to increase adjusted tax subsidies in 1986 . In general, the weakest results occur with the proportionate investment model because investment in 1986 is larger than in the base period . Table 4 reports the random walk and proportionate investment model results for the capital gain differential . For both models, the realized capital gain differential for HMT firms is significantly higher than the predicted capital gain differential, which is consistent with H2 . Table 5 reports the result from testing the implicit tax theory model, which by its construction is a joint test of the hypotheses and the implicit tax theory . If the theory is invalid, then a higher than expected pre-tax return could drive the results . Alternatively, higher than expected excluded tax subsidies could drive the results . With this caveat in mind, Panel A of Table 5 reports that 125 of the HMT firms had a higher after-tax return in the test period as opposed to the 91 firms that had the higher return in the base period . This result, which is significant, is consistent with increased tax subsidies in anticipation of rate changes .

Table 5 .

Non-Parametric Results For Implicit Tax Theory Model . HMT sample n = 216 ATROE]ft6 =ATROE aa ,~

Panel A

Mean Median Greater than count Wilcoxon c-statistic

ATROE.,sb

ATROE, x6

0 .133 0 .126 91

0 .150 0 .133 125 2 .65-

p-value

0 .0040 LMT sample n = 94

Panel B

Mean Median Greater than count

ATROE.,,

ATROE.,,

0 .011 0 .004 63

-0 .028 31

-(1 .069

Significant at 0 .05 level for one-tailed test Wilcoxon Z-statistic for signed-rank test Greater than counts number of times one value of matched pair is greater than the other value ATROE = predicted after-tax return on equity ATROE=after-tax return on equity

44

BRADLEY D . CHILDS

CONCLUSION Summary Prior research has examined the shifting of conforming income in response to the ordinary tax rate reductions enacted by TRA 86 . The results of this study indicate that HMT firms shifted tax subsidies, that is to say non-conforming income, into 1986 . While this tax strategy is consistent with other tax strategies that minimize tax outflows, the method of execution and the nature of the cost disincentives are different . The results of this study also indicate that HMT firms shifted the CGD tax subsidy into 1986. Limitations The results of this study should be interpreted with caution for several reasons . First, the LMT firms are contracting instead of expanding . To the extent that the control groups are inadequate, the results of this study are time dependent . Second, tax subsidy behavior is not definitively understood . Thus, some of the prediction models of tax subsidies may be misspecified . However, the reported results are robust to model specification . Third, implicit tax theory describes a relation between investment, economic income and tax subsidies . Accounting proxies for all of these underlying variables are likely to have measurement error . Future Research Directions One of the future research areas would be to investigate tax subsidies not studied in the present study . The research design of this study omitted, for reasons of excluding bias and distortion, several tax subsidies that firms would have had an incentive to shift . For example, a firm on the verge of restructuring could have accelerated its recognition of this event into the fourth quarter of 1986 . Doing so would have entitled the firm to additional tax deductions in a higher tax rate year. Alternatively, a fine that had already recognized restructuring expenses for book purposes could have accelerated economic performance of plant relocations, plant closings and termination payments, which would reverse the previously recorded tax asset. Another area of research lies with mixed income assets . Firms holding assets with potentially mixed income streams (ordinary and capital) had opposing incentives under TRA 86 for the timing of these asset dispositions . Presumably, assets with a higher ratio of capital gain income are more likely to have been realized in 1986 .

Tat Subsidy Increases in Anticipation of Tax Rate Changes

45

ACKNOWLEDGMENTS The author gratefully acknowledges useful comments and suggestions by Jim Angelini, Morris McInnes, two anonymous referees and the Editor, Tom Porcano .

NOTES I . Scholes and Wolfson (1992) define a natural tax clientele as taxpayers more likely than others to own assets with a specific tax status (i .e . AMT firms are more likely to own tax-favored assets). 2 . The incentive is limited because capital gain accruals are not taxed and some assets spin off ordinary and capital gains when sold . 3 . The alternative minimum tax (AMT) literature (Boynton, Dobbins & Plesko, 1992 ; Gramlich, 1991 ; Manzon, 1992 ; Dhaliwal & Wang, 1992) has examined the income shifting behavior of LMT firms, and they report that these firms shifted PTI from 1987, 1988 and 1989 (the period when there was a potential tax on book income) . Some of this shifted income was presumably shifted into 1986 . 4 . Calendar year firms faced a 6% decrease in tax rates, but the fiscal year-end firms experienced a double-digit decrease in tax rates . 5 . Scholes and Wolfson (1992) list the following assumptions for the implicit tax model : no transactions costs, no monitoring costs, no information costs, and buyers and sellers act as if they are unable to influence asset prices . 6 . One method to compute tax subsidies is to subtract current tax expense from the expected tax expense of a given PTI . Another method converts tax credits and rate differentials into NCI equivalents (i .e. tax preferences) . The tax preferences times the tax rate equals tax subsidies . Grossing up the tax subsidies by dividing by (1-tax rate) equals PTTS . For example, $100 of tax-exempt interest is the NCI . In 1986, the tax subsidy (i .e . tax savings) equals $l00*0 .46=$46. The PTTS equals $46/(l -0 .46) = $85 .19 .

7 . Specifically, they find that firms required to switch from cash tax accounting to accrual tax accounting make income-decreasing accruals . When these accruals had not been previously taxed, these accruals (i .e . NCI under the prior tax regime) had been much larger . 8 . See footnote 5 . 9 . Examples of capital assets that are extinguished are bonds that are retired and timber that is processed . 10 . Section 1231 property is subject to Section 1245 and Section 1250 depreciation recaptures. 11 . Tax subsidy data are usually non-normal which has prompted others to use nonparametric tests (see Shevlin & Porter, 1992 and Callihan & White, 1999) . 12 . The creation and utilization of NOL carryforwards is disputed in the literature . Wilkie (1992) considers this NCI item as a distortion, while Shevlin and Porter (1992) regard this NCI item as similar to other NCI items . 13 . Fiscal year firms had blended tax rates for their 1986 tax year .

46

BRADLEY D . CHILDS

14 . An alternative screen for HMT firms would eliminate those firms with any NOL carryforwards . This screen would exclude many HMT firms, because these firms have NOL carryforwards arising from foreign subsidiaries or purchased subsidiaries subject to the separate return limitation rules . 15 . Portfolios are based on size of tax subsidies in the base period . There are 10 portfolios for HMT firms and 5 portfolios for LMT firms . 16 . Both Wilkie (1992) and Callihan and White (1999) have treated stockholders' equity as including the deferred tax liability . This treatment is adopted in this paper . 17 . The LMT sample has 106 firms, but only 94 firms have positive stockholders' equity . The random walk model uses 106 firms, but the proportionate investment model uses only 94 firms because the denominator needs to be positive . 18 . Net capital gain realizations are computed by dividing the tax subsidy by 18% . The negative capital gain differentials reported in the minimum columns of Table 2 refer to capital loss realizations . 19 . The LMT sample contains all firms that were LMT in either the base period of the test period. Including only those firms that were LMT for both periods would decrease the size of this sample . As presently defined, this sample has many fewer firms than the HMT sample .

REFERENCES Boynton, C ., Dobbins, P ., & Plesko, G. (1992) . Earnings management and the corporate alternative minimum tax. Journal of Accounting Research, 30 (Supplement), 131-160 . Callihan, D., & White, R . (1999). An application of the Scholes and Wolfson model to examine the relation between implicit and explicit taxes and firm market structure . The Journal of the American Taxation Association, 21 (Spring), 1-19 . Cloyd, B ., Pratt, J., & Stock, T . (1996) . The use of financial accounting choices to support aggressive tax positions : Public and private firms . Journal of Accounting Research, 34 (Spring). Dhaliwal, D ., & Wang, S . (1992) . The effect of the book income adjustment in the 1986 alternative minimum tax on corporate financial reporting . Journal of Accounting and Economics, 18, 7-26 . Geiger, M . A ., & Hunt, H . G . 111. (1989) . Capital gain taxation : A critical analysis of historical and current issues . Advances in Taxation, 2, 21-39 . Gramlich, J . (1991 .The effect of the alternative minimum tax book income adjustment on accrual decisions. The Journal of the American Taxation Association, 13 (Spring), 36-56. Guenther, D . A . (1994) . Earnings management in response to corporate tax rate changes : Evidence from the 1986 Tax Reform Act. The Accounting Review, 69 (January), 230-243 . Guenther, D . A ., Maydew, E ., & Nutter, S . (1997) . Financial reporting, tax costs, and book-tax conformity . Journal of Accounting and Economics, 23, 225-248 . Johnson, C. (1992) . The undertaxation of holding gains . Tax Notes (May 11), 807-840 . Lopez, T ., Regier, P., & Lee, T. (1998) . Identifying tax-induced earnings management around TPA 86 as a function of prior tax-aggressive behavior . The Journal of the American Taxation Association, 20 (Fall), 37-56 . Manion, G . (1992) . Earnings management of firms subject to the alternative minimum tax . The Journal of the American Taxation Association, 14 (Fall), 88-111 . Maydew, E. (1997) . Tax-induced earnings management by firms with net operating losses . Journal of Accounting Research, 35 (Spring), 83-95 .

Tax Subsidy Increases in Anticipation of Tar Rate Changes

47

Mills, L. (1996) . Corporate tax compliance and financial reporting . National far Journal, 49 (September), 421-435 . Porcano, T. M ., & Shull, D . M . (1997) . A comparative analysis of capital gains-taking . Advances in Taxation, 9, 137-151 . Porcano, T . M . (1997) . An analysis of capital gains tax-induced earnings management . International Advances in Economic Research, 3, 395-408 . Repent. J . (1990) . Long-term capital gains, the long-term investment perspective, and corporate productivity . Tax Notes (Oct. I), 85-102 . Ricketts, R ., & White, C. (1997) . The capital gains tax and stock market returns . The Journal of the American Taxation Association, 19 (Supplement), 51--63 . Scholes, M ., & Wolfson, M . (1992). Taxes and Business Strategy: A Planning Approach . Englewood Cliffs, NJ. Prentice Hall . Scholes, M ., Wilson, P ., & Wolfson, M . (1992) . Financial responses to anticipated reductions in tax rates : The Tax Reform Act of 1986 . Journal of Accounting Research, 30 (Supplement),

161-185 . Shevlin, T ., & Porter, S . (1992). The corporate tax comeback in 1987 : Some further evidence. The Journal of the American Taxation Association, 14 (Spring), 58-79 . Slemrod, J ., & Feldstein, M . (1978) . The lock-in effect of the capital gains tax : Some time-series evidence . Tax Notes (Aug. 7), 134-140 . Wilkie, P. (1988) . Corporate average effective tax rates and inferences about relative tax preferences . The Journal of the American Taxation Association, 10 (Spring), 75-88 . Wilkie, P . (1992) . Empirical evidence of implicit taxes in the corporate sector . The Journal of the American Taxation Association, 14 (Spring), 97-116 .

EQUALITY OR SIMPLICITY: THE INCOME TAXATION OF RETIREMENT PLAN DISTRIBUTIONS Anthony P. Curatola, Janet Trewin and L. Melissa Walters-York

ABSTRACT The income tax rules concerning distributions from qualified retirement plans are considered by most to be a maze of rules defying logic . Some distributions qualify for a variety of special federal income tax treatments while others are taxed as ordinary income . In 1974, Congress proposed to equalize the total tax of taxpayers who receive distributions from retirement plans regardless of whether the distributions are received in a lump sum or as an annuity . The legislation provided a ten-year forward averaging rule for determining the tax to taxpayers receiving a lump sum distribution from a qualified pension plan . This rule was subsequently modified to five-year forward averaging in 1986. Congress recently repealed the five-year averaging rule and enacted a simplified method for determining an annuity's return on investment. The justification for the new legislation was "simplicity" rather than the original purpose - to prevent the bunching of taxable income into a single year as results from a lump sum distribution.

Advances in Taxation, Volume 13, pages 49-68 . Copyright C 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0 .7623-0774-9 49

50

A. P . CURATOLA, J . TREWIN AND L . M . WALTERS-YORK

Although Congress set out to equalize the income tax across the two forms of distributions, after two decades of tinkering, the law penalizes taxpayers who elect to receive the deferred compensation in the form of a lump sum distribution and, thereby, encourages the withdrawal of retirement assets over the life of the retiree all in the name of simplification . However, the result of the changes in the law is neither equality nor simplicity.

INTRODUCTION The income tax rules concerning distributions from qualified retirement plans are considered by most to be a maze of rules defying logic (Groot, 1976 ; Sacher, 1980 ; Crino et al ., 1985 ; Folz, 1986 ; Curatola, 1987) . Some distributions qualify for a variety of special federal income tax treatments under the Internal Revenue Code (IRC) (e .g . IRC Sec. 402(e)), 1 while other distributions are taxed as ordinary income (e .g . IRC Sec . 408(d)) . Distributions from qualified retirement plans (QRPs) that satisfy the conditions prescribed in IRC Sec . 402(e)(4)(A) and that are not explicitly excluded are known as qualifying lump sum distributions (LSD)? An LSD may be entirely cash, entirely stock, or part cash and part stock . LSDs in the form of cash receive long-term capital gain treatment to the extent they are attributable to pre-1974 contribution years . Any amount of a distribution attributable to post-1973 contribution years is taxed as ordinary income . However, the ordinary income portion of an LSD may be taxed under a separate five-year or ten-year forward averaging rule (5YA or TYA, respectively) . If the LSD is capital stock, the amount contributed toward the purchase of the stock by the employer is taxed as ordinary income at the time of distribution, and the net unrealized appreciation is not taxed until the stock is liquidated by the employee. Moreover, the net unrealized appreciation, as well as any appreciation since time of distribution, may be taxed as capital gain in the year of liquidation . The 104th Congress set out to simplify this area with the passage of H .R . 3448, known as the "Small Business Job Protection Act of 1996" (SBJPA '96) . This act, signed by President Clinton on August 20, 1996, eliminates some complexity by repealing the 5YA rule for all taxpayers, leaving the TYA rule for a select few, and providing a simplified method for determining an annuity's return on investment . The justification for repealing the separate tax was to achieve "simplicity" and, through the liberalized rules for a rollover of a distribution to an IRA, resolve the bunching of income problem resulting from an LSD .

Equality or Simplicity : Retirement Plan Distribution

51

Simplicity is certainly an appealing justification for tax legislation . However, simplicity should not supersede the policy reasons, if any, that were the initial motivation and justification for the enactment of tax legislation . This article evaluates the enactment and recent repeal of the special separate tax provisions that were legislated for retirement distributions in light of the original purpose of the provision as set out by Congress in its Committee Reports . If the tax legislation initially had merit, then it is important to determine whether the initial motivation is still valid and, more importantly, whether the elimination of the rules will truly achieve what those rules were anticipated to achieve . Additionally, whether simplicity has been achieved by the enacted legislation is evaluated .

HISTORY OF THE FEDERAL INCOME TAX ON RETIREMENT PLAN DISTRIBUTIONS Pre-Tax Reform Act of 1986 History At first glance, one surmises that Congress arbitrarily provided additional favorable tax treatment for distributions from certain retirement plans through the permitted use of TYA and, more recently, 5YA . A review of the historical development of these rules provides a different perspective . The TYA rule was not arbitrary and, in fact, had and still has theoretical appeal . The Revenue Act of 1942 allowed deferred compensation distributed in a lump sum from a qualified plan to receive long-term capital gain treatment . This special tax treatment was intended as a solution to the so-called bunched income problem of receiving in one taxable year an amount that had accrued over several years (HR Report No . 413) . By 1969, the then Congress believed this solution to have two fatal flaws . First, deferred compensation was taxed more favorably (i .e. as capital gains) than other compensation payments . As a result, a horizontal tax equality problem was perceived to exist . Second, the favorable tax benefits were seen to accrue to taxpayers with adjusted gross income in excess of $50,000 ; therefore, a vertical equity problem appeared to exist (HR Report No . 779) .' To assuage these flaws, section 515 of the Tax Reform Act of 1969 limited capital gain treatment to distributions of pre-1970 contributions from qualified pension plans with post-1969 contributions taxed as ordinary income . In an attempt to handle the income bunching issue of a lump-sum distribution, the act further provided that the ordinary income portion would be averaged using a seven-year forward averaging rule . 4 The details of the seven-year forward averaging rule inclusion and the its associated tax liability resulted in such complexity that Treasury was unable to

52

A . P . CURATOLA, J . TREWIN AND L . M . WALTERS-YORK

finalize regulations . Thus, the 93rd Congress introduced new rules in 1974 for computing tax liability . The new averaging rules were less complex and were designed to yield equality of tax treatment for various plans .' Specifically, Congress sought to equalize taxes (i .e . achieve a form of tax equality) among recipients of deferred compensation from QRPs, regardless of whether the payout form was structured as a lump sum or an annuity . This first of two goals is referred to as the equality policy. According to the House of Representatives report, the new legislation would retain the capital gain tax treatment for pre-1974 contributions and would tax post-1973 contributions as ordinary income (HR Report No .779) . The segregation of the distribution into two time periods was to be accomplished by establishing what percentage of participation time was associated with each period . Therefore, a recipient who participated in a plan from 1966 to 1985 (or 20 years) would have 40% (or eight out of 20 years) of the total distribution classified as pre-1974 contributions . This amount was eligible for capital gain treatment. The remaining 60% of the total distribution would be treated as ordinary income. Congress went further by permitting the ordinary income portion to be taxed separately under the TYA method [Section 2005(a) of Employment Retirement Income Security Act of 1974 (ERISA)] . The justification for this provision was that: (1) most distributees would have little or no other taxable income in the years following their retirement and (2) ten years represented the approximate life expectancy of a person age 65 . Therefore, the ten-year period was to approximate the period over which the income would have been spread had it been received in the form of an annuity and not an LSD (HR Report No . 779) . Thus, a retiree who received an LSD would pay approximately the same tax on the distribution as a retiree who received an annuity .' The new provisions also were made available to self-employed individuals who received distributions from HR 10 plans (also known as Keogh plans) . The main thrust of the new provisions appears to have been an attempt at horizontal tax equality, both within plans and across plans, for distributions to those who are 65 or older . That is, a person would pay approximately the same tax amount if he/she elects to receive an LSD or an annuity distribution . Prior to 1987, the use of TYA was permitted for LSDs regardless of the recipient's age . Furthermore, TYA could be elected repeatedly by an individual who was younger than 59 but only once by an individual who was 59 or older (Sec . 402(e)(4)(B)) . Congress had a dual motive in implementing the one-time election for those 59 or older (HR Report No . 779) . First, the spouse of a decedent could elect TYA for a distribution that was received on behalf of the deceased spouse while

Equality or Simplicity: Retirement Plan Distribution

53

retaining his or her own election, i .e . one election per taxpayer . Second, the retiring spouse could not make multiple distributions and, thereby, further reduce his or her tax . Thus, the separate TYA rules did not provide an additional tax benefit to the taxpayer . On the other hand, as noted by the House Committee, the original purpose for permitting the favorable tax treatment (i .e . capital gains and TYA) to LSDs for individuals under 59 was to mitigate the effect of the progressive tax structure on individuals receiving all of their benefits in a single year as a result of separating from their employer . Prior to 1975, there was no investment vehicle available to receive a distribution of deferred compensation from a QRP and permit the amount to remain deferred until retirement . As a result, the multiple election of TYA for LSDs had merit for younger taxpayers receiving LSDs . Unfortunately, the provision also provided an incentive to take advantage of the tax-favored receipt of the pre-retirement benefits and use the proceeds for non-retirement purposes . The individual retirement account (IRA) was introduced in 1975 with a provision to accept the transfer (direct or rollover) of assets from a QRP . With the introduction of the IRA, a non-retired taxpayer (i .e. under 59) had a place to transfer an LSD from a QRP and continue to defer the income tax on the deferred compensation until the amounts were distributed from the IRA . Hence, there was no longer a need to provide TYA to LSDs received by this group But it is surprising that this liberalized rule for the under 59 recipient did not change until the passage of section 1123 of the Tax Reform Act of 1986 (TRA '86), when the election of TYA was made a one time election and only for an individual who is 5942 or older (Sec . 402(e)(4)(D)) . Recall, the intention of Congress until 1986 was to achieve tax equality . To evaluate the achievement of horizontal tax equality for distributions received by the 65 and older population, two issues need to be examined : (I) Were the assumptions upon which the retirement plan distribution rules were based reasonable AND are these assumptions still reasonable after subsequent tax legislation ; and (2) Does the tax equity sought for qualified plans apply equally to withdrawals from other deferred compensation plans such as IRAs? Reasonableness of Assumptions The two assumptions underlying the TYA rule are the life expectancy of a person 65 years of age and the amount of other taxable income that is available to the retiree . A review of Table I indicates that ten years was a conservative estimate for both men and women who were age 65 even during the periods 1959 to 1961 and 1969 to 1971 . Additionally, Table I reveals that

54


life expectancy for both men and women has increased subsequent to the 1974 legislation enacting TYA . Life expectancy between 1971 and 1992 increased more than 18% for men to 15 .4 years and more than 16% for women to 19 .2 years . These data indicates that the 10-year life expectancy of retirees assumed by TYA is not a reasonable assumption if the goal of Congress is to equalize the tax liability of an LSD and an annuity distribution from a QRP . Although TYA mitigates the tax burden by spreading it over 10 years, it does not necessarily result in lower taxes when compared to the tax liability that results from an annuity over the retiree's expected life . Yet, it is consistent with the second goal of Congress, to encourage a retiree to take his/her deferred compensation over his/her life since it results in a lower tax or, more importantly, more disposable income to the retiree (see Table 2) . The second assumption underlying TYA is difficult to evaluate because specific information for pre-1974 years is not available . Therefore, a comparison of available data from that period is used to assess the reasonableness of assuming little or no taxable income other than retirement plan distributions . The starting point for the comparison is total money income$ that was available in 1971 to the 65 and older population . Table 3 summarizes the percentage distribution of individuals age 65 and older by money income . As shown in Table 3, the median income for all persons 65 and older was $3,071 ($5,358 for married couples and $2,049 for non-married persons) and only 4% of this age group had money income in excess of $15,000 . Since all money does not produce federal income tax liability, these statistics are an overstatement of taxable income and need to be adjusted downward. Approximately 87% of the age 65 and over population received Social Security

Table l .

Selected Life Expectancy of Those Age 65 by Sex and Year . Year

Male

Female

1959-61 1969-71 1979-81 1985 1990 1992 1994 1996

12.9 13 .0 14 .2 14 .5 15 .1 15 .4 15 .5 15 .7

15 .5 16.5 18 .4 18 .5 18 .9 19 .2 19 .0 18.9

Source : Statistical Abstracts of the United States (Annual editions 1995, Table No . 115 ; 1996,

Table No . 118 and 1998, Table No . 129)

55


N V D

.0 a

a b

N 0

0 N T 0 S N

Q

I a

V

N

W 0

V fA

N

0 V h

fA

0C[7a

O

7 U U

C S 0 0 d3

N

Q

N n

Q

o o N

04 a j R V7 aa a k rJ

0 70

F N

a h

0

oo o

F

o 0

o

m

a a E

0 0 m

y

Q

0

A c m

E Q

F

F

0 ? ti M o r

0 0 7 E Q 0 0 E 0

Q

F

0


56

N O 3

mg m -+

n ooo N a mv,y v, v,m~l

a

co N a~no m roar v,NN o y 4

E

a z W 'O 7 \O ~0 V N-

-m-.

"" '"`nv„o S5

r r^

o0

v

b

00

o m o r 'o 'o o N oo 'o v

0 0 zs

P a

M

h

T E

O z

E0

s

ON

a7a7

~.g,a~.rn v

o EO

N N m m ~O U Yf b C I I I I I I I I I I z$g$g$g$$$$$

~


57

benefits in 1971 (Grad, 1977, Table 12) . The median amount of these benefits received was $1,697 for all individuals receiving benefits, $2,341 for married couples receiving benefits, and $1,427 for non-married persons receiving benefits . Assuming the money income and the Social Security benefit distributions for the age 65 and older population equal the median amounts, then a rough estimate for taxable income can be made. That is, taxable income was approximately $1,374 ($3,071-$1,697) for the 65 and older population as a whole, $3,017 ($5,358-$2,341) for married couples and $622 ($2,049-$1,427) for non-married persons . However, the distribution of Social Security benefits is unavailable, so the rough estimate previously made is questionable . Yet, there is another point to be considered . A taxpayer does not incur a federal income tax liability on all taxable income . Congress has legislated exclusions, personal exemptions, and standard deduction provisions for all taxpayers .9 For the 65 and over population, additional provisions such as an additional personal exemption and the elderly income tax credit were available . To simplify the analysis, only the basic minimum provisions are considered (i.e . dividend exclusion, personal and old age exemption) . In 1971, a taxpayer did not incur a federal income tax liability on taxable income below $2,900 ($200 + (4 $675)) in the case of filing a joint married return or $1,450 ($100 + (2 * $675)) in the case of a non-married single return . When these figures are coupled with the Social Security benefits received by the 65 and over population, it can be concluded, as assumed by Congress, that this age group had little or no taxable income .

TAX REFORM ACT OF 1986 The 93rd Congress introduced 5YA for LSDs received by those taxpayers who were age 59'/2 or older . Under the five-year forward averaging (5YA) rules, the applicable tax rates used in the calculation were those applicable for the year of the distribution (IRC Section 402(e)(1), amended by Act Section 1122(a)(2)(A) of Public Law 99-514, 10/22/86) . If a taxpayer was 50 years of age, or older, before January 1, 1986, the taxable amount could be taxed separately under the 5YA or TYA rules (Act Section 1122(h)(3) of Public Law 99-514) . Unlike the 5YA rules, the election of TYA required the use of the 1986 tax rates . Under either method, the tax liability was less than it would have been had the entire distribution been taxed as ordinary income in a single year . Further, the tax liability was still more than the taxpayer would have paid had the amount been taken as an annuity for life or rolled over to an IRA and distributed over the average life expectancy of a taxpayer who was age 594 2 or 70 1/2 .

58


cv E

m N v,

b 7 7 00 ,p h t

m m N N- ~n m- N

3

N .-+ ~{ O O Cl O r

F

v 00 U

J en

N V7

t en

fn en

en 00

en Nin

$g

-0-.--N v vv ~n~n e vV

n ,o

Za,~~nvmenenN-

V7

N ,O

en- en

v ON tie

00

v w

OO

N- m Qv

N

O r

d h h 7 O7 M

en N O r- t

W

a A eo

E0 0

U a O` 0` 0' a 0\ O, o.aa,o,o.a,a,o. 0 S2aa.o,a,ao.o,aa0'ONONao,aaa rno,a,a,a010'Cl ^^N o

o y

0 N


c E0 3

N

N

0 °

r0O V,v nww

59

a

`n

aC

C V

mo U

h G

h

C

0 .C.

aa O o

a a a a aa aa aa aa aa aa aa aaa a±a ap; aN aN rn a a a;aoyaaa -+ N en V ~O r- W a sn sn M Sn Vf Sn SA M Hi Sn Sn Sn 4n I I

I

I I

I I

o 7 C

O C O

60


To establish the reasonableness of the second assumption - little or no other income for this group - at the time of the Tax Reform Act of 1986, an analysis similar to the one previously given is presented for the population 65 and older in 1982 . 70 Table 4 shows that the median total money income for all units aged 65 and older was $8,790, but approximately 29% of the group had income in excess of $15,000 . As was the case for Table 3, these amounts include nontaxable income . A breakdown of money income excluding Social Security benefits is given in Table 5 . The median income in this case was $3,200, and only 16% of the 65 and over population as a whole had income over $15,000 . The federal income tax laws have been modified since 1974 with respect to the calculation of taxable income . For the 1982 taxable year, a married couple could have up to $7,400 ($7,600 including $200 of income in dividends) of taxable income and a single person could have $4,000 ($4,100 including $100 of income in dividends) of taxable income before incurring any federal income tax liability . Considering this information and the information found in Table 5, it seems reasonable to conclude that the little or no other taxable income assumption was met by most taxpayers (half of the married couples and approximately 70% of the non-married persons) . Finally, similar analysis is given for the population age 65 and older in 1994 (i .e . the most recent available data) . Table 6 shows that the median total money income for all units age 65 and older was $15,094 (including nontaxable income) . This means approximately 50% of the group had income in excess of $15,000 including nontaxable income . Table 7 gives a breakdown of money income excluding Social Security benefits . The median income in this case drops to $5,092, and only 26% of the 65 and over population as a whole had income over $15,000 . For the 1994 taxable year, a married couple could have up to $12,750 of adjusted gross income and a single person could have up to $7,200 of adjusted gross income before incurring any federal income tax liability . Considering this information and the information found in Table 7, a case can be made that the little or no other taxable income assumption is reasonable since it would be met by approximately 60% of the married couples and nearly all of the nonmarried persons . In conclusion, the second assumption appears to be violated more as we move from 1974 to the present. That is, the number of taxpayers 65 and older with taxable income has been increasing since 1974 . Yet, the majority of taxpayers appear to satisfy this assumption .


c E

3

61

b b V~ - r- V1 -+ 'b N M P N m N P O g N ,~ M O- N M r O . 0O y b 'O 7 V M O b N rB

-

rn

N of W N- 7 N V O- a o N- N P 01 O O~ M O O - N v~ v~ v~ vt vt v1 O vt v~ e} I- V V N f9

M

v~ b V O M N b M 0 O N 1~ V O N h V M P M N M ~p M f~ n b b 7 7 7 M- ~D [r~ W '~ M O O M

mg

o NNN m Irv eoiMmooMb e'ov, -dodo ooo---ri oinaM^ a_oa;

cc

c N fA

V1 t U n- N 7 1~ N N vl W cc m M N P M M V N O O- N V V7 t h 7 7 M M M m O~ i0 N O

M N

o v C)

P 0 0` O T U 0~ T 0' T 0' 0 P P P P

U P 0, 0, Ot D` U O1 D\ O, 01 0, 01 T Ot pi 0, N - a n' T i O 00 U O-^^= 'n V O4 O N M r W T s9 f9 (9 y3 N3 ap 7

Gy 1

1

1

1

1

1

1

1

1

I

I

I

I

E

c

o z f'^

- .N viv`nen wrn~ _ `~ Yi f9 f9 H3 f5 Yi 44 H Vi 0 fA fA'4

I

v

I

I

I

~n~NO Y3 44 H1

M 44

V a

,C„

62

A . P . CURATOLA, J. TREWIN AND L . M . WALTERS-YORK

N

0 b

c

b H w

E

3

O c O .7

m g ^, O

N N

oco~nNV .-~nav,oo-Oman 01 M ~O 7 y a m N N N N ~O 7 N- m

'o

m at

°° h N

L

A

vi o 0o m r m o, o r v o0 t (V r vi m G ~G h 7 V 7 m m m 0o

b0 ~ ro ~ C d

w

-

'O ci

0 e

aI c

G

0

C)

G

.. ~'

N

^ W 1p eT 7 7 m N N f+ N

O

7 N

7

F a

a U N N U O O G A b

d

M

7

U

o 'O N V 7 r V N O~ o, '0 o, m M M m m N m N N m W W p h N N D•

S V r

O v c

N

c 0 O U

r

H N

N

A ~.y

7,

vl

p Z F

m ~O

h

a O\ T D\ a O` 0\ ~` Ua O~S O~ P T O~ U D` T ^ P ~, 4 N m h V r W a Nf M Vi 4 4y 4e 4 . vi z s vi 44 I I I

c E C) E V n

v 9

m m o r N v, 7 -+ N m' V 7 M M N (J N r ~O 7 m N

' in m-

N A O F o

9p y7 L

n C F

g

O Z

U 7 T O M M 'O I

N

U N ., E

0 S V

Q y r~ fA (A (A fA 4 i9 (9 19 Yi Vi di 4 M to fA

h

u


63

IRAs and SEPs

In 1973, the 93rd Congress acknowledged that not all employees were able to participate in qualified pension plans . However, this Congress felt that employers who did not establish plans provided their employees with a compensating differential, i .e . compensation in addition to that which would otherwise have been paid (HR Report No . 779) . In other words, employees who were not participants in an employer-sponsored qualified plan received additional cash compensation in lieu of plan benefits . Some empirical evidence that higher pension benefits were, at least partially, offset by lower wages is provided by Schiller and Weiss (1980) . The obvious problem is that the employee could not invest the additional compensation in a personal investment fund for retirement without first paying income tax on the earnings and then paying tax on the fund's annual earnings . To address this shortcoming, Congress, in Section 2002 of ERISA, provided a retirement vehicle (i .e . an IRA) for employees not covered by a qualified plan . Employees whose employer did not sponsor a qualified plan now were eligible to make annual deductible contributions to their personalized IRA . As a result of this legislation, all employees have the ability to participate (either through company sponsorship or individual sponsorship) in some form of retirement plan . However, several differences exist between QRPs and IRAs . First, the limitations placed on annual contributions differ dramatically . Second, and relevant to this article, unlike distributions from QRPs, withdrawals from IRAs are not eligible for the TYA or 5YA rules . Rather, the taxable amount withdrawn from an IRA is added to gross income . The justification for this tax treatment, found in the committee report (HR Report No . 779), is as follows : [Slince contributions to the account, etc ., will be made with tax free dollars and income of the account, etc ., will not he taxed as earned, the individual's basis in the account, etc ., is to be zero .

Although the argument for taxing IRA distributions as ordinary income is appealing, it applies equally to distributions from QRPs . Contributions to QRPs are made with tax-free dollars and income of the QRP is not taxed as earned . Thus, the individual's basis in the QRP is zero . There are no real differences between QRPs and IRAs ; hence, the taxation of the distribution from either plan should be the same. Withdrawals from IRAs are at the discretion of the employee ; therefore, retirement benefits can be spread over ten years, and approximately the same tax liability can be obtained under both a qualified plan and an IRA . Indeed, since a 65-year old man (woman) has a life expectancy of approximately 15+

64

A. P. CURATOLA, J . TREWIN AND L . M . WALTERS-YORK

years (19+ years), a logical decision on his (her) part might be to choose a payment schedule greater than ten years . However, a person who elects, for whatever reason, an LSD from his/her IRA is penalized by not being allowed the same tax treatment as for an LSD from a QRP. The logical question to ask is, Why? Congress wanted to encourage individuals to withdraw from an IRA over retirement years and not in a lump sum . This objective was another reason offered by Congress for not allowing the TYA rule for IRA withdrawals (HR Report No. 779) . Of course, this objective is equally applicable to QRP . The net result of this Congressional action was that the TYA provision was made available to some taxpayers to equalize the tax burdens associated with LSDs and annuities, and the ordinary income provision was made available to other taxpayers so that they would make withdrawals over the retirement years and not in a lump sum . Such inconsistent tax treatment across distributions has served only to heighten taxpayer's anxiety about the tax equality and complexity of the tax system. 1997 and Beyond

Most recently, section 1401(a) of SBJPA'96 repealed the 5YA rules for distributions received after 1999 but retained the transition rules pertaining to the use of TYA for those who attained age 50 by January 1, 1986 . The Committee rationalized that the 5YA rules are complex and also increase the likelihood of error even though inexpensive software packages capable of performing the necessary calculations have been introduced in recent years . This 5YA rule calculation is probably one of the simpler ones in the taxation of retirement income . The Committee rationalized that the liberalization of the rollover rules in the Unemployment Compensation Amendment of 1992 fulfilled the original intent of the rules, to equalize the tax liability resulting from the bunching of income from an LSD in a single year . This is true to a degree . The rollover rules have been relaxed somewhat. Pursuant to Regulation Section 1 .401(a)(31)-l (effective as of October 19, 1995), qualified plans are required to provide employees with a direct rollover option. Under this option, an employee may elect to have an eligible rollover distribution paid directly to an IRA (account or annuity), or to another qualified plan . However, qualified plans are not required to accept direct rollovers . Since a qualified plan risks the loss of its qualified status if a non-qualified rollover is accepted as a direct rollover, plans have not quickly adopted rules to accept direct rollovers . Treasury has issued, and will probably continue to issue, regulations to alleviate this problem but for the moment many recipients of LSDs will, more than likely, be limited to an IRA rollover.


65

The result of the rollover election is the need of the taxpayer to distribute the rollover assets by means of the minimum distribution rules found in IRC section 401(a)(9) . The calculation for the minimum distribution amount is simply the account balance divided by the àpplicable life expectancy' (or Applicable Divisor) . Although the account balance is not difficult to determine, the applicable life expectancy (ALE) is another matter . It depends on the designated beneficiary or beneficiaries (i .e . spouse, non-spouse, non-person, or none) on each IRA and the ALE method (i .e. annual recalculation or ALE minus one) elected at the time of the first distribution . These elections have a serious impact on the annually calculated distribution amount for the taxpayer . Moreover, there is no clear cut rule on the optimal elections for a taxpayer. The final decision to elect the "best" ALE method rests on the financial situation, health and age of both the taxpayer and the designated beneficiary . Once the minimum distribution amount is determined, the taxpayer needs to determine the taxable portion of the distribution and whether he or she has established a basis in any IRA . Under prior law, the distribution amount was allocated between the taxable and nontaxable portions by means of the annuity table exclusion method. By electing this method, the expected return is based on a variety of factors (age, contract type, refund feature, sex of annuitant, and pre July 1986/post June 1986 investments) . The method is quite complicated and confusing . Relief from this set of procedures is provided by the simplified method, added by section 1403 of SBJPA'96 . The investment recovery amount (or excluded amount) is calculated as the contract investment divided by the anticipated number of monthly payments, where the anticipated number of monthly payments is based on the person's age bracket, not the life expectancy, at the start of the annuity . The age brackets and corresponding number of payments are given in amended IRC Section 72(d) . Although the ability to rollover deferred compensation from a QRP to an IRA is relatively simple, the distribution calculations are likely to be complex . Thus, even though Congressional intent was to simplify the rules applicable to QRPs, simplicity certainly has not been achieved . From Here to There

In 1974, Congress proposed to equalize the total tax of taxpayers who receive distributions from retirement plans regardless of whether the plan assets are received as a lump sum or as an annuity . Two assumptions were at the heart of the legislation : recipients had a life expectancy of ten years, and they had little or no other taxable income. Given these two assumptions, the legislation

66


provided a tax that was based on a separate TYA rule for taxpayers receiving an LSD from a QRP . An analysis of the TYA assumptions was performed for the period 1971 through 1994. The limited taxable income for the 65 and older population assumption was found to be reasonable . However, the ten-year life expectancy assumption may or may not be seen as reasonable at the time of the enactment of TYA but has clearly become less reasonable as time has passed . Life expectancy exceeded ten years when the TYA rule was enacted and has increased over time . The passage of TRA `86 slashed the marginal income tax rate for individuals from 50% to 28% and eliminated numerous tax rate brackets . This flattening of the rate structure helped to mitigate the tax inequality between the bunching of income problem addressed by the forward averaging rules and the taxation of annuities . However, it did not eliminate the inequality . Since life expectancy has significantly increased beyond ten years, one would have expected Congress to lengthen the averaging method to about 15 years in order to maintain tax equality between LSDs and annuity distributions ." Instead, Congress introduced 5YA for distributions from QRPs and partially repealed TYA . This has served only to further erode tax equality between LSDs and annuities . More recently, Congress, beginning in the year 2000, will have eliminated horizontal tax equality between the two forms of distribution with the repeal of the 5YA rule . The result of this latest legislation is to put distributions from QRPs and IRAs on an equal playing field . That is, LSDs from QRPs and IRAs are equally penalized for the sake of "simplicity" and taxpayers are encouraged to withdraw their deferred compensation over their retirement years . The creation of the TYA, and later the 5YA, rules added a degree of complexity to the calculation of the tax liability attributable to a QRP . Yet, the availability and relatively low cost of tax software today makes the various tax calculations for a distribution a minor inconvenience to many taxpayers . More importantly, the rules may be seen to achieve, in part, some measure of horizontal equality across distributions taken in the form of an LSD or an annuity . Additionally, since the LSD distribution incurred a higher total tax under the TYA and 5YA rules (due to life expectancy), a taxpayer had an incentive to take a distribution in the form of an annuity or elect to rollover the LSD into an IRA and take periodic distributions . Hence, both of the initial objectives were partially achieved by the earlier Congress . The 104th Congress set out to achieve simplicity by repealing 5YA for LSDs . In so doing, it appears to have forgotten that the primary reason for TYA rules was to provide tax equality across distributions (LSDs and annuities) as well as to encourage the distribution of the deferred compensation over a taxpayer's


67

retirement years . What the 104th Congress has achieved by its actions is : (1) to provide a tax penalty to those who elect an LSD from a QRP, and, (2) to increase tax complexity to those who rollover an LSD to an IRA and are forced to deal with the minimum distribution rules for IRAs .

NOTES I . Unless otherwise stated, all statutory provisions refer to the Internal Revenue Code of 1986 as amended . 2 . Unless otherwise stated, references to LSDs are assumed to be qualifying distributions from qualified plans . 3 . In House Report No . 779, the committee stated, "The more significant benefits under this treatment apparently accrued to taxpayers with adjusted gross income in excess of $50,000, particularly in view of the fact that a number of lump sum distributions of over $800,000 have been made" (House Report 93-779, 1974-3 CB 387) . 4 . The seven-year forward rule came out of the Conference Committee . The House and Senatehad both proposed five-year forward averaging rules . However, their proposals differed with respect to the inclusion of other income in the year of distribution and the need to recompute the tax at the end of the five years . The specifics of the different proposals can be found in HR Report No. 413, S Report No . 552 and JC Report No . 782 . 5 . It is interesting to note that during this time of debate, distributions from selfemployed plans were eligible for a five-year forward averaging method . 6 . To fairly compare the total tax payments on these forms of distribution, the aggregatedistribution, without inclusion of any earnings on that amount, is considered .on its reinvestment . 7 . See General Explanation of the Tax Reform Act of 1986, "Blue Book" prepared by the staff of the Joint Committee on Taxation, page 721 . 8 . Grad (1977, 38) defined total money income as the sum of all income received by the aged unit, before deduction for taxes, from the following sources : (1) earnings ; (2) Social Security and Railroad Retirement benefits ; (3) dividends,interest (on savings or bonds), income from estates or trusts, net rental income, or royalties ; (4) public assistance or welfare payments such as old-age assistance, aid to families with dependent children, aid to the blind, and aid to the permanently and totally disabled ; (5) unemployment compensation, government workers pensions, veterans payments, or workers' compensation ; and (6) private pensions, annuities, alimony, regular contributions from persons not living in the household, and other . 9 . For the tax year 1971, a taxpayer could exclude up to $100 ($200, if married filing jointly)of dividends received from a domestic corporation, 13% of adjusted gross income as a standard deduction with a maximum of $1,500, $675 for each personal exemption, and an additional $675 for the taxpayer and/or spouse if 65 or older . 10. Data for 1982 were selected because they were available during the legislative debates for the Tax Reform Act of 1986 . 11 . One could argue that forward averaging was no longer needed as a result of the Tax Reform Act of 1986 (TRA'86) since Congress created a "flat" tax by reducing the number of tax rate brackets to two . However, a few new tax rate brackets have crept back into the income tax law since TRA'86 .

68

A. P. CURATOLA, J . TREWIN AND L . M . WALTERS-YORK

ACKNOWLEDGMENTS This research was funded in part by grants awarded by the International Foundation of Employee Benefit Plans and Drexel University . The authors would also like to thank the anonymous reviewers for their helpful comments .

REFERENCES Crino, M. D ., Curatola, A . P., & Samson, W . D . (1985) . Tax Implications of Lump Sum Distributions . Personnel, 62, 20-22. Curatola, A . P . (1987) . Retirement Plan Distributions and Federal/State Income Tax Equality . Tax Notes, February 23, 817-821 . Employee Retirement Income Security Act of 1974 . (1974) . HR Doc . No. 2, 93rd Cong ., 1st Sess . Folz, D. A. (I986a) . When Rollover IRAs are Best . Trusts and Estates, January, 39-42. Folz, D . A . (1886b) . Estate Tax Strategies for Rollover IRAs . Trusts and Estates, February, 22 28 . Grad, S . (1977) . Income of the Population Aged 60 and Older, 1971 . Staff Paper No . 26 . U.S . Department of Health, Education, and Welfare, Social Security Administration, Office of Research and Statistics. Washington, D .C. : Social Security Administration . Grad, S. (1984). Income of the Population 55 and Older, 1982. U.S . Department of Health, Human Services, Social Security Administration, Office of Retirement and Survivors Insurance and Office of Policy . Washington, D .C . : Social Security Administration . Grad, S . (1996). Income of the Population 55 and Older, 1994 U .S . Department of Health, Human Services, Social Security Administration, Office of Retirement and Survivors Insurance and Office of Policy . Washington, D .C . : Social Security Administration . Groot, G . (1976) . Qualified Plan Distributions : Tax Deferral, ERISA, and the IRA . Fordham Law Review 45, No . 2 (November), 389-407 . HR Rep . No. 413 . (1969) . 91st Cong., 1st Sess . Conference Committee Report . HR Rep . No. 779. (1974) . 93rd Cong ., 2nd Sess . Conference Committee Report . Internal Revenue Code of 1986 as amended . (1996) . Klein, W. A . (1976). Policy Analysis of the Federal Income Tax Text and Readings . Mineola, New York: The Foundation Press . The Revenue Act of 1942. (1942) . HR Doe. No . 7378, 77th Cong ., 2nd Sess . Revenue Act of 1978 . (1978) . HR Doc . No. 13511, 95th Cong ., 1st Sess. S Rep . No . 552 . (1969) . 91st Cong ., 1st Seas . Sucher, C . P . (1980) . How and When a Rollover IRA Should be Used to Receive a Lump-sum Distribution. Taxation for Lawyers, 8(5) (March/April), 288-292 . Schiller, B . R ., & Schiller, R . D . W. (1980) . Pensions and Wages : A Test for Equalizing Differences . The Review of Economics and Statistics, LX II (November) : 529-538. Small Business Job Protection Act of 1996. (1996) . HR Doc . No . 3103, 104th Cong ., 2nd Sess. Tax Reform Act of 1969 . (1969). HR Doc . No . 13270, 92st Cong ., 1st Sess. Tax Reform Act of 1986 . (1986) . HR Doc . No . 3838, 99th Cong ., 2nd Seas . U .S . Department of Commerce, Bureau of Census . (1995) . Statistical Abstract of the United States, 1995 . Washington, D . C . : Bureau of Census .

A SURVEY OF TAX EVASION USING THE RANDOMIZED RESPONSE TECHNIQUE Jodie Houston and Alfred Tran

ABSTRACT We conducted a mail questionnaire survey using both the randomized response (RR) technique and the direct questioning (DQ) technique to directly estimate the prevalence and type of income tax evasion . We also assessed the effectiveness of the RR technique in reducing response and non-response biases and examined the relationship between tax evasion and key demographic variables. Of the respondents completing the RR survey instrument, 5 .5% admitted tax evasion by under-reporting income, and 6 .5% admitted tax evasion by over-claiming deductions. The corresponding proportions obtained from the DQ survey instrument were 1 .7% and 4.2% respectively . The RR technique was ineffective in reducing non-response bias, but the estimated proportions of tax evasion obtained by the RR technique are higher than those obtained by the DQ technique . A relationship was,found between the demographic variables examined and tax evasion . However, interpretation of the results was restricted by the lack of statistical significance of the differences.

Advances in Taxation, Volume 13, pages 69-94. Copyright O 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0 .7623-0774-9 69

70

JODIE HOUSTON AND ALFRED TRAN

INTRODUCTION Despite interest in tax evasion, very little research has been carried out in Australia and not much is known about the extent of the problem . Empirical investigation into tax evasion is needed to understand the extent and the cause of the problem, but current research is hampered by difficulties in obtaining sensitive information about tax evasion from individuals . One of the main limitations facing researchers investigating tax evasion is the inability to directly observe individual evasion behavior . As such, most empirical evidence is based on individuals' self-reports (i .e . surveys) to describe evasion behavior .' Surveys of tax evasion are complicated by the sensitive nature of the topic . In general, tax evasion is perceived to be an illegal and socially undesirable behavior . Individuals are reluctant to admit to having evaded tax . The threat of penalties, prosecution and stigmatization can induce individuals either to lie about their tax evasion behavior (response bias), or to refuse to take part in the study because they wish to avoid answering sensitive questions (non-response bias) .' Response and non-response biases in a survey affect the validity and the generalizability of the results, making reliable estimates of tax evasion difficult to obtain . The problem facing researchers is how to encourage participants to respond, and then to provide a truthful response in surveys . A suggested solution is the Randomized Response (RR) technique first developed by Stanley Warner (1965) . The RR technique was designed to reduce both response bias and nonresponse bias in surveys which ask sensitive questions . It uses probability theory to protect the privacy of an individual's response and has been used successfully in several sensitive research areas, such as abortion, drugs and assault . There have been calls from the tax research community to use the RR technique to investigate tax evasion (see, for example, Elffers et al ., 1988 ; Roth et al ., 1989 ; Harwood et al ., 1993), but a review of literature shows that little has been done . Motivated by the need to gather more reliable and meaningful data on tax evasion and to improve the research methods, we conducted a survey designed to achieve the following objectives : (1) to directly estimate the proportion and type of tax evasion (i .e . underreporting income, over-claiming deductions, and over-claiming rebates and credits) of individuals in Australia ; (2) to assess the effectiveness of the RR technique in reducing response and non-response biases in surveys asking sensitive questions ; and (3) to examine the relationship between tax evasion and key demographic variables .

A Survey of Tax Evasion Using the Randomized Response Technique

71

The rest of this article is organized as follows . The next section briefly reviews the development of the RR technique and its applications in tax research. This is followed by a description of the hypotheses developed and tested in the study . Next, the research design is described, and then the results are reported . The article concludes with a discussion of the limitations of the study, and some improvements in designing surveys using the RR technique are suggested .

LITERATURE REVIEW The Warner Model

The RR technique was proposed originally by Warner (1965) . The innovative approach was designed to protect the privacy of survey respondents when they were asked sensitive questions . In the Warner design, the respondents are given two logically opposite questions and are instructed to answer one or the other depending on the outcome of a randomizing device . For example, suppose the sensitive characteristic is tax evasion . The respondent may be asked to toss a dice, and the outcome determines which question they answer : 1, 2, 3 or 4 : 5 or 6 : Answer :

Question 1 : 1 have evaded tax . Question 2 : I have never evaded tax . True or False

When the respondent answers `true' or `false,' the researcher does not know whether the respondent is answering Question 1 or Question 2 . Thus the privacy of the respondent is protected . The use of probability theory allows the researcher to estimate the proportion of affirmative responses to Question I (rr) and the associated sampling variance using the following equations :

or Thus, and where :

P(True) = P(Question 1) P(TruelQuestion 1) + P(Question 2) P(TrueIQuestion 2)

(1)

X= pm + (I -p)(1-'rr)

(2)

-ir= (X+ p-1) / (2p-1)

(p x 0 .5)

Var(f)=[ar(1-'rr)/n)+[p(1-p)/n(2p-1)2)

(3) (4)

~r = the estimated proportion of `true' responses to Question I ; k = the observed proportion of `true' responses ; p = the probability of answering Question 1 ; and n = the sample size .

72


Warner (1965) claimed that the RR technique had the potential to reduce both response bias and non-response bias resulting from sensitive survey questions . The main drawback was the increase in variance of the estimator due to the introduction of the randomizing procedure into the design . Because of this inflated variance, Warner stressed the importance of using the RR technique only for sensitive issues, so as to offset the increased variance of the estimate with the lower mean square error produced by more truthful reporting . The Unrelated Question Design An important improvement to the Warner model was proposed by Horvitz et al . (1967) who suggested the use of an unrelated question with a non-sensitive characteristic . I For example : Question 1 : Did you cheat on your tax return last year?

Question 2 : Did you watch the 6 :00 pm news yesterday? This unrelated question approach requires two independent samples with different selection probability (p, # p) to estimate two parameters : n x for the sensitive behavior, and err, for the non-sensitive behavior . It has the improvement of reducing the sensitivity of the design, as only one of the questions relates to the sensitive topic. However, the samples also are used to estimate the distribution of the unrelated question (i.e . watching news) which may not be of interest to the researcher . This technique is referred to as the unrelated question design with an unknown distribution . 4 The design was expected to further reduce response bias and improve the efficiency of the estimate . The estimated proportion of affirmative responses to the sensitive question and the associated variance are calculated using the following equations :

where :

'nx=(k,(1-p2)-X2(l-P,)l/(p,-PI)

(5)

Var('fr)=[1/(p,-P2)] [X1(1-X,)(1-p2)2/n, + X2(1-X2)(1 - p,) 2 / n2l

(6)

'r the estimated proportion of `yes' responses to the sensitive question ; X1, X2 = the proportion of `yes' responses for samples 1 and 2 respectively ; PP p2 = the probability of answering the sensitive question for samples 1 and 2 respectively ; and n 1 , n 2 = the size of samples 1 and 2 respectively .


73

The unrelated question design was further improved by Greenberg et al . (1969) . They examined a similar design where the distribution of the non-sensitive question is known in advance . Knowing the distribution of the non-sensitive question offers a substantial improvement in the precision of the estimate of the sensitive characteristic and reduces the number of samples to one, as there is now only one parameter to estimate . The unrelated question design with a known distribution uses a simplified version of the original unrelated question equations to estimate the proportion of the sensitive characteristic and the sampling variance :

where :

[X - (I -p)

'R y l l p

(7)

V ar('R x) = K(l -X) / np2

(8)

'R . =

m = the estimated proportion of `yes' responses to the sensitive question ; Tr y = the known proportion of `yes' responses to the nonsensitive question ; A = the observed proportion of `yes' responses ; p = the probability of answering the sensitive question ; and n = the sample size .

Although the concept of the unrelated question design was introduced by Horvitz et al . (1967), 5 Greenberg et al . (1969) offered a more comprehensive treatment of the refined RR technique and a theoretical proof that it provided a significant improvement on the Warner design by increasing the precision and efficiency of the sensitive estimate, especially when the distribution of the nonsensitive question is known . Further extensions of the RR technique include the use of polychotomous measures and quantitative measures .' Despite the variety of extensions and variations that have emerged since Warner's original design, the unrelated question design developed by Greenberg et al . (1969) has remained one of the most popular RR techniques used by researchers investigating sensitive issues . The RR technique has been used in a few studies in the tax area . They are briefly reviewed below . The Use of RR Technique in Tax Research An early taxpayer compliance study employing the RR technique was carried out by Aitken and Bonneville (1980) 8 who compared the RR technique to a locked-box . The results appeared to be promising : the admission of tax cheating was higher for the RR technique sample than the locked-box sample . These results were later compared to a study done by Yankelovich et al . (1984) who

74


did not use the RR technique . Aitken and Bonneville (1980) found a significantly greater number of respondents admitting to cheating on their taxes than in Yankelovich et al . (1984) . However, little reliance could be placed on the comparison as there were numerous differences between the two studies, such as different surveying techniques, questions and time frames, and a lack of statistical testing . Despite the encouragement offered by Aitken and Bonneville (1980), the randomized response technique did not appear to have been widely used in tax evasion research . In 1993, Harwood et al . (1993) published a research note on the potential benefits of using RR technique in tax compliance research . Their study did not explicitly deal with taxpayer compliance but examined the relationship between the income level of paid tax preparers9 and non-compliance . They briefly reviewed different RR designs and their application in other areas to illustrate the use of the RR technique for sensitive issues . Harwood et al . (1993) offered an adequate description and overview to draw the tax research community's attention to the potential usefulness of the RR technique . The RR technique was used again in the tax area by Larkins et al . (1997), who surveyed tax practitioners . They investigated the ethical issues involved in the tax practice, and more importantly, compared the effectiveness of the RR technique to direct questioning (DQ) . The results indicated that the RR technique did not reduce response bias and non-response bias compared to DQ . Larkins et al . (1997) concluded that the RR technique was unsuitable for surveying tax practitioners about their ethical behavior because the existence of an affinity between CPAs might have resulted in a desire to protect the reputation of the profession . A more likely reason might be that the questions asked, which addressed the compliance of CPAs with the Statements of Responsibilities in Tax Practice, might not have been sufficiently sensitive to warrant the use of the RR technique as the Statement only provided advisory guidelines . Despite the conclusion, Larkins et al . (1997) urged the use of RR techniques in tax research, and emphasized the importance of comparing and validating the RR technique in other tax settings . Overall there has been very little work done using the RR technique in the tax area, with only one study directly addressing the issue of taxpayer compliance (Aitken and Bonneville 1980) . This lack of research reinforces the importance of first establishing the validity of the RR technique in tax research and what improvement, if any, it can add to the traditional DQ technique .

HYPOTHESES DEVELOPMENT We conducted a mail questionnaire survey of Australian individual taxpayers using both RR and DQ techniques to directly estimate the proportion and the


75

types of income tax evasion (under-reporting income, over-claiming deductions, and over-claiming rebates and credits) for the 1997/98 tax year . We designed two survey instruments : one used the RR technique to ask sensitive questions, and the other used the traditional DQ technique . We tested the following two hypotheses to assess whether the RR technique was effective in reducing non-response and response biases in surveys asking sensitive questions : HI : The response rate will be higher for individuals receiving the RR survey instrument than for those receiving the DQ survey instrument . H2 : The proportion of individuals admitting to tax evasion will be higher for those completing the RR survey instrument than for those completing the DQ survey instrument . The first hypothesis tests whether the RR technique reduces non-response bias in the survey . The second hypothesis tests whether response bias is reduced by use of the RR technique and is based on the assumption that a higher proportion of respondents admitting evasion indicates more truthful reporting ." Six more hypotheses also have been developed to investigate the relationship between tax evasion and six taxpayer demographic variables . Table I presents a summary of the empirical results from previous studies which examined the relationship between taxpayer demographic variables and tax evasion . Care should be taken in interpreting the empirical findings summarized in Table I because there are factors which could restrict the comparability of results between studies and contribute to the observed inconsistency, such as those described below . (1) Different research methods have been employed to collect tax evasion data and this may account in part for the observed inconsistency in results between the studies .'' (2) Different target populations have been used ; for example, convenience samples of students, telephone listings, tax agents and other specific subsets of the population . (3) Not all the studies use the same definition of tax evasion, many use the term `non-compliance' which is a broader term, encompassing both intentional and unintentional behavior. (4) The main dependent variable in all these studies is tax evasion (or non-compliance) but only a few have directly estimated tax evasion . Many use hypothetical scenarios or taxpayers' attitudes and opinions, then adopt the assumption that intentional behavior is the same as actual behavior .'-

76


Table 1 .

Association of Tax Evasion and Demographic Variables in Prior Studies .

Author, Date Vogel, 1974** Mason & Calvin, 1978 Mason & Calvin, 1984 Porcano, 1988 Collins, Milliron & Toy, 1992 First model*** Contingency models**** Wahlund, 1992

Income Level

Age

Occupation*

Education

Gender

0

+/-

+ denotes a positive association of the demographic variable with tax evasion . - denotes a negative association of the demographic variable with tax evasion . 0 denotes that the association is indeterminate. * Occupation has been defined in these studies in two ways : (a) self-employed versus employee, and (b) blue-collar versus white-collar . ** The findings of Vogel (1974) were actually based on measures of taxpayer attitudes : taxpayers were asked if they considered themselves the type of person to evade tax . Vogel (1974) is included here as the measures found are close to those from a direct assessment of evasion . *** These are the results from the first model developed by Collins et al . (1992) following traditional guidelines . **** These are the results from the contingency models . +/- indicates that both positive and negative associations were found for different contingencies (i .e. using a tax preparer and the individuals preparing the return themselves.)

(5) There are differences between studies in the number and type of variables that are associated with tax evasion . The absence of relevant variables can produce a confounding effect if they are not properly controlled . Guided by the results of previous studies, we hypothesized the following relationships between the demographic attributes of taxpayers and their propensity to evade tax . H3 : Taxpayers in a lower income bracket will have a higher proportion of

evasion than taxpayers in a higher income bracket . H4 : The proportion of evasion will be higher for younger taxpayers than for older taxpayers . H5 : Taxpayers with a higher level of education will exhibit a higher propor-

tion of evasion than taxpayers with a lower level of education . 116 : Self-employed taxpayers will exhibit a higher proportion of tax evasion than employees .


77

H7: Men will display a higher proportion of evasion than women . The tax preparer (or tax agent in Australian terminology) variable has recently been included in tax evasion studies ." Relevant findings from the tax preparer literature suggest that many tax preparers display aggressive behavior against the tax authorities, aiding taxpayers in avoidance and non-compliance . We therefore hypothesized that : H8 : Taxpayers using a tax agent will display a higher proportion of evasion than taxpayers who do not use a tax agent .

RESEARCH DESIGN We designed two survey instruments : one instrument used the RR technique to ask the sensitive questions about tax evasion (a copy is included as an appendix), while the other used the traditional DQ technique . Otherwise, the two instruments were identical .'^ Both survey instruments offered respondents the protection of anonymity . Respondents of the RR instrument had the added protection of the randomizing procedure . We used the unrelated question design with a known distribution in the RR instrument . We made a few decisions to select the most suitable RR design. Randomizing Procedure The randomizing procedure is vitally important to the success of the RR technique, as it shows respondents that their answers are being protected by probability theory and that the researcher cannot know which question they have answered . The most common randomizing devices used in mail RR surveys are bank notes supplied by the respondent, and random number charts supplied by the researcher . The use of bank notes has been popular in accounting and tax research (see Berry et al ., 1987 ; Harwood et al ., 1993 ; Larkins et al ., 1997) as the serial number on a bank note is considered sufficiently random for the purposes of the RR technique and easily accessible to most individuals . We used the last three digits of the serial number on a bank note of the respondent . The only problem was the potential lack of a bank note,' so respondents were instructed to use the last three digits of their telephone number (which are sufficiently random) if a bank note was not handy . An important parameter to determine is the probability of answering the sensitive question, p, which has an impact on the variance of the estimate . The smaller the level of p (i .e . the fewer respondents who are instructed by the randomizing device to respond to the sensitive question), the greater the protection

78

JODIE HOUSTON AND ALFRED IRAN

offered to the respondents . However, this also means that the sampling variance of the estimator (refer to equation (8)) will increase . Thus, there is a trade-off between respondent jeopardy and estimation efficiency . For example, other things being equal, using a probability of p = 0 .5 will increase the variance by a factor of 4, whereas using a probability of p = 0 .7 will only increase the variance by a factor of 2. This means that a RR design using a probability of 0 .7 would require a sample only about half as large as one using a probability of 0 .5 to achieve the same level of efficiency . Because of this, researchers using the RR technique are encouraged to make p as large as they dare (see Lanke 1975), especially for small samples where the level of sampling error is higher . Soecken and Macready (1982) recommended that p be chosen between 0 .7 and 0 .85 to obtain sufficient efficiency in the design and still protect the privacy of respondents . We struck a balance between respondent jeopardy and estimation efficiency and chose p = 0 .7 in the study . Non-Sensitive Question and Known Distribution In choosing a non-sensitive question and a known distribution, previous studies have used known demographic distributions for certain populations, 76 or have asked respondents if they were born in a certain month," but these measures can be unreliable, with problems of memory recall, respondent knowledge and the validity of the demographic statistics used . Because of these limitations, we followed the practice of several more recent studies (Berry et al ., 1987 ; Larkins et al ., 1997) and used the serial number on a bank note to create a known distribution . The choice of the known distribution will have an impact on the variance of the sensitive estimate . The probability of getting either type of `yes' response is given by : X = p'r x+(1 - p)'rry where :

(9)

7r = the true proportion of respondents with the sensitive behavior ; and ary = the proportion of `yes' response to the non-sensitive question .

The observed proportion of `yes' responses (k) increases as the known distribution (ary) increases . This leads to a larger numerator in equation (8), and results in a higher variance and a less efficient estimate of the prevalence of the sensitive behavior . Setting ar y to zero is in effect direct questioning, as any `yes' response obtained only could refer to the sensitive question . A smaller iry leads to a smaller variance of the sensitive estimate . On the other hand, a


79

larger Tr y provides more protection to respondents as there is a greater likelihood of more respondents answering `yes .' Again, a trade-off exists between respondent protection and estimation efficiency . In the RR survey instrument, we asked the respondents to use a digit in the bank note's serial number as a randomizing device, directing them to answer either the sensitive or non-sensitive question . The non-sensitive question also uses the same digit to create a known distribution of answering `yes,' which has a probability of '/3 or 33 .3% . Ideally, the chosen known distribution for the non-sensitive question should be as close as possible to the sensitive attribute being estimated . With the benefits of hindsight, we found that the known probability for the non-sensitive question we used was too high . The large n y we used has substantially inflated the sampling variance of the estimator for the sensitive attribute . Survey Procedure

A mail questionnaire survey was used because the use of RR technique required larger samples for effective data analysis . We also wanted to survey across all of Australia in order to obtain a representative sample of Australian individuals . The target population for this study was members of the Australian public who derive income and are subject to income tax . Two random samples were drawn using Australia on Disc (May 1999 version) which is a CD-ROM containing all the latest residential telephone directories across Australia and includes a software to draw random samples from the database . The sample size was 500 for the DQ survey instrument, and 1,500 for the RR survey instrument . The larger sample for the RR instrument was meant to compensate for the inflated sampling variances caused by the randomizing procedure . We used a screening question at the beginning of the survey instruments to make sure the respondents were taxpayers who had lodged a tax return for the 1997/98 tax year. An advance letter was mailed prior to the survey instruments to explain the purpose of the study and the procedure, inform respondents of the survey and encourage participation . Ten days after the survey instruments were dispatched, a follow-up letter was posted to thank those who had responded and to remind those who had not to complete and return the questionnaire using the reply paid envelope . Statistical Procedures

Z-tests were used in hypotheses testing . All comparisons involving RR data used the estimated proportion of evasion and the sampling variance based on equations (7) and (8) to calculate the z-score, using the standard formula :

80

JODIE HOUSTON AND ALFRED TRAN z = (m,-*2 ) I [Var(at,) + Var(B2 )j 112

where :

(10)

it=estimated proportion of the respondents admitting tax evasion ; and Var('rz) = variance of the estimated proportion .

To calculate the proportion of respondents admitting to two types of evasion (under-reporting income and over-claiming deductions), the joint distribution of two evasion types was estimated using the following formula :" '

xIx2

= [ X -P,( 1- P2)TTxi'n yz

(1-P,)P2"y,"R .2

-(l-p,)(1-p2)'Ry,'Ry2] / (P,P2)

where :

(11)

°ix,x2=the joint probability of the two sensitive characteristics (evasion types) ; A=the proportion of respondents answering `yes' to both questions ; PP p2 = the probability of answering two sensitive questions 1 and 2 respectively ; ,trx =the estimated proportion of `yes' responses to the sensitive question I or 2 ; and -try = the known proportion of `yes' responses for the nonsensitive question 1 or 2 .

RESULTS Non-Response Bias

The effectiveness of the RR technique in reducing non-response bias was tested in Hl . The results are summarized in Table 2 . There are statistically significant differences (at the 0 .05 level) between the gross response rate of the RR instrument (27 .8%) and that of the DQ instrument (34 .6%), and between the useable response rates of the RR (22 .3%) and DQ (27 .9%) instruments . The null hypothesis that the two instruments had the same response rates is rejected, but the direction of rejection is opposite to the one hypothesized in HI, with the DQ instrument having a significantly higher response rate than the RR instrument . Thus, the RR technique has failed to reduce non-response bias . The following explanations could account for this result : (1) The survey instrument using the RR technique was much more time consuming to complete and involved carrying out relatively complex instructions and procedures to complete it .

A Survey of Tax Evasion Using the Randomized Response Technique Table 2 .

Response Rates and Non-response Bias . Randomized Response Instrument

Number originally sent Number returned undelivered Effective mail-out Responses received* Invalid responses** 1997/98 tax return not yet lodged*** Usable responses*

81

(%)

Direct Questioning Instrument

(%)

1,500 225 1,275

100

500 66 434

100

354 26

27 .8 2 .0

150 9

34 .6 2 .1

44 284

3 .5 22 .3

20 121

4 .6 27 .9

e-score

-2 .61"

--2 .29°

* The response rates and usable response rates were calculated based on the effective mail-out. ** Returned survey instruments were classified as invalid and were not included in the data set if the entire questionnaire was left blank, or the majority of the questionnaire had not been completed, or the respondents clearly did not follow the RR technique properly . *** Survey instruments returned by respondents who had not yet lodged a return for the 1997/98 tax year were also excluded, as the tax evasion questions referred to the 1997/98 tax year . n Statistically significant at the 0 .05 level .

(2) Some individuals receiving the RR instrument did not believe that the survey results could be useful because of the randomizing procedure . This is evidenced by one letter and a few telephone calls in which the recipients expressed this concern to us . (3) A randomizing device was not handy so the recipients did not bother to complete the survey ." (4) There might be a general distrust about the RR technique, as it is a relatively unknown technique and respondents may have felt their privacy was not protected sufficiently . Response Bias

The second hypothesis examined the effectiveness of the RR technique to reduce response bias . The results summarized in Table 3 show that the estimated proportions of admitted evasion in the survey using the RR technique are higher than those in the survey using the DQ technique for the evasion types 'underreporting income' (RR 5 .5% ; DQ 1 .7%) and 'over-claiming deductions' (RR 6 .5% ; DQ 4 .2%) . However, the differences between the two techniques are not statistically significant at the 0 .05 level . The higher estimated prevalence of tax evasion obtained using the RR technique may suggest that the use of the RR

82


Table 3 .

Prevalence of Tax Evasion and Response Bias .

Type of Evasion

Prevalence RR Instrument

Prevalence DQ Instrument

Z-score

5 .5% 6 .5% -1 .6%*

1 .7% 4 .2% 0 .0%

1 .22 0 .66 n/a

5 .0%

n/a

Under-reporting income (Q2) Over-claiming deductions (Q3) Over-claiming tax rebates or tax credits (Q4) Under-reporting income and/or over-claiming deductions

7 .1%**

* A negative estimated proportion is possible using equation (7) when the realized mz is close to zero and the realized Jr, is smaller than its theoretical value (1/3) . ** This proportion is computed using 7r 1 +xr,2-n 1 2 and equation (11) .

technique has reduced response bias . On the other hand, the higher estimated proportions also may be attributed to the randomizing procedure and the responses to the non-sensitive questions . The estimated proportion of taxpayers admitting òver-claiming tax rebates and tax credits' is zero for both RR 20 and DQ instruments . Therefore, H2 is rejected . The effectiveness of the RR technique in reducing response bias cannot be established statistically . Since the proportions of admitted tax evasion in the survey using the DQ technique are very low (e .g . 1 .7%, or 2 out of 121 valid responses, admitted evasion by under-reporting income), testing the association between demographic variables and tax evasion has little meaning . As such, the results of hypotheses testing reported below are related to the RR survey only . Income Level

Consistent with the hypothesized direction in H3, the results in Table 4 indicate that respondents in the lower income group tended to have a higher proportion of tax evasion by under-reporting income and by over-claiming deductions (6 .9% and 9 .3% respectively), compared to respondents in the higher income group (3 .6% and 2 .2% respectively) . However, the differences are not statistically significant and H3 is rejected . Age

H4 examines the relationship between age and tax evasion . The results in Table 5 confirm that the younger age group was less compliant than the older age


Type

of

Income Level and Tax Evasion (RR Instrument) . Level of Taxable Income

Evasion

Under-reporting income

Up

to

$38,001

Overclaiming deductions

Up

to

$38,001

Table 5 .

No . of

Responses

of

of

Evasion


$38,000

175

6 .9

104

3 .6

$38,000

176

9 .3

and above

104

2.2

,z-score

0 .56

1 .18

Age and Tax Evasion (RR Instrument) .

Age Group 18-45

Responses

years

of

Proportion Evasion (%)

111

6 .3

and above

170

5.0

18- 45

years and above

III

10 .2

46

171

4 .1

46

Over-claiming deductions


and above

No . of

Type

83

,z-score

0 .21

0 .97

group, displaying a higher proportion of evasion by under-reporting income (6 .3% vs . 5 .0%) and by over-claiming deductions (10 .2% vs . 4 .1%), but none of the differences proved to be statistically significant . Thus, H4 is rejected . Education Level H5 tests whether taxpayers with a higher level of education exhibit a higher proportion of evasion than taxpayers with a lower level of education . Contrary to the hypothesized direction, a negative relationship between education and tax evasion was found . Taxpayers without tertiary education tended to have higher proportions of tax evasion (6 .4% by under-reporting income and 9 .4% by over-claiming deductions) than taxpayer with tertiary education (4 .8% by under-reporting income and 3 .8% by over-claiming deductions), but none of the differences are statistically significant, so H5 is rejected .


84 Table 6 .

Type of Evasion Under-reporting income


Education and Tax Evasion (RR Instrument) . Education Group*

No. of Responses

Proportion of Evasion (%)

Non-tertiary Tertiary

138 142

6 .4 4.8

0 .27

Non-tertiary Tertiary

139 142

9.4 3 .8

0 .92

z-score

*Non-tertiary education includes taxpayers with no formal education and those that have completed education up to year 12 . Tertiary education includes taxpayers who have a tertiary education other than university and those who have completed university education .

Occupation In H6, we hypothesized that self-employed taxpayers would exhibit a higher proportion of tax evasion than employees . A statistically significant difference was found between the self-employed and employee groups . As shown in Table 7, self-employed respondents admitted a significantly higher proportion of tax evasion by under-reporting income (16 .5%) than respondents who were employees (3 .6%) . H6 is supported in the hypothesized direction . The difference in the proportions of over-claiming deductions also is in the hypothesized direction (9 .2% for self-employed versus 7 .5% for employee) but is not statistically significant.

Table 7 .

Type of Evasion Under-reporting income


Occupation and Tax Evasion (RR Instrument) . Occupation

No. of Responses

Proportion of Evasion (%)

Self-employed Employee

79 144

16.5 3 .6

1 .68*

Self-employed Employee

79 144

9 .2 7 .5

0.23*

* Statistically significant at the 0.05 level (one-tailed test) .

z-score


Gender and Tax Evasion (RR Instrument) . No. of

Type

of

Evasion

Gender


Female


85

Responses

of


z-score

Male

101 180

8 .3 4.0

0 .70

Female Male

101 181

11 .2 3 .9

1 .13

Gender H7 hypothesizes that men would display a higher proportion of evasion than women . The results in Table 8 indicate a higher proportion of evasion occurred among women (8 .3% under-reporting income, and 11 .2% over-claiming deductions) than men (4 .0% under-reporting income, and 3 .9% over-claiming deductions), pointing to a different direction from the hypothesized one . However, the differences are not statistically significant . H7 is rejected . The Influence of Tax Agents The final variable examined was the influence tax agents had on individuals' evasion behavior . The results presented in Table 9 confirm that higher proportions of evasion are present among taxpayers who used the services of a tax agent to prepare their 1997/98 tax returns (6 .2% under-reporting income and 7 .4% over-claiming deductions) compared to those who did not use tax agents (0 .8% under-reporting income and 0 .8% over-claiming deductions) . This is consistent with the hypothesized direction of H8 . However, the differences are not statistically significant, so H8 is rejected.

Table 9 .

Type of

Tax Agent Use and Tax Evasion (RR Instrument) .

Evasion

Under-reporting income Over-claiming deductions

Return Prepared by Tax Agent

Responses

Yes No

223 57

6 .2 0.8

0.82

Yes

224 57

7 .4 0 .8

0 .99

No

No . of of


c-score

86


SUMMARY AND CONCLUSION Of the respondents completing the RR survey instrument, 5 .5% admitted tax evasion by under-reporting income and 6 .5% admitted evasion by over-claiming deductions . The corresponding proportions obtained from the DQ survey instrument were 1 .7% and 4 .2%, respectively . The RR technique was ineffective in reducing non-response bias, and its effectiveness in reducing response bias could not be established statistically . Some relationships were found between the demographic variables examined and tax evasion . The association of the demographic variables and tax evasion confirmed some new trends found in tax evasion research, such as women evading more than men and the relevance of the tax preparer variable . However, interpretation of the survey results was restricted by the lack of statistical significance of the differences in hypotheses testing ." The only statistically significant result was for the occupation variable, where a significantly higher proportion of tax evasion by under-reporting income was found among respondents who were self-employed compared to respondents who were employees . There are several ways to improve the efficiency of the design and hence the significance of the results . The first consideration is the type of RR technique chosen, as some RR techniques are more efficient than others . The RR technique used in this study (unrelated question with a known distribution) is one of the most efficient RR designs, as it uses a known distribution for the non-sensitive question, leaving only the sensitive attribute to be estimated ." The second consideration is the choice of parameters used in operationalizing the RR technique since this can affect the efficiency of the estimates, particularly the choice of p (the probability of answering the sensitive question) and rri, (the known distribution for the non-sensitive question) . The choice of parameters and their relationship to the sampling variance of the estimator are discussed in the research design section . The general rule is that the closer p is to 1 and ary is to zero, the greater the efficiency of the design ." However, the jeopardy to the respondents also must be taken into consideration . The jeopardy level of the RR design cannot be ignored for the sake of efficiency, as this would be a refutation of whole reasoning behind using a RR technique . The tradeoff between efficiency and jeopardy is the dilemma of using the RR technique and presents no easy solution . Researchers must either deal with the increased inefficiency of the estimates affecting the significance of the results or, if choosing an overly efficient RR design, run the risk of respondents refusing to participate due to high levels of respondent jeopardy . As noted earlier, with the benefits of hindsight, we found that the


87

known probability for the non-sensitive question we used was too high . The large ar y we used substantially inflated the sampling variances of the sensitive estimators . Another consideration is the sample size and the response rate . Sample size is constrained by the availability of resources . Higher response rates produce smaller variances of the estimates, and this will increase the efficiency . Low response rates are a concern for most surveys addressing sensitive issues . We expected to increase response rate by using the RR technique . Unfortunately, we found that ordinary people receiving a RR technique survey instrument might find the instructions difficult to comprehend, and even when the instructions were comprehensible, they still might have difficulties in appreciating the usefulness of the survey results . Furthermore, when the RR technique is used, the respondents inevitably have to spend more time to read and follow the instructions . All these factors contribute to the low response rate, which is contrary to expectations . Thus, it is a challenge to researchers to write clear RR instructions which are easy to comprehend and follow, and are able to convince the respondents that the procedure will protect their privacy, yet also provide useful data for the researchers . In this study, the data analysis was restricted to univariate analysis . The extension of the RR technique to multivariate analysis, such as multiple regression and analysis of variance, has been discussed in the RR technique literature but none of the suggested theoretical models''- 4 for adjusting the RR data for multivariate analysis have been attempted by tax researchers . In a recent study, van der Heijden et al . (1998) successfully used the RR technique and regression analysis to investigate social security fraud . Future tax research using the RR technique may consider using multivariate analyses to test more sophisticated theories of tax evasion .

NOTES I . Tax audits carried out by the Australian Taxation Office focus more on measuring non-compliance, and can only infer evasion behavior, rather than directly measure it . Experimental designs cannot directly measure tax evasion either, as many rely on the assumption that intended behavior, and behavior occurring under laboratory conditions, is equivalent to actual behavior . 2 . Response bias refers to any systematic distortion of the respondents' true answer and this endangers the validity of the survey measurements . Non-response bias arises when the respondents answering the survey are not representative of the sample, and there are systematic differences between the respondents who do and do not respond to the survey questions . 3 . This suggestion is attributed to Simmons and the unrelated question design is often referred to as Simmon's model .

88


4 . The unknown distribution refers to the distribution of answers to the unrelated question which is estimated as part of the design . 5 . The unrelated question design with a known distribution was mentioned only briefly in the Horvitz et al . (1967) article . 6 . See Fox and Tracy (1986) for a review . 7 . Although the unrelated question was proposed by Simmons and first published by Horvitz et al . (1967) it is referred to throughout this paper as being developed by Greenberg et al . (1969) as this was the more popular and detailed paper outlining the technique, and has been used as the basis for most studies . 8 . This article was unavailable . The reported results were taken from Roth et al . (1989 : 343) . 9 . In Australia tax preparers are referred to as tax agents . 10 . This is a common assumption made in comparison studies involving the RR technique (for example, see Larkins et al ., 1997) . 11 . Elffers et al . (1992) compared three different measures of tax evasion . No correlation was found to exist between self-reports and the experiment data, and the authors suggested that the lack of correlation might have been caused by the different methodologies measuring different aspects of tax evasion . 12 . There is some doubt as to the validity of this assumption (Hite, 1988) . 13 . For example, see Collins et al . (1992) who examine tax evasion using a contingency approach, with one of the contingencies being whether a taxpayer uses the services of a tax preparer to prepare their return . 14. The only other difference between the two surveys is the omission in the DQ instrument of two questions in section III that evaluate the RR technique used. 15 . Lack of a randomizing device can lower the response rate and reduce the effectiveness of the RR technique (Larkins et al ., 1997) . 16 . For example, targeting respondents with a prison record would allow researchers to use reliable demographic data as the details may be available from their prison record . 17, See Bradburn and Sudman (1979) who used birth months ; though this study was flawed through over use of birth months creating respondent suspicion . 18. See Fox and Tracy (1986 : 52) . 19. Respondents were given the option of using the last three digits of their telephone number, but this may have been forgotten as it was only mentioned once in the instructions, and the questions all refer to the serial number on a bank note . 20 . A negative estimated proportion is possible using equation (7) when lr x was close to zero and the realized w was smaller than its theoretical value ('/ 3) . 21 . Other studies in the accounting and tax areas using the RR technique also encountered similar problem of obtaining results in hypotheses testing which are not statistically significant (Berry et al ., 1987 ; Larkins et al ., 1997) . 22 . The relative efficiency of the unrelated question approach with a known distribution compared to Warner's model was illustrated in Section II . 23 . Recall that setting p equal to 1 results in direct questioning, as all the respondents are directed to answer the sensitive question, and setting 'r r equal to zero means that all `yes' responses can only refer to the sensitive question . 24 . See Rosenberg (1979) and Eriksson (1976) .


89

ACKNOWLEDGMENTS The authors appreciate the helpful comments of the participants at the 2000 Tax Research Network Conference in Birmingham, the United Kingdom . The financial support of the Australian Research Council's Small ARC Grants is gratefully acknowledged .

REFERENCES Aitken, S . S ., & Bonneville, L . (1980) . A general taxpayer opinion survey . Prepared for Office of Planning and Research, Internal Revenue Service. Washington : CSR, Inc . Berry, L. E., Harwood, G . B ., & Katz, J. L . (1987) . Performance of auditing procedures by governmental auditors : some preliminary evidence . The Accounting Review, 62, 14-28 . Bradbum. N ., Sudman, S ., & Associates (1979) . Improving interview method and questionnaire design. San Francisco : Jossey-Bass Publishers . Collins, J ., Milliron, V ., & Toy, D. (1992) . Determinants of tax compliance : a contingency approach . The Journal of the American Taxation Association, 14, 1-29 . Elffers, H ., Weigel, R . H ., & Hessing, D. J . (1987) . The consequences of different strategies for measuring tax evasion behavior . Journal of Economic Psychology, 8, 311-337 . Elffers, H ., Robben, H. S . J ., & Hessing, D . J . (1992) . On measuring tax evasion . Journal of Economic Psychology, 13, 545-567 . Eriksson, S . (1976) . Regressions analysis of data from randomized interviews. Confidentiality in surveys . Report No . 17 . Department of Statistics, University of Stockholm . Fox, J . A ., & Tracy, P . E. (1986) . Randomized response : a method for sensitive surveys . Beverly Hills, CA : Sage Publications Inc . Greenberg, B . G ., Abul-Ela, A . A ., Simmons, W .R., & Horvitz, D .G . (1969) . The unrelated question randomized response model : theoretical framework . Journal of the American Statistical Association, (June), 520 .539 . Harwood, G . B ., Larkins, E . R ., & Martinez-Vazquez, J . (1993) . Using a randomized response methodology to collect data for tax compliance research . The Journal of the American Taxation Association, 15(3), 79-92 . Hite, P.A. (1988). An examination of the impact of subject selection on hypothetical and selfreported taxpayer noncompliance . Journal of Economic Psychology, 9, 445-466. Horvitz, D . G ., Shah, B . U ., & Simmons, W . R . (1967) . The unrelated question randomized response model. Proceedings of the Social Statistics Section . American Statistical Association . Lanke, J . (1975) . On the choice of the unrelated question in Simmons version of randomized response. Journal of the American Statistical Association, 70, 80-83 . Larkins, E . R., Hume, E . C ., & Garcha, B . S . (1997). The validity of the randomized response method in tax ethics research . Journal of Applied Business Research, 13(3), 25-32 . Mason, R ., & Calvin, L . (1978) . A study of admitted income tax evasion . Law and Society Review, (Fall), 73-89 . Mason, R ., & Calvin, L . (1984) . Public confidence and admitted tax evasion. National Tax Journal, 37(4), 489-496. Porcano, T . M . (1988) . Correlates of tax evasion . Journal of Economic Psychology, 9, 47-67 . Rosenberg, M. J . (1979) . Multivariable analysis by a randomized response technique for disclosure control . Unpublished doctoral dissertation, University of Michigan .

90


Roth, J . A., Scholz, J . T., & Witte, A . D . (Eds) (1989) . Taxpayer compliance - Volume 1 : An agenda for research . Philadelphia: University of Pennsylvania Press. Soeken, K. L ., & Macready, G . B . (1982) . Respondents' perceived protection when using randomized response . Psychological Bulletin, 92 (September), 487-489 . Van der Heijden, P . G . M ., van Gils, G ., Bouts, J ., & Box, J. (1998) . A comparison of randomised response, CASAQ, and direct questioning ; eliciting sensitive information in the context of social security fraud. Kwantitative Methoden, 19, 15-34 . Vogel, J . (1974) . Taxation and public opinion in Sweden : an interpretation of recent survey data . National Tax Journal, (December), 499-513. Wahlund, R . (1992) . Tax changes and economic behavior : the case of tax evasion . Journal of Economic Psychology, 13, 657-677 . Warner, S . (1965) . Randomized response : a survey technique for eliminating evasive answer bias. Journal of the American Statistical Association, 60, 63-69. Yankelovich, Skelly & White, Inc . (1984) . Taxpayers Attitudes Study : Final Report. Prepared for the Internal Revenue Service . Washington : Government Printing Office .

APPENDIX Randomized Response Survey Instrument THE AUSTRALIAN NATIONAL UNIVERSITY Survey of Taxpayers All responses are anonymous and will be kept strictly confidential . If you have not lodged an income tax return for the 1997/98 tax year (the year ended 30 June 1998), please pass on this questionnaire to someone else in your household who did lodge a 1997/98 tax return for completion . If no one in your household has lodged a tax return for 1997/98, please tick `No' in question 1, then answer questions 5 to 14 . 1 . Did you lodge a tax return for the 1997/98 tax year? (Please tick one box .) Yes

Please answer all the questions in Sections 1, 11 and III .

No

Please answer all the questions in Sections II and III only .

SECTION I : In this section we would like to ask you some potentially sensitive questions about your 1997/98 tax return, but we don't want to put you on the spot, so we are using a procedure that makes it safe for you to respond

A Survey

of Tax

Evasion Using the Randomized Response Technique

91

truthfully to each question without anyone ever knowing which question you actually answered . First, take a bank note from your wallet or purse and look at the LAST three digits of the serial number on the bank note . (If you don't have a bank note handy, please use the LAST three digits of your telephone number.) Do not make any note of these numbers on the questionnaire . We will ask three pairs of questions on the next page . You answer only one question in each pair, depending on the serial number on the bank note, which only you know . We will not know which question in the pair you answered ; we will only be able to statistically draw some conclusions about all the respondents as a group.

2 . If the LAST digit of your bank note's serial number is I, 2, or 3, answer question 2A . Otherwise, answer question 2B . 2A . Is the LAST digit of the serial number an even number? 2B . In filing your 1997/98 tax return, did you intentionally omit some of your income that was subject to tax? Your answer to 2A or 2B is : (Please tick one box.)

Yes

No

3 . If the SECOND LAST digit of your bank note's serial number is I, 2, or 3, answer question 3A . Otherwise answer question 3B . 3A . Is the SECOND LAST digit of the serial number an even number? 3B . In filing your 1997/98 tax return did you intentionally claim deductions (expenses and losses) that you were not entitled to? Your answer to 3A or 3B is : (Please tick one box.)

92


4 . If the THIRD LAST digit of your bank note's serial number is 1, 2, or 3, answer question 4A . Otherwise answer question 4B . 4A .

Is the THIRD LAST digit of the serial number an even number?

4B . In filing your 1997/98 tax return did you intentionally claim tax rebates or tax credits that you were not entitled to? Your answer to 4A or 4B is : (Please tick one box .)

Yes

No

SECTION II : Please complete the following background information which will help us prepare a profile of the respondents .

5 . How old are you? (Please tick one box.) Under 18 years of age 18-30 31-45 46-60 Over 60 years of age 6. Are you male or female? (Please tick one box.) Male Female 7 . Which of the following best describes your occupation during the 1997/98 tax year? (Please tick one box .) Self-employed


L ]

93

Employee Unemployed Retired

L

8 . Was your 1997/98 tax return prepared by a tax agent? (Please tick one box.)

L-1

Yes

F-I 9 . Which is the highest level of education that you have achieved? (Please tick one box.)

I

I

No formal education Completed primary school

L I

Completed high school (year 10)

L L

Completed college (year 12)

I

Completed tertiary education other than university (eg TAFE) Completed university

10 . What was your taxable income (net of deductions) for the 1997/98 tax year? (Please tick one box.)

J

Below $5,400 $5,401-$20,700 $20,701-$38,000

U J

$38,001-$50,000 Over $50,000

94


SECTION III : Please indicate the extent of your agreement or disagreement with the following statements by CIRCLING a number to help us evaluate the questionnaire we use.

Strongly agree

Agree

Neither agree nor strongly disagree Disagree disagree

11 . All of the questions and instructions were clear in their meaning .

2

3

12 . I felt that my privacy was protected by the anonymity of my response .

2

3

4

5

5

13 . I understood the method of selecting which question in a pair to answer using the serial number on a bank note .

I

2

3

4

5

14 . 1 felt that my privacy was further protected by the procedure used in questions 2 to 4.

1

2

3

4

5

Thank you very much for your time . Please return the questionnaire in the enclosed reply paid envelope.

AN EXAMINATION OF THE "REALISTIC POSSIBILITY" STANDARD'S INFLUENCE ON TAX PRACTITIONER AGGRESSIVENESS Charles F . Kelliher, Dale Bandy and Andrew J . Judd

ABSTRACT: Professional standards are used as a means of regulating professional behavior . This relationship raises the question of how standards can be effectively employed . This study considers whether tax preparer aggressiveness is influenced by the preparer's familiarity with the relevant professional standard. The current "realistic possibility" standard replaced the "reasonable basis" standard as the previous standard was considered too low a threshold. Its application did not afford an effective form of ethical guidance controlling tax practitioner aggressiveness (Bandy et al ., 1993 ; Graetz, 1987) . The results of this study indicate that practitioner aggressiveness is inversely related to familiarity with the standard . This finding in turn suggests that a program of familiarization may be needed to achieve compliance with professional ethical standards .

Advances in Taxation, Volume 13, pages 95-121 . Copyright © 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0-7623-0774 .9 95

96

C . F. KELLIHER, D . BANDY AND A . J . JUDD

INTRODUCTION The economic theory of regulation suggests that the accounting profession acts in a way that supports regulation by the government (Ayers et al ., 1989) . Regulation theory, based on the work of political scientists Bentley (1908) and Truman (1951), proposes that regulation exists, in certain instances, to benefit the regulated parties . Regulated parties comply with regulations in order to protect the favored standing they enjoy . Regulations limit the access of potential competitors to the market creating protected rights for regulated parties . Further, the public relies on the regulation process and accepts regulated status as an indication of professional qualification . For CPAs, the benefits of regulation stem from the provisions of Circular 230, which allow only CPAs, lawyers, and enrolled agents the right to practice before the IRS . This right allows fees to be earned through representing a client's interest in obtaining rulings, negotiating settlements, and resolving other matters with the IRS . Failure to maintain the practice standards specified in Circular 230 may result in fines, censure, damaged reputation, litigation, revocation of the right to practice before the IRS, and the loss of related revenues . Theoretical research in economics and finance suggests that neither increases in the precision nor the level of professional standards will mitigate the extent to which practitioners make overly aggressive reporting decisions (Kane, 1977, 1979, 1981 ; Finnerty, 1988 ; Cuccia et al ., 1995 ; Hackenbrack & Nelson, 1996) . These studies suggest such revisions to regulations merely encourage innovations designed to achieve, through other means, the outcomes consistent with decision-makers' incentives . Cuccia et al . (1995) found that CPAs act in a manner consistent with this theory in case studies relating to tax practice . Their findings are discussed in greater detail in the next section . The question addressed by the current study is : "How does familiarity with professional standards affect tax advice given by CPAs?" . In two case studies, we test the relationship between familiarity with the realistic possibility standard and tax compliance . Through experience, practitioners have the opportunity to internalize professional standards . Many experiences affect the professional environment in which CPAs practice . Some of these are : (1) personal experiences with clients, the IRS, and fellow practitioners ; (2) increased understanding through repeated application of standards ; (3) firm policies ; and (4) AICPA guidelines . We test whether greater familiarity with a professional standard contributes to compliance with that standard . The remainder of this article is organized as follows . The next section presents the background and development of hypotheses relevant to the study . This section is followed by a discussion of the research method and a presentation

Realistic Possibility Standard's Influence on Tax Practitioner Aggressiveness

97

of the results . Finally, the results are summarized and opportunities for future research are discussed .

BACKGROUND AND HYPOTHESIS DEVELOPMENT In 1976, Congress enacted the first statutory provision attempting to regulate tax practitioner aggressiveness . Internal Revenue Code (IRC) Section (Sec .) 6694 provided for the assessment of a $100 (now $250) penalty against a tax return preparer if the understatement of the taxpayer's liability was, in any part, due to a position for which there was not a "reasonable basis" for being sustained on its merits . Both the accounting and the legal professions found the "reasonable basis" standard to be such a low threshold that its application did not afford an effective form of ethical guidance controlling tax practitioner aggressiveness (Bandy et al ., 1993 ; Graetz, 1987) . In 1988, the American Institute of Certified Public Accountants (AICPA) adopted a higher ethical standard . This standard applies to practitioners who recommend tax return positions or sign tax returns . The AICPA Tax Division's Statements on Responsibilities in Tax Practice (STRP) (1988 Revision) No . I, "Tax Return Positions," states that CPAs must have "a good faith belief that the [tax return] position has a realistic possibility of being sustained administratively or judicially on its merits if challenged ." In 1989, Congress modified IRC Sec . 6694 by replacing the "reasonable basis" standard with the "realistic possibility" standard . The House Report specifically stated, "the committee has adopted this new standard because it generally reflects the professional conduct standards applicable to lawyers and certified public accountants ." The regulations issued for IRC Sec . 6694 adopted the American Bar Association's one-in-three standard as the threshold for determining whether the realistic possibility standard is met . Though the AICPA adopted a non-quantified standard, if CPAs act in a manner consistent with regulation theory, their tax advice regarding issues that lack clear-cut answers should be within the one-in-three range expressed in the Treasury Regulations for the Internal Revenue Code and in Circular 230 . While the AICPA's "good faith belief' position (or non-quantifiable standard) is not as strict as the Treasury standard, it is likely that CPAs would adhere to the more stringent standard to preserve the benefits of their regulated status . The obvious question from a regulation perspective is : "What impact do these guidelines have on professional practice?" . As noted, theoretical research in economics and finance suggests that neither increases in the precision nor in the level of standards will mitigate the extent to which practitioners make overly aggressive reporting decisions (Kane, 1977, 1979, 1981 ; Finnerty, 1988 ;

98


Cuccia et al ., 1995 ; Hackenbrack & Nelson, 1996) . Each of these studies suggests practitioners find ways to circumvent regulations . In fact, Finnerty (1988) suggests that a significant portion of financial innovation is attributable to the avoidance of tax, accounting, and other regulatory requirements . Studying 103 recent financial innovations, Finnerty concludes that tax avoidance is a primary motivate for 27 innovations (e.g . master limited partnerships), financial reporting advantage is a primary motive for 7 innovations (e .g . in-substance defeasance), and other regulatory avoidance is a primary motive for 20 innovations (e .g . interest rate swaps) . That is, much of what takes place in the financial arena is regulation circumvention . Kane (1979, 1981) proposes that loophole mining and fabrication have become a main business of regulatees . Kane (1981 : 359) adds, "When the costs of avoiding traditional forms of regulations are low, regulatory intentions are easily frustrated ." He further states (1981 : 358), "regulatory intentions are easily frustrated [because] any new regulation kicks off an immediate search for feasible avoidance or circumvention activities Cuccia et al . (1995) studied whether replacing a standard that employs a vague, verbal disclosure threshold with a standard that employs a more stringent numerical threshold mitigates the aggressiveness of reporting decisions . They mailed an experimental instrument to tax managers of Big Six firms . The participants were asked to decide whether the proceeds obtained from the settlement of a defamation of character lawsuit should be included in the client's taxable income. One experimental case tested the impact of a vague, verbal standard on the aggressiveness of reporting decisions while the other experimental case examined a more stringent, numerical threshold . In addition, both cases manipulated the client's stated risk preference . In one version, the client was described as conservative, wishing to avoid risk (include the proceeds in taxable income), while the other version contained an aggressive client who was willing to take risk (exclude proceeds from taxable income) . Both cases also tested the relationship between client's desires and subject's interpretation of the hypothetical standard to justify aggressive positions . According to Cuccia et al ., when a verbal standard is in place, tax practitioners use the latitude inherent in the verbal standard to support aggressive reporting . When a numerical standard is in place, tax practitioners may use instead the latitude available in assessing evidential support to justify an aggressive reporting decision . When actual professional standards are replaced with fictitious standards in an experimental setting, as was the case with Cuccia et al ., it is unclear whether practitioner behavior is restrained by concerns suggested by regulation theory (e .g . fear of loss of license to practice or damage to professional reputation) .

Realistic Possibility Standard's Influence on Tax PractitionerAggressiveness

99

Cuccia et al . rationale is based on a tenant of economic theory that suggests individual actions are motivated by financial self-interest . The economic theory of regulation suggests that regulated professionals act in ways that support government regulation (Ayers et al ., 1989) . The regulated parties benefit from complying with regulations because compliance assures continued protection . Regulation theory adds to the concept of self-interest the idea that decisions are not only influenced by factors such as the immediate potential loss of a client, but also by the desire to preserve the long-term benefits of regulation . Regulation theory and self-interest should be viewed as complimentary explanations of behavior . The basic premise of regulation theory is that individuals conform to regulations in order to receive the "economic rent" that regulation provides . Accordingly, regulation theory broadens the concept of self -interest to recognize that decisions may be influenced by the long-run benefits of regulation. In the short-run practitioners may face the loss of a client, but in the long-run they preserve "economic rent" . In order for practitioners to behave in a manner consistent with regulation theory, it is necessary that they have a minimum familiarity with the standard . When they are unfamiliar with the standard they lack the understanding necessary to fulfill expectations . When they are unaware of requirements associated with professional standards, they remain aware of the potential immediate loss of revenue associated with the loss of a client. Without knowledge of the professional standard they cannot properly consider the long-run implications of aggressiveness . Hence, familiarity with the professional standard is essential to compliance because familiarity with the standard enables the practitioner to understand the long-run implications of the immediate decision . Further, other behavioral theories suggest that factors other than financial selfinterest may affect individual actions . The term fiscal psychology was adopted by Schmolders (1959) to refer to an area of public finance research concerned with the psychological foundations of financial activity . Schmolders (1970) considered tax compliance to be a behavioral matter affected by fiscal psychology . This theory is based on the premise that governmental authorities can function effectively only when the public approves of the policies enough to willingly comply (Engstrom & Giles, 1972 ; Easton, 1975 ; Sarat, 1977 ; Bardach & Kagan, 1982 ; Tyler, 1986 ; Milliron & Toy, 1988) . Milliron et al . (1989) observe that financial self-interest appears as an important, but not overriding criteria . They conclude that taxpayers also judge policy alternatives on the basis of fairness, simplicity, and economic growth . Collins et al . (1990) studied the impact of penalties on tax practice . Their study concluded that practitioners are driven by concern for professional reputation, and that firms move to minimize the risk of penalty by changing office practices when penalty provisions were changed.

100

C . F . KELLIHER, D. BANDY AND A . J . JUDD

The study by Cuccia et al . (1995) used fictitious standards created for purposes of their experimental manipulations . This design did not address the impact of changing professional expectations or the possible risks of damaged professional reputations . It is unclear whether the responses of these of subjects were influenced by their lack of familiarity with the fictitious standard, and it also is unclear how practitioners compensate for the lack of familiarity gained from experience with an actual practice standard . As practitioners gain more experience with a professional standard, they have the opportunity to internalize that higher standard . Additionally, sustained exposure to a standard may integrate that standard into a practitioner's value system . Cuccia et al . acknowledge these possible influences in their study when explaining their decision to use a fictitious standard in their experiment . They state, "The `realistic possibility' standard in effect currently for tax preparers was not used because we were concerned that subjects' responses would be influenced by guidelines provided by their firms, the American Bar Association, and the Internal Revenue Service" (p . 234, footnote omitted) . Such influences, however, contribute to the professional environment in which practitioners provide advice . It is these very influences that may result in conformity with professional guidelines regardless of their form . Thus, the question addressed in this study is : "How does familiarity with the realistic possibility standard affect tax advice given by CPAs?" . The answer to this question could indicate whether professional standards are an effective means of regulating tax practitioner behavior . The reason for the change to the realistic possibility standard was a concern that the prior reasonable basis threshold was too low and thus allowed too much latitude in recommending aggressive tax return positions . If tax preparers who are familiar with the newer standard are less aggressive than those who are less familiar, the newer standard has the potential for achieving the desired behavioral change . Such a finding would provide support for the premise that professional standards can be used to effectively regulate professional behavior . Although there may be a basic behavioral tendency of practitioners to circumvent regulation, it is unclear whether this tendency may dominate the desire to preserve the benefits of regulation ; especially as familiarity with a regulatory standard grows . Over time as professionals become more familiar with a new or revised standard, they are able to better understand its meaning and implications for continued professional certification and professional reputation, and this familiarity may moderate their behavior . Changes in behavior may not be based solely on being aware of a new or revised professional standard . Over time, professionals also have the opportunity to internalize professional expectations associated with the changed standard .


101

In our study, the impact of familiarity with the realistic possibility standard on practitioner aggressiveness is tested in two cases . In one case, subjects are given latitude to assess evidentiary matters, and in a second case this latitude is limited . Individuals more familiar with a professional standard should behave more consistently with the provisions of that standard than those individuals who are less familiar with it. The reason given for the enactment of the realistic possibility standard was concern that the prior "reasonable basis" standard was too low a threshold to prevent practitioners from recommending overly aggressive tax return positions . An obvious question is whether this standard raised the threshold of support tax practitioners need before recommending a favorable tax return position . Despite a client's preferences, practitioners who are familiar with the professional standard know that there is a minimum level of support that must be present before recommending a tax return position to a client . This leads to the following hypothesis : H : Tax preparers who are more familiar with the realistic possibility standard will provide advice that is more consistent with the Treasury's one-in-three standard.

RESEARCH METHOD The research instrument was mailed to 1,000 randomly selected members of the AICPA's Tax Division . Responses were received from 238 individuals, representing a response rate of nearly 24% .' Subjects were presented two cases in which they were asked whether a taxpayer was entitled to a tax deduction . One case dealt with a worthless security deduction and the other with an education expense deduction . Subjects also were asked to indicate their familiarity with each of the two tax issues and with the realistic possibility standard . This design resulted in one dependent variable (the recommendation to deduct or not deduct) for each of the cases and two independent variables - client aggressiveness and familiarity with the realistic possibility standard . Familiarity with each of the two tax issues (measured on a seven-point scale), years of tax experience, and gender were added as control variables . Independent Variables Familiarity with the Realistic Possibility Standard Subjects were asked to indicate on a seven-point scale their familiarity with the realistic possibility standard . The end points of the scale were labeled "none" and "very familiar ." Because of low cell counts in the extreme responses, cells I and 2 (6 and 7) were pooled into common categories .

102


Client Aggressiveness The client aggressiveness factor was manipulated at two levels . The aggressive client had "a strong desire to claim a deduction ." Alternatively, the risk averse client wanted "to claim a deduction as long as he can avoid a challenge from the IRS ." Subjects received only one version of the instrument in a betweensubjects design so that the experimental manipulation would not be obvious . Control Variables While the primary variables of interest in this study are familiarity with the realistic possibility standard and client aggressiveness, there are many other subject attributes that may add insight into tax practitioners' decision-making behavior (Roberts, 1998) . The model-building strategy used in this study was to include only those factors that had some relationship with the dependent variable . The objective was to reduce the number of variables included in the model . As more variables are added, the model becomes more dependent on the observed data (and becomes more sample specific) due to the increase in the estimated standard errors . Based on the univariate tests, the following control variables were included in the model . Familiarity with the Tax Issue Subjects indicated on a seven-point scale their familiarity with each tax issue - education deduction and worthless stock deduction . Once again, the cells with low cell counts (responses 2 and 3) were collapsed into a common category . There were no "1" responses to either tax issue . Tax Experience and Gender Tax experience and gender often are cited as additional factors that may affect the risk preference of tax practitioners' judgments . Since there was some relationship between the dependent variable and experience and gender, these variables were added to the model as control variables? Unlike some prior work that suggests a difference in the risk-taking behavior of men and women, there was only a small gender difference in this study . The rather weak relationship between gender and the dependent variables may be attributable, in part, to the small number of women respondents (n = 29) . Dependent Variable Recommendation In both sections (worthless stock and education cases), subjects were asked whether it was appropriate to claim the deduction . The instrument forced the


103

subject into a dichotomous - yes or no - choice. Additionally, in the education deduction case, responses were analyzed to determine the "threshold" at which the subject would move from not recommending the deduction to recommending the deduction . That is, subjects could chose to recommend the deduction at the 10, 20, 30, 40, or 50% threshold or they may have not recommended the deduction in any case . This decision strategy or "threshold" variable was measured on a six-point scale, with higher values indicating a higher threshold . Research Instrument

The research instrument contained three sections . The first two sections presented the two tax issues requiring the subjects to decide whether the client should claim the deduction in question . The final section collected demographic data to describe the sample and to later classify the subjects into various categories . The first case in this study asks subjects to recommend whether to claim a worthless stock deduction in a temporally linked series of scenarios describing the experiences of a shareholder in a corporation that experiences progressively worsening financial conditions (see Appendix A for a complete copy of the research instrument) . Subjects had to interpret the facts and apply provisions from an unclear law . This manipulation allowed consideration of the realistic possibility standard in a case involving an area of tax law where the appropriate treatment of the item is unclear because available authority does not contain specific criteria that establish a "bright line" test . No odds were included to serve as an anchor suggesting whether the tax preparer was in the "ball park" and to help decide what other factors might be considered . The introductory paragraphs describe briefly the worthless stock deduction and provide information regarding a hypothetical corporation along with facts relating to the performance of the corporation during 1991. The subjects then were asked to decide the appropriateness of claiming the loss deduction in 1991 . If they advised the client to claim the deduction in 1991, they were instructed to go to the second section of the questionnaire . If subjects decided that claiming the deduction was inappropriate, they received additional facts that represented the corporation's performance for the subsequent year . This process was repeated for up to five years . The second case asked subjects to decide the appropriateness of claiming a deduction for education expenses incurred by a client while attending graduate school . Five case scenarios, presented in random order across subjects, contained different combinations of trial-level court cases that favored/opposed the client's

104

C . F . KELLIHER, D . BANDY AND A . J. JUDD

position .' The probabilities of success manipulated throughout the five scenarios were 10, 20, 30, 40, and 50% (see appendix A) . Although CPAs do not ordinarily encounter quantifiable odds when giving advice, research identifying some cases both allowing and denying the deduction comes close to a situation where CPAs might employ odds when applying the realistic possibility standard . For each scenario, subjects were asked whether it was appropriate to claim the deduction . By omitting other factors such as low confidence in the evidentiary matter or ambiguous tax provisions, the principal uncertainty faced by subjects was the weighing of legal authority . Thus, the practitioner subjects were required to provide advice in a situation where there is little latitude to interpret evidentiary matters or exploit ambiguity in the tax law .

RESULTS Subjects Table 1 shows descriptive statistics and frequency distributions for the sample . The 238 respondents had an average of 17 years tax experience, ranging from 2 to 45 years . Only eight individuals had fewer than 5 years experience and over 70% of the sample had more than 10 years of tax experience . While the sample is highly skewed toward experienced practitioners, it is judged to be representative of CPAs who regularly give tax advice to individual taxpayers .'

Table 1 . Descriptive Statistics : Sample . Number of responses'

Variable

Category

Gender

Male Female

208 29

Experience

2-10 years 11-20 years Greater than 20 years

65 115 53

Current employment

Big Six CPA firm Regional CPA firm Local CPA firm Self-employed Law firm Other

15 II 118 61 2 30

'Note : Some totals add up to less than 238 because of missing data .


105

Subjects were randomly distributed across the United States and most were employed in public accounting (60%) or were self-employed (26%) . Of those in public accounting, roughly one-half were employed by local firms . Education Deduction

A logistic regression model was used to analyze and isolate the effects of the independent variables and test the hypothesis . The response variable was the subject's decision "threshold" measured on a six-point scale while the independent variables were familiarity with the realistic possibility standard and client aggressiveness . Years of tax experience, familiarity with the education deduction, and gender were the control variables . Table 2 shows the results of the logistic regression model . Familiarity with the realistic possibility standard (p = 0.0830), client aggressiveness (p = 0 .0005) and tax experience (p = 0 .0340) all were related to the subject's recommendation "threshold" . Table 3 reports the number (and percent) of CPAs in each cell who advised the client to claim the education deduction . It appears that some CPAs have a threshold lower than the one-in-three regulatory test . Twenty-two percent of the CPAs advised the client to claim the deduction in the 10% probability case, while nearly thirty-percent of the subjects recommended the deduction in the 20% probability case . Still, the large increase in the percentage of CPAs recommending the deduction in the 30% probability case suggests that most CPAs comply with the one-in-three threshold contained in the Treasury's realistic possibility standard . The percentage of subjects recommending the deduction nearly doubled from 30% (in the 20% probability case) to 56% (in

Table 2 .

Logistic Regression Model : Education Deduction .

Independent Variables

p-value

Client Aggressiveness Familiarity with the realistic possibility standard Familiarity with the education deduction Tax experience Gender

0 .0005 0 .0830 0 .4410 0 .0340 0-2079

Dependent Variable - Decision strategy or "threshold", measured on a six-point scale, with higher

values indicating a higher threshold . It measured when the subject would move from not recommending the deduction to recommending the deduction . That is, subjects could chose to recommend the deduction at the 10, 20, 30, 40, or 50% threshold or they may have not recommended the deduction in any case.

106

C . F . KELLIHER, D . BANDY AND A . J . JUDD

the 30% probability case) . Of course, tax advice is normally more complicated and subtle than simply counting how may court cases favor a particular position, and other factors can influence professional judgment regarding expectation of success . Client Aggressiveness In all five cases, CPAs provided more aggressive advice to clients having a "strong desire" to claim the deduction (Table 3) . It appears that these CPAs

Table 3 .

Number (and %) of Subjects Recommending Education Deduction and Recommendation Threshold .

Never

Threshold at which subject moved from not deduct to deduct 10% prob. 20% prob. 30% prob . 40% prob. 50% prob .

27 (12 .4%)

49 (22.6%)

24 (11 .1%)

57 (26 .3%)

38 (17 .5%)

Number recommending education deduction in : 10% prob. 20% prob . 30% prob . 40% prob . Totals (n=238)

22 (10.1%)

50% prob .

52 (21 .8%)2

74 (29.8%)

132 (55.5%)

174 (73 .1%)

189 (79.4%)

21 (17 .8%) 31 (25 .8%)

25 (21 .2%) 49(40 .8%)

57 (48.3%) 75 (62.5%)

79(66.9%) 95 (79.2%)

86 (729%) 103 (85 .8%)

22 (52.4%) 15 (38 .5%) 11 (17 .7%) 17 (30.9%) 9 (23 .1%)

25 (59 .5%) 25 (64 .1%) 29(46 .8%) 29 (52 .7%) 24(61 .5%)

32 (76 .2%) 31 (79 .5%) 40(64 .5%) 40(72 .7%) 31 (79 .5%)

37 32 47 44 29

6(30.0%)

12(60 .0%) 28 (58 .3%) 38 (58 .5%) 34(50 .1%) 20(54 .1%)

17 35 53 44 25

16(80.0%) 39 (81 .2%) 54(83.1%) 52 (77.6%) 28 (75.7%)

Client aggressiveness Avoid challenge (n= I18) Strong desire (n = 120)

Familiarity with the realistic possibility standard" 1,2 (n = 42) 3 (n=39) 4 (n = 62)

17 (40.5%)^ 11 (28 .2%) 11 (17 .7%)

5 (n=55) 6,7 (n = 39)

10(18 .2%) 3 (7 .7%)

(88.1%) (82.1%) (75.9%) (80.0%) (74.3%)

Familiarity with the education deductiona 2,3 (n=20) 4 (n=48)

6 (30.0%0 11 (22 .9%)

5 (n = 65) 6 (n = 67) 7 (n = 37)

20(,30.8%) 6(8 .9%) 9(24.3%)

14(29.2%) 25 (38.5%) 17(25 .4%) 12(32.4%)

(85 .0%) (72 .9%) (81 .5%) (65 .7%) (67 .6%)

" Where : I = None and 7 = Very familiar . " Where percentage is computed as follows : cell count/row total (52/238 = 21 .8% or 17/42 = 40 .5%)


107

were influenced by client preferences once some minimum level of support was attained . However, this result does not imply unethical behavior, as it is appropriate to allow more aggressive clients to take more aggressive positions once a minimum confidence threshold is met. Familiarity With the Realistic Possibility Standard Familiarity with this professional standard was related to practitioner advice in scenarios where there was little evidence in support of the client's position (i .e. the 10% and 20% probability scenarios) . In these two instances, CPAs who were more familiar with the realistic possibility standard were much less likely to recommend the deduction . Specifically, 7 .7% of CPAs who were more familiar with the standard (responses 6 and 7) recommended the deduction in the 10% probability scenario compared to 40 .5% of those who were less familiar (responses I and 2) . In the 20% probability scenario, 23 .1% of those who were more familiar recommended the deduction compared to 52 .4% of those who were less familiar . That is, their recommendations were more consistent with the Treasury Regulation's one-in-three standard . This suggests that the professional standard is achieving its intended effect on those practitioners who are more familiar with its provisions . That is, tax preparers who are familiar with this professional standard are less likely to take aggressive tax return positions when the facts do not support the tax return position . Not surprisingly, familiarity with the standard did not have much affect on the recommendation in the last three scenarios (i .e . the 30% through 50% probability cases) . One would expect the judgments of both groups to converge once the client's position has exceeded the one-in-three threshold . Thus, whether one is familiar with the standard or not, increasing the quantity of evidentiary support above the realistic possibility threshold does not significantly affect CPAs' advice . This evidence provides support for the hypothesis that when there is little support for the client's position, CPAs who are more familiar with the realistic possibility standard provide advice that was more consistent with the realistic possibility standard . Familiarity With the Education Deduction There were no significant relationships between practitioners' advice and their specific knowledge of the education deduction . This was not surprising since these cases provided few facts permitting preparers to utilize personal knowledge of the education deduction . They had little opportunity to rely on their knowledge of the provision . The fact that there was no significant difference suggests that greater familiarity with this provision does not automatically incline preparers toward aggressive positions .

108


Table 4 .

Number (and %) of Subjects Recommending Worthless Stock Deduction. 1991

Yes No

1 (0 .4%) 237 (99 .6%)

Number recommending deduction in : 1992 1993 1994 7 (2.9%) 230 (97 .1%)

46 (20 .0%) 184 (80.0%)

88 (47 .8%) 96 (52 .2%)

1995 89 (92 .7%) 7 (7.3%)

Note : The sample size decreases each year because the subjects were asked to indicate the earliest

year that they would recommend claiming the deduction . Once they recommended the deduction they were instructed not to consider any subsequent years .

Worthless Stock Deduction Table 4 reports frequency data for each year to show when practitioners recommended claiming the worthless stock deduction . Since few subjects recommended the deduction in either 1991 or 1992 (eight combined for the two years), only the data from 1993-1995 were included in the subsequent logistic regression analysis .' While most of the CPAs recommended that the worthless stock deduction should be claimed in either of the last two years (1994 or 1995), there was no clear "best" answer . The results of the separate univariate logistic regression model run on each tax year are presented in Table 5 .1 Also included in Table 5 is the number (and %) of subjects in each cell recommending the worthless stock deduction . Familiarity With the Realistic Possibility Standard In 1993, tax preparers who were more familiar with the realistic possibility standard were less likely to recommend that clients claim the worthless deduction early . The difference was significant at the endpoints of the scale (p = 0 .051 on a univariate test of the extreme responses) . Only 10 .8% of the subjects who were most familiar with the realistic possibility standard (responses 6 and 7) recommended the deduction in 1993 . This compares to 20% in the intermediate categories (responses 3, 4, and 5) and 29 .3% of the subjects who were less familiar with the standard (responses 1 and 2) . That is, subjects who were less familiar with the standard were almost three times as likely to recommend the deduction early (in 1993) compared to subjects who were more familiar with the standard . It appears that practitioners who were more familiar with the realistic possibility standard behaved in a manner more consistent with the standard . This result suggests that more familiar practitioners may have better appreciated the importance of compliance with professional standards . Beyond 1993 the proportion in each group who recommended the deduction was

109

Realistic Possibility Standard's Influence on Tax Practitioner Aggressiveness Table 5 .

Number (and %) of Subjects Recommending Worthless Stock Deduction and Univariate Logistic Regression Model . 1993

Number recommending deduction in : 1994

1995

88 (37 .0%)

89 (37 .4%)

43 (36 .4%) 45 (37 .5%) 0 .8425

46 (38 .9%) 43 (35 .8%) 0 .2564

10 (24 .4%) 15 (39 .5%) 24 (41 .4%) 24 (44 .4%) 15 (40 .5%) 0 .4358

17 13 23 19 16

6 (30.0%) 11 (26 .2%) 21 (35 .6%) 36 (47 .4%) 14 (35 .0%) 0 .8108

7 (35 .0%) 13 (31 .0%) 23 (39 .0%) 29 (38 .2%) 16 (40 .0%) 0 .6835

Totals (n=238) 46 (193%)b Client aggressiveness Avoid challenge (n = 118) 24 (203%) Strong desire (n = 120) 22 (18 .3%) p-value 0 .5035 Familiarity with the realistic possibility standard' 1,2 (n=41) 3 (n=38) 4 (n=58) 5 (n=54) 6,7 (n=37) p-value

12 (29 .2%)° 9 (23 .7%) 11 (19 .0%%) 10 (18 .5%) 4(10.8%) 0 .3464

(41 .5%%) (34 .2%) (39 .7%) (35 .2%) (43 .2%) 0 .7853

Familiarity with the worthless stock deduction , 2,3 (n=20) 4 (n=42) 5 (n =59) 6 (n =76) 7 (n=40) p-value

'

7 (35 .0%) 12 (28 .6%) 12 (20 .3%) 9 (11 .8%) 6 (15 .0%) 0 .0523

Where: I = None and 7=Very familiar . Where percentage is computed as follows : cell count/row total (46/238=19 .3% or 12/41 =29 .2%)

approximately equal . Although the statistical support for the hypothesis was not strong, there is some indication that the realistic possibility standard is having the desired effect. Client Aggressiveness

In all three years, the number of tax preparers recommending the worthless stock deduction to conservative and aggressive clients was nearly identical . Subjects did not respond to the client's aggressive desire where the evidentiary matter did not support the aggressive position . Familiarity With the Worthless Stock Deduction

CPAs who were more familiar with the worthless stock deduction provision were less likely to recommend claiming the deduction early . In 1993, the result

110

C. F . KELLIHER, D . BANDY AND A . J . JUDD

was statistically significant (p = 0 .0523) . For example, in 1993 only 13% of the subjects who were more familiar with the deduction (responses 6 and 7) recommended the deduction versus 35% of the subjects who were unfamiliar with the deduction (responses 2 and 3) . Perhaps these subjects were better able to understand the complexity of the worthless stock deduction, the importance of claiming the deduction in the proper period, and the type of evidence necessary to support the tax return position . This result contradicts the finding in Duncan et al . (1989) who found that knowledge of the provision contributed to greater aggressiveness . The Duncan et al . result may be explained by the fact that in their study preparer knowledge of the relevant provisions was measured using the percentage of the subjects' clients who invest in tax shelters - the larger the percentage of clients who were invested in tax shelters the more knowledgeable the preparer . It could well be that the index is, in reality, one of acceptance or even tolerance of tax shelters rather than familiarity with tax-sheltered investments . Our results suggest that preparers who are more familiar with the tax issue are better able to distinguish between overly aggressive and permissibly aggressive positions . That is, they are better qualified to know when the facts support the desired tax return position and when they do not support such a position . Tax preparers who are unfamiliar with a tax issue may take overly aggressive tax positions simply because they fail to recognize when they are taking a position lacking the minimum level of support required by the tax law . Alternatively, these individuals may forgo favorable positions simply because they do not recognize that the minimum level of support required by law is present, and this leads to undue influence by the client's preferences . Further, individuals who are unfamiliar with the tax provision may lack the common benchmark required for uniform behavior .

CONTRIBUTIONS AND FUTURE RESEARCH The findings of this study extend the body of literature regarding tax practitioners' compliance with professional standards . The results of the study indicate that practitioners who are more familiar with the realistic possibility standard provide less aggressive advice when there is little support for the client's position . In the education deduction case, practitioners were given quantifiable odds and little additional information in order to create a situation where they could not rely on evidentiary detail to support aggressive positions . Practitioners who were more familiar with the standard were less aggressive . In the worthless stock case, practitioners were provided with evidentiary details relating to a possible deduction . Again, practitioners who were more familiar with the


III

standard were less aggressive in the critical threshold year . One explanation is that the more familiar practitioners were with the standard the better they appreciate the importance of compliance with the standard . The results of the two cases together suggest that the change in professional standards has produced the desired behavior of inhibiting preparer aggressiveness . A reduction in overly aggressive positions is consistent with regulation theory and implies more consistency in practice . The dramatic growth in corporate tax shelters and the growth in tax consulting compared to tax compliance represent the current business context in which the regulation of tax professionals exists . An obvious question is whether aggressive behavior crosses the line that the realistic possibility standard establishes . Clearly some tax planning schemes are abusive while others are not . In the past, most of tax practice revenue was generated from compliance related activities, and as a result, the realistic possibility standard was more often associated with compliance decisions . The expansion of tax services results in a growing number of planning decisions where professionals must determine whether a particular course of action falls within the bounds established by the realistic possibility standard . This growth in consulting practice means that the realistic possibility standard is even more important . Because familiarity with standards apparently can influence aggressive behavior, the profession should consider efforts to increase practitioner familiarity with professional standards . Continuing education programs, articles in professional journals, coverage in college curricula, and professional examinations each can contribute to greater familiarity with the standard and more consistent behavior by practitioners . Future studies may consider which strategies are the most effective means for increasing familiarity with the standard . Future studies also may consider how CPAs who are most familiar with the standard deal with their clients . That is, do CPAs who are most familiar with the standard state their concerns more adamantly when dealing with aggressive clients'?

NOTES I . To test for possible non-response bias t-tests compared the mean responses of the first third of the sample with the responses of the really late responses . The really late group included the last thirty-five subjects (or approximately 15% of the sample) . There were no significant differences between the two groups based on their familiarity with the realistic possibility standard (p = 0 .340) ; familiarity with worthless stock deductions (p=0 .232) ; or familiarity with education deductions (p = 0 .240) . Also, there were no significant differences between the demographic variables gender (p=0 .356) or tax experience (p=0 .176) . There was no difference between the treatments and response

112


date (p = 0.384) . No significant differences were found between the two groups for either dependent variable . 2. The univariate tests between the subjects' recommendations and years of tax experience and gender found some relationship in several of the cases/scenarios . Years of tax experience is moderately correlated with familiarity with the realistic possibility standard (r=0 .174 ; p=0 .008) . Despite this correlation the model appeared stable . The estimated coefficients and standard errors were nearly identical in both the full model that included the control variable and the reduced model that included only the two treatment variables . The estimated coefficients and standard errors in the full model were comparable to the results from the univariate model fit for each independent variable . Overall, the multivariate model including the control variable did not produce large uninterpretable estimated coefficients and the estimated standard errors were reasonable . 3 . A repeated measures logistic regression model with version of the questionnaire as the independent variable and education deduction as the dependent variable revealed no significant order effect (p=0 .2754) . 4 . The current employment characteristics of the sample were similar to the membership breakdown of the AICPA . Thirty percent of the subjects were self-employed (AICPA - 23%) ; sixty-three percent indicated regional and local firms (the AICPA's firms with 2-9 members and more than 10 members - 63%) ; and 7% were employed by Big-6 firms (the AICPA's largest 25 firms - 20%) . The percentages do not include two categories : "law firms" (2) and "other" (30) . 5 . An additional logistic regression model was run after combining tax practitioners that recommended the worthless stock deduction in 1991, 1992 or 1993 . This was done to avoid losing information on the most aggressive tax practitioners . This model reported similar results . 6 . While a repeated measures model was run (with missing data), only the results of the univariate model are reported in the paper . Because of the declining sample size, it is not a true repeated measures design, however, the results were consistent .

REFERENCES American Institute of Certified Public Accountants, Federal Taxation Executive Committee . (1991) . Tax Return Positions . Statement on Responsibilities in Tax Practice (1991 Revision) . Ayers, F ., Jackson, B ., & Hite, P . (1989) . The economic benefits of regulation : Evidence from professional tax preparers . The Accounting Review, 300-312 . Bandy, D ., Judd, A . J., & Kelliher, C. (1993) . Dealing with shades of gray : The realistic possibility standard . Journal of Accountancy, (December), 51-58 . Bardach, E., & Kagan, R. (1992) . Going by the Book: The Problem of Regulatory Unreasonableness . Temple University Press . Bentley, A. (1908) . The Process of Government . University of Chicago Press. Collins, K., O'Neil, C ., & Cathey, J. (1990). Tax practitioners' reactions to return preparer penalties . Advances in Taxation, 15-43 . Cuccia, A. D ., Hackenbrack, K ., & Nelson, M. W. (1995) . The ability of professional standards to mitigate aggressive reporting . The Accounting Review, 227-248 . Duncan, W., LaRue, D., & Reckers, P . (1989) . An empirical examination of the influence of selected economic and noneconomic variables in decision making by tax professionals . Advances in Taxation, 91-106 .


113

Easton, D . (1975) . A re-assessment of the concept of political support . British Journal of Political Science, 435-457 . Engstrom, R., & Giles, M . (1972) . Expectations and images : A note on diffuse system support for legal authorities . Law and Society Review, 631-636. Finnerty, J . D . (1988) . Financial engineering in corporate finance : An overview . Financial Management, (Winter), 14-33 . Graetz, M . (1987) . Too little, too late . The Tax Times, (February, 17) . Hackenbrack, K ., & Nelson, M . W . (1996) . Auditors' incentives and their application of financial accounting standards . The Accounting Review, 43-59 . Kane, E . J . (1977) . Good intentions and unintended evil : The case against selective credit allocation . Journal of Money, Credit, and Banking, (February), 55-69 . Kane, E. J . (1979) . The three faces of commercial-bank liability management . In : R . Lombra & H . Kaufman (Eds), The Political Economy ofPolicymaking . Beverly Hills : Sage Publications . Kane, E . J . (1981) . Impact of regulation on economic behavior: Accelerating inflation, technological innovation, and the decreasing effectiveness of banking regulation . Journal of Finance, 355-367 . Milliron, V . C ., & Toy, D . R . (1988). Tax compliance: an investigation of key features . The Journal of the American Taxation Association, 84-104 . Milliron, P ., Watkins, R ., & Karlinsky, S . S . (1989). Policy judgments of taxpayers : An analysis of criteria employed. Advances in Taxation, 201-221 . Roberts, M . L. (1998) . Tax accountants' judgment/decision-making research : A review and synthesis . The Journal of the American Taxation Association, 78-121 . Sarat, A . (1977) . Studying American legal culture : An assessment of survey evidence . Law and Society Review, 427-498 . Schmdlders, G. (1959) . Fiscal psychology : A new branch of public finance . National Tax Journal, 340-345 . Schmblders, G. (1970) . Survey research in public finance - a behavioral approach to fiscal theory . Public Finance. 300-306 . Truman, D . (1951) . The Governmental Process : Political Interests and Public Opinions . New York : Knopf. Tyler, T . (1986). R . Lau & D . Sears (Eds), Justice and Leadership Endorsement . Political Cognition . Erlbaum : Hillsdale, NJ .

114


APPENDIX A RESEARCH INSTRUMENT {cover letter) Dear xxx : We are currently conducting research that investigates how CPA's interpret Statement on Responsibilities in Tax Practice (1988 Revision) No.], which provides CPA's with guidance when recommending tax return positions to clients . Tax accountants often encounter situations where the tax law is unclear as to how an item should be handled on a client's return . The Tax Division of the American Institute of Certified Public Accountants revised the Statement on Responsibility in Tax Practice No . I to provide that a CPA may recommend a tax return position without disclosure if it meets the "realistic possibility of success" standard . This means that "the CPA has a good faith belief that the position has a realistic possibility of being sustained administratively or judicially on its merits if challenged ." Accountants, of course, are faced with the challenge of interpreting this standard and applying it to situations they encounter in practice . The purpose of this research is to determine how CPA's interpret the realistic possibility standard. Please help us by completing the enclosed three-part questionnaire and returning it in the self-addressed return envelope at your earliest convenience . It should take only 10-15 minutes to complete the questionnaire . All your responses are confidential and the results will be aggregated . Thank you in advance for your participation . Sincerely, xxx

115


SECTION I - WORTHLESS SECURITIES DEDUCTION CASE IRC Sec . 165(g) allows a loss deduction for securities that become worthless during the current tax year. To qualify for this loss deduction, a taxpayer must be able to establish that both the security is worthless and that it became worthless during the claimed year of loss . Some factors that the courts have considered in resolving disputed worthless stock deductions are : insolvency, receivership, liquidation, cessation of business activities, market for the firm's principal product(s), opinions regarding future profitability, and the market for the firm's stock . Assume a client wants to claim a worthless stock deduction as long as he can AVOID A CHALLENGE from the IRS . [Alternative version : Assume a client has expressed a STRONG DESIRE to claim a worthless stock deduction] The stock was purchased in 1985, in an initial offering from a small electronic control manufacturer specializing in high-tech defense, medical, and industrial applications . From 1986 to 1990, the firm reported modest operating profits and the stock's price has never fluctuated by more than 10% . In light of the client's expressed desire and the realistic possibility standard please indicate the earliest year you would claim a worthless stock deduction of $7,500 as an ordinary loss under IRC Sec . 1244 for a taxpayer who has $75,000 of taxable income without including this deduction . If you conclude that the taxpayer should claim the worthless stock deduction in any of the following fact situations answer "Yes" for that year and do not consider the fact situations for any subsequent year(s) . Then proceed directly to Section II of the questionnaire . In 1991, another firm is granted a patent on a product that is superior to the

principal product of your client's firm and because of this competition, the firm records its first operating loss in history . Management is confident that intensified research and development (R&D) efforts will make the firm's product competitive in the future . Is it appropriate to claim the deduction in 1991?

Yes

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100 ; where 0 = no confidence, and 100 = complete confidence) :

116

C. F . KELLIHER, D. BANDY AND A . J . JUDD

In 1992, operating losses continue, the firm's research and development efforts

have produced no marketable results, the firm's cash flow and borrowing potential are not sufficient to fund the additional R&D that management feels will produce favorable results by next year . Shareholders are offered additional shares of stock to provide the needed R&D funds . No shareholders purchase any additional shares of stock . The only individuals providing funds to the corporation during 1992 are the majority shareholder and top management who loaned the corporation sufficient funds to continue operations and the R&D efforts . Is it appropriate to claim the deduction in 1992? _

Yes

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100 ; where 0 = no confidence, and 100 = complete confidence) : In 1993, the firm ceases its manufacturing operations but continues in its R&D

efforts . Some operating assets are sold to provide needed R&D funds but creditors are successful in having the firm placed in receivership to prevent the sale of more assets . This year the majority shareholder is the only individual who loans additional funds to the corporation, but top management and some R&D personnel agree to reduced compensation in exchange for stock that allows the firm to stay solvent. Is it appropriate to claim the deduction in 1993?

Yes

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100 ; where 0 = no confidence, and 100 = complete confidence) : In 1994, the R&D efforts result in the firm being granted a patent that will

allow it to produce a competitive product, but the majority shareholder does not have the funds to begin production . Reservations about the market life of this patent inhibit other current and prospective shareholders from contributing additional funds to begin production . The receiver decides to liquidate the firm to pay off the creditors' claims leaving the current shareholders with an interest in the patent that they try to market. Several potential purchasers have made serious inquiries about the patent, but no firm offer is received . Is it appropriate to claim the deduction in 1994? - Yes

No.


Realistic Possibility Standard's Influence on Tia Practitioner Aggressiveness

117

During 1995, a purchaser for the patent is found, but while negotiating for the

sale another firm is granted a patent superior to the one owned by the shareholders . The sale never takes place and all shareholder efforts to recover their investment cease . Is it appropriate to claim the deduction in 1995?

Yes

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100 ; where 0 =no confidence, and 100 =complete confidence) :

118


SECTION II - EDUCATION DEDUCTION CASES (In random order) The client is attending graduate school and wants to deduct the education expenses he is incurring as long as he can AVOID A CHALLENGE from the IRS . CASE 1 - Research has identified 3 trial level court cases in which taxpayers in similar situations were denied deductions for such expenses and 2 where such expenses were allowed . Is it appropriate to claim the deduction?

Yes

No .


CASE 2 - Research has identified 7 trial level court cases in which taxpayers in similar situations were denied deductions for such expenses and 3 where such expenses were allowed. Is it appropriate to claim the deduction?

Yes

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100; where 0 = no confidence, and 100 = complete confidence) : .

CASE 3 - Research has identified 9 trial level court cases in which taxpayers in similar situations were denied deductions for such expenses and I where such expenses were allowed . Is it appropriate to claim the deduction?

Yes

. No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100; where 0 = no confidence, and 100 = complete confidence) :

CASE 4 - Research has identified 4 trial level court cases in which taxpayers in similar situations were denied deductions for such expenses and I where such expenses were allowed .


Is it appropriate to claim the deduction?

Yes

119

No .

Indicate the level of confidence that your decision complies with the realistic possibility standard (enter a number between 0 and 100 ; where 0 = no confidence, and 100 =complete confidence) :

CASE 5 - Research has identified I trial level court case in which taxpayers in similar situations were denied deductions for such expenses and I where such expenses were allowed . Is it appropriate to claim the deduction?

Yes .

No .


120


SECTION III - DEMOGRAPHIC INFORMATION Also, please answer the following questions to help us with the study . 1 . Sex : _Male -Female . 2. How many years of tax experience have you had? _ (years) . 3 . Current employment : 6 CPA Firm Big Law Firm Regional CPA Firm Self-employed Local CPA Firm Other (specify) 4 . Current position and/or title : 5 . When dealing with client issues where the tax law is unclear, in the majority of cases I normally (select one) : resolve the doubt in favor of the client. ask my supervisor to decide how the matter should be treated . resolve the doubt in favor of the IRS . other (please specify) : 6 . Assume there have been 10 cases dealing with an issue . In a typical situation, how many of the trial-level court cases must have been decided in favor of the taxpayer in order to meet the "realistic possibility" standard? (please enter a number between 0 and 10) . 7 . Indicate your familiarity with the realistic possibility standard (circle the number that applies) : 1 None

2

3

4

5

6

7 Very Familiar


121

8 . Indicate your familiarity with worthless stock deductions (circle the number that applies) : 1

2

3

4

5

None

6 7 Very Familiar

9 . Indicate your familiarity with education deductions (circle the number that applies) : 1 None

2

3

4

5

6

7 Very Familiar

THE USE OF THE LATENT CONSTRUCTS METHOD IN BEHAVIORAL ACCOUNTING RESEARCH: THE MEASUREMENT OF CLIENT ADVOCACY J. David Mason and Linda Garrett Levy

ABSTRACT This article presents procedures that can be followed in behavioral research when indirect measures are needed for a variable of interest . It is hoped that these procedures will be used to develop a database of scales that can increase generalizability across accounting studies . The suggested framework was used to develop a measure of client advocacy, an important construct when studying the judgments and decisions of tax professionals . This measure of client advocacy may be defended for both its reliability and its validity, and the usefulness of this scale is illustrated in two studies. Davis and Mason (1997) tested whether different levels of advocacy affected judgments of similarity in tax authority evaluation . Including advocacy allowed for a more specific interpretation of their results . Measuring advocacy at two points allowed Levy (1996) to examine whether advocacy is an inherent attitude or can be influenced by other variables, such as client preferences .

Advances in Taxation, Volume 13, pages 123-139 . Copyright © 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved . ISBN : 0-7623-0774-9 123

124

J . DAVID MASON AND LINDA GARRETT LEVY

INTRODUCTION The field of behavioral tax research continues to grow . The validity of the research findings derived from affective scales used in the studies depends on the validity of the measures employed . As such, it is important that variables of interest be carefully measured and that common scales for important constructs be implemented . Variables integral to tax studies include taxpayers' propensity to follow their accountants' advice, taxpayer audit and penalty risk preference, tax preparer audit and penalty risk preference, and client advocacy . Unfortunately, the measures used to quantify these and other tax variables are limited, and single-item measures continue to be used . In mature fields such as consumer research, an existing database of scales is available for researchers (e .g . Bruner & Hensel, 1991) . The advantage of using such a database of scales is increased generalizability across studies . Because no such database of scales currently exists for accounting tax researchers, scale development is an important issue for behavioral tax research . The tax environment includes unique aspects, such as the existence of a formal and well-defined evidence base, the need for persuasion and defensibility in research, and infrequent outcome feedback (Davis, 1995) . The role of tax professionals as client advocates is central to tax practice ; however, client advocacy may introduce bias into professionals' judgment and decision making. Roberts (1998) noted that in each of the nine tax practitioner judgment and decision making studies which included client advocacy as a variable, client advocacy significantly affected judgments and decisions . Clearly some measure of the professional's attitude toward the client is an important component of the tax professional's judgment and decision making processes . Despite the demonstrated importance of client advocacy in studies of tax professional judgment and decision making, the use of different measures has limited comparability of results . For example, Johnson's (1993) scale focussed on understanding the legal requirements of advocacy, while Bums and Kiecker's (1995) focussed on ethical orientation of the professional . Literature regarding scale development is scattered in different sources, hindering researchers from efficiently following the recommended steps of scale development (Churchill, 1979) . The purposes of this article are : (1) to present in one source a research method that can be used to develop effective measurement scales, and (2) to illustrate the construction of a scale to measure client advocacy . This guide could serve as a tool for future researchers, increasing the comparability across tax studies . The scale described in this paper represents an advance over prior advocacy scales because its focus has been refined to that of loyalty to the client . Further,

The Measurement of Client Advocacy

125

this scale improves on prior measures because it is a multi-item measure yet explains more variance in responses using fewer items than previous scales .' Finally, the results of two behavioral tax studies that used this advocacy scale are reported . The remainder of this article is organized as follows . The application of the attitude scale method to behavioral research is demonstrated by reviewing the development of a scale for client advocacy . The results of two studies that used this advocacy scale are presented . Finally, conclusions, limitations, and suggestions for future research are presented .

MEASURING LATENT CONSTRUCTS Background of the Problem

Unlike the physical sciences where measurement is largely a matter of finding satisfactory physical instruments to measure the phenomenon of interest (e .g . the speed of a computer), social scientists are interested in measuring not only actual events and behaviors, but also the contents of processes of people's minds . In such situations, it is necessary to ask people to respond to stimuli and then measure their responses to such stimuli . The peculiar problems involved in constructing scales to measure the contents of people's minds make scaling an important and special topic in the social sciences (Simon & Burstein, 1985) . It is the attributes of objects, not the objects themselves, that are measured (Nunally & Bernstein, 1994) . The rigor and skill with which the measure is developed may determine whether the construct has been captured by the measure (Churchill, 1979) . Given an imperfect world, attributes that are not directly observable, and imperfect measures, it is more likely that differences between individuals on the measure may reflect factors other than differences between individuals on the true attribute. Both random and/or systematic factors may account for the differences (Churchill, 1979) . Examples that may explain differences between the measure and the true object include : (1) true differences in other relatively stable characteristics which affect the score (e .g . a person's willingness to express his or her true feelings) ; (2) differences due to transient personal factors (e .g . a person's mood or state of fatigue) ; (3) differences due to situational factors (e .g. where the interview is conducted) ; (4) differences due to variances in administration (e .g . instructions given by administrator) ; (5) differences due to sampling of items (e .g . the specific items used on the questionnaire) ; (6) differences due to lack of clarity of measuring instruments (e.g . vague or ambiguous questions which are interpreted differently by those responding) ;

1 26


and (7) differences due to mechanical factors (e .g . a check mark in the wrong box or a response which is coded incorrectly) (Selltiz, Wrightsman & Cook, 1976) . A measure is valid when the differences in observed scores reflect true differences in the characteristic one is attempting to measure . If independent but comparable measures of the same construct of a given object agree, the measure is reliable (i .e . random error is zero) . Note that a valid measure is reliable, but a reliable measure is not necessarily valid . The fundamental objective therefore is to develop a measure such that the measured scores approximate the true scores as closely as possible . Because constructs are not observable, the researcher never knows what the true scores are, and the measures are necessarily inferences . The quality of these inferences is critical and depends directly upon the procedures used to develop the measures and the evidence supporting their "goodness" . This evidence typically takes the form of a reliability or validity index . Developing a Scale to Measure Client Advocacy The procedures followed in developing multi-item measures should provide a valid basis for the inferences that are made . The basic steps include specifying the construct domain, generating items and validating content, purifying the measure, assessing reliability, assessing construct validity, and developing

Techniques or Recommended Coefficients

steps 1 . Specify the Domain of the Construct

Literature search

2 . Generate Sample of Items

Literature search Experience survey

3 . Purify the Measure

Coefficient alpha Item-to-total correlations Factor analysis

4 . Assess Reliability with New Data

Coefficient alpha Item-to-total correlations Factor analysis

5 . Assess Construct Validity

Criterion validity

6 . Develop Norms

statistics for summarizing distribution of scores

Fig . 1 .

Steps In Developing Latent Contructs Measures .

The Measurement

of Client

Advocacy

1 27

norms (Churchill, 1979) . The steps are summarized in Figure 1 . The utility of each step in developing a "good" measure is generally undisputed, although differences may exist among researchers as to how to carry out each step . The following discussion illustrates how the scale development steps were used to develop a client advocacy scale . Specify the Domain of Construct

Because the strength of the inferences is linked to the initial step of specifying the construct domain, it is critical that the researcher be exacting in defining the construct. This requires a careful review of the literature. In addition, it is important to verify that the definition of the construct developed by the researcher is consistent with the views of professionals knowledgeable in the field . To formulate a definition of client advocacy, practitioner publications (e .g . Blaustein, 1982), official pronouncements of the AICPA (e .g . Statements of Responsibilities of Tax Practice), and academic publications (e .g . Ashley, 1991 ; Johnson, 1993) were consulted . These publications suggested that the basic notion was one of primary loyalty to the taxpayer . This was emphasized in both the official pronouncements and the academic literature . After having developed the construct, the author-generated definition was reviewed by four tax professionals . The tax professionals concurred that the following definition reflected their understanding of the concept : Advocacy is a state of mind in which one feels one's primary loyalty belongs to the taxpayer . It is exhibited by a desire to represent the taxpayer zealously within the bounds of the law, and by a desire to be a fighter on behalf of the taxpayer .

Generate Items and Validate Content

Once the construct has been defined, the creation of items that will be used to measure the construct may begin . This involves generating a pool of items or statements designed to capture the specified domain . For exploratory research, techniques such as literature searches, experience surveys, and insight-stimulating examples may be productive in developing the initial pool . The author as well as others knowledgeable in the area may generate these items . In the early stages of item generation, developing items that tap each dimension of the construct at issue should be emphasized . Toward the end of this stage, the focus should shift to item editing to make the statements as precise as possible . After further editing, refinement would await actual data from pretesting . As a rule of thumb, the edited pool of items should contain no less than 25 items for each dimension of the construct . The advocacy construct as defined was believed to contain a single dimension, client loyalty . Thirty statements were generated as the initial pool of items,

128

I . DAVID MASON AND LINDA GARRETT LEVY

15 by the scale developer and 15 using an experience survey . The experience survey involved a tax professor and a tax attorney, both with tax practice experience, who offered insights and ideas into the phenomenon . As a result of the experience survey, one statement was eliminated due to similar wording, resulting in an initial pool of 29 items. The 29 items were edited further to increase the precision of the items and to ensure that approximately half of the items in the pool were negatively worded . This is done to reduce `yea-' or `nay-' saying tendencies (Churchill, 1979) . At this point in the development of the measure, the initial pool of items should be evaluated for content validity (Zaichkowsky, 1985) by having expert judges rate each item on how well it represents the construct . The expert judges should be asked to rate each item as very representative, somewhat representative, or not representative of the construct . Items that are rated as not representative should be dropped from the pool . Although no set number of expert judges is required, it is common practice to use from three to five judges (e .g . Zaichkowsky, 1985 ; Lichtenstein et al ., 1990) . A panel of four expert judges judged the 29 items in the client advocacy measure for content validity . Two of the judges were Ph .D . students in accounting with extensive professional backgrounds in tax . The other two judges were accounting professors whose primary research and teaching areas are tax . The four judges were given the conceptual definition of advocacy and then asked to rate each of the 29 items as either "very applicable," "somewhat applicable," or "not applicable" to the advocacy definition . In addition, the judges were asked to modify wording, as they believed necessary . The following a priori retention rule was used (Bearden et al ., 1989) : if three of the four judges rated an item as at least "somewhat applicable," the item was retained . This procedure resulted in the retention of 22 of the 29 items in the pool . Based upon suggestions from the judges, the 22 remaining items were edited further to clarify ambiguous or unclear wording .

Purify the Measure After carefully editing the pool of items, the next step is to collect data using the measure . These initial data are used to further refine the measure and test its internal consistency . Purification usually results in the elimination of items . The first recommended measure of internal consistency is the coefficient alpha (Nunnally, 1967 ; Churchill, 1979 ; Normally & Bernstein, 1994) .' Coefficient alpha is the estimated correlation of the test with errorless true scores . Thus, the lower the coefficient alpha, the poorer the sample of items performed in capturing the construct of interest . A high alpha indicates that the test correlates well with true scores . Conversely, a low alpha suggests that some

The Measurement of Client Advocacv

129

of the items in the pool do not relate to the common core and should be eliminated . Determining which items are not related may be accomplished by calculating the correlation of each item with the total score and plotting the correlations by decreasing order of magnitude (Churchill, 1979 ; Nunnally & Bernstein, 1994) . Items with low correlations are eliminated . In addition, items that exhibit a substantial or sudden drop in the item-to-total correlations should be deleted as well . This purification process provides preliminary estimates of reliability and validity . For the advocacy scale, the 22 items were administered for purification to a pretest sample of 34 public accountants from local CPA firms and 30 Internal Revenue Agents . The surveys were hand delivered or mailed to the respondents after they were contacted by phone to request their participation . Completed surveys either were picked up by or mailed to the author . Thirty-three of the 34 tax practitioner CPAs completed and returned the survey ; 20 of the 30 Revenue Agents completed and returned the survey . The participants indicated the extent of their agreement with each item on a 7-point Likert scale, with I representing strongly disagree and 7 representing strongly agree . The analyses performed on the pretest data followed the recommendations discussed above . First, item-to-total correlations and Cronbach's alpha were examined . Items were sequentially eliminated until the removal of an item no longer produced an increase in the coefficient alpha . This procedure eliminated 7 items with corrected item-to-total correlations of less than 0 .40 (Lichtenstein et al ., 1990) . In the second step of purification, the reliability and structure of the remaining 15 items were determined through principal components factor analysis . Factor analysis provides empirical confirmation of the number of dimensions conceptualized in the construct (Churchill, 1979) . However, care must be exercised in its use and interpretation because it can generate spurious results, especially if performed before the purification steps suggested above (Nunnally & Bernstein, 1994) . No restrictions were placed on the number of factors to be extracted. For the 15 items, two factors were extracted accounting for 51 .5% of the variance. Items then were eliminated until all remaining items had loadings of greater than 0 .60 (Netemeyer et al ., 1993) on a given factor for both the principal components and a varimax rotated factor pattern . Six of the 15 items were eliminated during this process . The remaining 9 items all loaded on a single factor explaining 51 .6% of the variance . For these 9 items, Cronbach's alpha was 0 .88, To summarize, in the purification stage good variability is desired, and 13 of the 22 items were eliminated, leaving a scale with 9 items . The mean score on the 9 item scale was 44 .4, with a standard deviation of 8,51 . The scores ranged from 19 to 59 . The maximum possible score was 63, and the minimum

130


I . In an instance where no judicial authority exists with respect to an issue and where the Code and Regulations are ambiguous, I feel that the taxpayer is entitled to take the most favorable tax treatment . strongly disagree 1

strongly agree

2

3

1

4

I

5

6

7

2 . Generally speaking, my loyalties are first to the tax system, then to the taxpayer . strongly disagree

CI

2

strongly agree

1

3

1

I 4 1

6

7 J

3 . 1 feel I should apply ambiguous tax law to the taxpayers benefit. sagree

trongly

strongly agree

6 I 7

1

4 . When examining a tax return, I tend to point out to taxpayers reasonable positions they could have taken which would have contributed to minimizing their tax liability . strongly disagree

r ,

strongly agree 6

-L-2 -L,

7

5 . I believe it is important that I encourage taxpayers to pay the least amount of taxes possible . strongly agree strongly disagree ~, 4 2

T- 5 _F6

17

6 . 1 always interpret unclear/ambiguous laws in favor of the taxpayers . strongly disagree .. ~

1

strongly agree

2

3

4

5

7.

.J

7 . It is important to use trends in the law by trying to establish a pattern of more favorable treatment for the taxpayer and then extending this pattern to the taxpayer's position . strongly agree

strongly disagree

I

2

-

3

1

4

5

6

7 l

8 . Where no judicial authority exists with respect to an issue, I feel that the taxpayer is entitled to take the most favorable tax treatment . strongly disagree

1

1

strongly agree

2

3

4

5

1

6

7

9 . The taxpayer has the right to structure transactions in ways that yield the best tax result, even if the law is unclear in an area. strongly disagree I

I

L

2

strongly agree 1

3

1

4 Fig . 2 .

5

7 Advocacy Instrument .


131

possible score was 9 . This 9 item advocacy scale was administered to participants in a subsequent study to assess reliability with a different sample (see Fig . 2) . Assess Reliability with New Data Although coefficient alpha is the basic statistic for determining the reliability of a measure based on internal consistency, it does not adequately estimate errors caused by factors external to the instrument, such as differences in testing situations or in respondents over time . A comparison of the results of the initial sample with the results of a subsequent sample may operate as a measure of the external consistency of the instrument . If the construct is more than a measurement artifact, the results should be reproduced when the purified instrument is submitted to a new sample of subjects . The purified advocacy scale was administered to 150 participants in a subsequent study (Davis & Mason, 1997) . The responses of 3 subjects had missing information and were dropped from the analysis . Of the 147 usable responses, 58 participants were from the IRS and 89 were from public accounting firms . The analysis performed on the responses followed the same approach used on the initial data . In the first step, item-to-total correlations and Cronbach's alpha were examined . All items had corrected item-to-total correlations of greater than 0 .40 (Lichtenstein et al., 1990), and Cronbach's alpha for the 9 items was 0 .83 .3 These results provide support for the assertion that the results in the pretest were not due to mere chance and that the construct is more than a measurement artifact . In the second step, the reliability and structure of the 9 items was determined by using a principal components factor analysis with no restrictions placed on the number of factors extracted . Two factors were extracted accounting for 57 .10% of the variance. Recall that the 9 items loaded onto a single factor in the pretest . An examination of the individual factor loadings for the two factors showed that all 9 of the items loaded on the first factor with loadings of 0 .53 or greater . Only one item suggested a second factor . Item 2 had a factor loading of 0 .53 on factor one and 0 .54 on factor two, suggesting that Item 2 might be providing information on more than one factor . However, the conclusion that all 9 items load on a single factor was not rejected because the Eigenvalues for factor I was much greater than that for factor 2 (3 .88 vs . 1 .27) and all 9 items had loadings of 0 .53 or better on factor 1 .° Assess Construct Validity Using the above steps should produce a measure that is reliable and has content validity . However, it may or may not produce a measure that has construct

132


validity . To establish construct validity, the researcher must determine : (1) the extent to which the measure correlates with other measures designed to measure the same construct and (2) whether the measure behaves as expected . Given that there is only one other multi-item measure of advocacy in the literature (Johnson, 1993) and the domain of the construct for that measure was not the same as that used here, it was not possible to determine the extent to which the measure in the current study correlated with that of Johnson (1993) . 5 However, another way of judging construct validity is to test whether the measure behaves as expected . One way of testing this is seeing if it can differentiate known groups . The data used to develop the measure in the current study included information from both IRS agents and public accounting tax practitioners, making it possible to determine whether the advocacy scale could distinguish between the groups . It was expected that tax practitioners would score higher on the advocacy scale than would IRS agents . This expectation was based upon professional accounting and Internal Revenue standards . For example, CPAs have an ethical duty to

Table 1 .

Descriptive Statistics for Advocacy Measure .

Panel A : (Davis & Mason, 1997) Group

n

Mean

Standard Deviation

Minimum

Maximum

89 58 147

4603 36 .29 42 .19

7 .74 7 .52 9 .00

24 15 15

63 56 63

n

Mean

Standard Deviation

Minimum

Maximum

107 15 122

53 .9 53 .8 53 .9

6 .5 6 .2 6 .4

24 45 24

63 63 63

Group

n

Mean

Standard Deviation

Minimum

Maximum

CPA firm participants Industry and other Total

107 15 122

53 .3 54 .3 53 .4

6 .4 5 .8 6 .4

24 45 24

63 63 63

CPA firm participants IRS participants Total Panel B. Measure I (Levy, 1996) Group CPA firm participants Industry and other Total Panel C: Measure 2 (Levy, 1996)


133

function as advocates of the client in tax matters (AICPA, 1992) . In contrast, the Internal Revenue Manual provides that IRS agents have an ethical duty to " . . . perform this work in a fair and impartial manner, with neither a government nor a taxpayer point of view ." (Internal Revenue Manual §4015 .1(1) and §4015 .3) The advocacy scores support the prediction that public accounting practitioners are stronger taxpayer advocates than are IRS employees . Table 1 (Panel A) reports descriptive statistics for the first administration of the advocacy measure . CPA tax practitioners scored significantly higher on average than did IRS agents (46 .03 vs . 36 .29, respectively ; t-statistic = 7 .54, p < 0 .001) . In addition, the range of responses was lower for IRS agents (15 to 56) than for public accounting tax practitioners (24 to 63) . The ability of the advocacy scale to distinguish between public accounting practitioners and IRS personnel supports the construct validity of the measure . These results can be used as the initial data upon which norms for this measure may be developed . Develop Norms Raw scores typically are not very informative about the characteristic being measured . To gain more information, norms should be developed which allow an individual's score to be compared to the score achieved by others . Churchill (1979) suggests calculating a mean and standard deviation or other statistics indicative of central tendency . The need for such norms is particularly strong when new measurement instruments are introduced . As results from replications of the measure are reported, the norms will continue to be refined . For the client advocacy measure, the initial results suggest that CPA firm participants have an average advocacy score of 46 and IRS participants an average of 36 . As additional replications are performed, these norms will be refined and the extent of their generalizability may be determined . For example, as discussed in the next section, public accountants in a second study using the advocacy scale averaged 53 . The different types of participants in the two studies may explain the difference in the scores, and a reconciliation of these results is presented in the next section of the paper . The results of two studies in which researchers examined the influence of advocacy on judgment, discussed in the next section, provide examples of how the advocacy measure can be used . Figure 3 contains a detailed summary of the steps and sub-steps of the process . Carefully following all procedures should help produce valid construct measures .

USE OF ADVOCACY SCALE IN BEHAVIORAL STUDIES The potential influence of differences in client advocacy on tax professionals' information processing and judgments should be measured and controlled in

134


Specific Steps

Implementation

Specify the Domain of the Construct :

• Formulate definition of client advocacy • Determine if definition is consistent with professionals' perception of concept

• Consulted practitioner publications, textbooks, and official pronouncements • Tax professionals reviewed definition

Generate Sample of Items :

• Develop initial pool of items capturing definition of client advocacy • Evaluate initial pool of items for content validity • Clarify ambiguous or unclear items

• Generated 30 statements as initial pool of items • half generated by researchers • half generated from experience survey • Expert judges rated each item on how well it represents construct ; non-representative items dropped, resulting in refined initial pool of 22 items • Items were edited

Purify the Measure :

• Collect data using measure • Analyze data collected • Examine reliability and structure of remaining items

• Items administered to pretest sample (completed by 53 subjects) • Subjects indicated extent of agreement with each item • Examined item-to-total correlations and Cronbach's alpha • Eliminated items until removal of an item did not increase coefficient alpha • Procedure eliminated 7 items • Conducted principal components factor analysis • Eliminated items with low factor loadings • After this step, 9 items remained

Assess Reliability with New Data :

• Compare results of initial sample with those • Purified advocacy scale administered to new of another sample sample (147 usable responses obtained) • Examined item-to-total correlations and Cronbach's alpha • Conducted principal components factor analysis Fig. 3 .

Steps In Developing Client Advocacy Scale .

135


SDecific Steps

Implementation

Assess Construct Validity : • Determine extent to which this measure of client advocacy correlates with other measures of client advocacy • Determine whether measure behaves as expected

• Not possible given lack of client advocacy scale using common domain

• Data used to determine if advocacy scale could distinguish between IRS agents and public accounting practitioners

Develop Norms : • Calculate mean or other sta stics indicative of central tendency

Fig. 3 .

• initial results indicate mean advocacy score of 46 for public accounting practitioners and 36 for IRS agents

(continued) . Steps In Developing Client Advocacy Scale .

most experimental studies of tax accountants' judgment and decision making (Roberts, 1998) . Two studies have used the advocacy scale discussed in this paper to control for the effects of client advocacy . One study, Davis and Mason (1997), formulated and tested a similarity model of tax authority evaluation . Participants in the study completed a task that required the application of case law to a client fact situation . The manipulated independent variables were outcome of precedent (the outcome of the court case in the task materials ; either favorable or unfavorable to the client) and distinctive features (whether the taxpayer situation or the court case contained a high number of distinctive features) . Two independent variables were measured : client advocacy (measured using the scale described in this paper) and the direction of comparison (whether the participants compared the case to the taxpayer situation or vice versa ; measured by participants' responses) . The dependent variables were the importance of each feature (operationalized by having subjects weight each feature in the case on a 7-point Likert scale), the similarity of the court case to the client situation (operationalized by having participants indicate the degree of similarity on a 7-point Likert scale), and the likelihood of success (operationalized by having participants fill in a number from 0 to 100 which represented their determination of the likelihood of success in court if the clientfavored treatment were challenged) . Davis and Mason used the similarity model to predict that the weighting of common relative to distinctive features would be higher (lower) at higher levels

136


of advocacy when the court case's outcome was favorable (unfavorable) . They found support for the predicted interaction, but only for common features . In other words, advocacy affected similarity judgments in the following way . The influence of a favorable outcome for a court ruling on the common feature weights increased (decreased) as the level of client advocacy increased (decreased) . The use of the client advocacy scale allowed Davis and Mason (1997) to carefully test whether different levels of advocacy affected judgments of similarity in tax authority evaluation . As previously stated, advocacy did affect the aggressiveness of tax authority judgments, and the inclusion of advocacy in that study allows for more specific interpretation of these results . The common feature weights of strong advocates were more influenced by favorable precedent than those of weak advocates . The second study to use the 9-item client advocacy scale was Levy (1996) . Levy proposed a tax authority evaluation model that distinguished between evidence strength judgments, reporting position conclusions, and client recommendations . Participants in the study made judgments on a simulated client fact situation based on given authority (including statutory and case law) . The manipulated independent variables included penalty standard (high or low level of support required to take tax return position without penalty), documentation (half the subjects were required to write a memo explaining their conclusions and half were not), and client preference (the client was described as conservative or aggressive) . Client advocacy was measured using the scale described in this paper. Participants completed the scale at two points in the study . The first measurement was of participants' "normal" advocacy attitude . The second measurement was of participants' advocacy attitude given a specific client risk preference . Participants responded to three dependent variables : evidence strength rating (similar to the likelihood of success judgment in Davis and Mason), reporting position judgment (operationalized as the "correct" reporting position on a 7-point Likert scale), and recommendation to the client (subjects chose either the client-favored or disfavored position) . Levy predicted that strong advocates would rate similar evidence as being more favorable to clients than would weak advocates, would make more aggressive conclusion judgments, and would be more likely to recommend the client-favored position than would weak advocates . Measuring advocacy at two points allowed an examination of whether advocacy attitudes are inherent or can be influenced by other variables, such as client preferences . Levy found that the second measure of advocacy had a significant positive effect on evidence strength ratings (F = 7 .06, p = 0.01) and a marginally significant positive effect on reporting position judgments (F = 3 .57, p = 0 .06) . The higher a subject


137

scored on the advocacy scale measuring advocacy attitude given client preferences, the higher he or she rated evidence strength and the more aggressive reporting position judgments he or she made. The second advocacy measure also had a significant positive effect on recommendations (Wald test statistic = 3 .70, p = 0 .05) . As advocacy scores increased, subjects tended to make more aggressive choices . Advocacy does affect tax research judgments, generally increasing practitioners' favorable interpretation of evidence . However, practitioners seem able to put aside their natural advocacy attitudes in favor of those desired by the client . This finding could serve as evidence that practitioners do not influence their clients to take aggressive positions as suggested by the IRS (Shapiro, 1987) . Panels B and C of Table 1 report descriptive statistics for the advocacy measure . The mean advocacy score was higher than that reported for the public accounting participants in Davis and Mason (1997) . This could be due to the difference in the type of participants in the two studies . The participants in Davis and Mason (1997) included participants from national, regional, and local public accounting firms, while those in Levy (1996) were primarily (83%) sole proprietors . In addition, the subjects in Levy (1996) had more tax experience (average of 12 .5 years) than those in Davis and Mason (average of 6 .1 years) . It is possible that sole proprietors, especially those with more tax experience, identify more closely with their clients than do practitioners working within a firm,

CONCLUSION This article presents procedures that can be followed in behavioral research when indirect measures are needed for a variable of interest . These procedures are illustrated in the context of development of a scale to measure client advocacy . Finally, the use of the client advocacy scale in two behavioral tax studies is discussed . The results of the client advocacy scale development reported in this paper suggest that the latent constructs method can be an important and useful research tool that should not be overlooked when seeking to develop measures of variables . Using this method results in measures that can be defended as reliable and valid . As with any research method, there are shortcomings to its use and care must be exercised in its application . However, as a result of following the suggested framework, the behavioral researcher now has a measure of client advocacy that may be defended not only for its reliability, but also for its validity . The usefulness of this advocacy measure was illustrated in two behavioral studies . Accountants in tax practice could use the client advocacy scale

138


to assess the degrees of client loyalty exhibited by their employees . This would allow them to make sure that employees understood the appropriate client loyalty position within their firms . Several avenues exist for future research . Further validation of the advocacy measure presented in this article should be conducted by including the advocacy measure in future studies where client advocacy is a variable of interest or where it should be included as a control variable . In addition, the suggested scale development method can be used to develop measures of other variables of interest. In the tax area, such variables may include taxpayers' propensity to follow their accountants' advice, audit risk preference, and penalty risk preference .

NOTES 1 . For example, Johnson's (1993) measure uses 18 items, while this scale uses nine . Cuccia and McGill (2000) used Johnson's scale and reported that responses loaded on one factor explaining 25% of the variance . The scale described in this article loaded on one factor in two studies and explained between 41% and 52% of the variance . 2 . This is based upon the domain sampling model which is the most logically defensible one for basic research . This model holds that the purpose of any particular measurement is to estimate the score that would be obtained assuming all the items in the domain are used . Generally, though, only a sample of the items is used . To the extent the sample of items correlates with true scores, it is a good sample . The key assumption is that all items, if they belong to the domain of the construct, should have an equal amount of common core, and thus should be highly correlated . Low correlation between items would indicate some items are not from the appropriate domain (Nunally 1967) . 3 . The drop in Cronbach's alpha should be investigated in further validation of the scale . 4 . The fact that Levy (1996) found a single factor bolsters the original conclusion of a single factor . Factor analysis confirmed that one factor underlies the 9-item scale, and coefficient alpha for the measure was 0 .80 . 5 . Although Johnson (1993) measured client advocacy, the construct measured in that study differed from the construct measured here . Examining the items on Johnson's scale suggested that the construct of interest was more focused on understanding the legal requirements of advocacy than client loyalty as construed here . To measure this dimension of client advocacy, a new advocacy scale was developed .

ACKNOWLEDGMENTS The authors appreciate the suggestions of the discussant and participants at the AAA Annual Meeting in New Orleans . This paper is based on data obtained as part of each author's doctoral dissertation at the University of Colorado at Boulder . The authors would like to acknowledge the advice and suggestions received from their respective doctoral dissertation committees, especially Jon Davis, Betty Jackson, and Marlys Lipe.


139

REFERENCES American Institute of Certified Public Accountants, Inc . (1992) . Statements on Responsibilities in Tax Practice (1988 Rev .) No . I (SRTP #1) . Ashley, K . (1991) . Modeling Legal Argument: Reasoning with Cases and Hvpothetic'als . Boston : MIT Press . Bearden, W . O ., Netemeyer, R, G ., & Teel, J . E . (1989) . Measurement of Consumer Susceptibility to Interpersonal Influence, Journal of Consumer Research, 15 (March), 473-48L Blaustein, R . (1982) . How To Do Business With The IRS: The Complete Guide For Tat Professionals . Englewood Cliffs, NJ : Prentice-Hall, Inc. Bruner, G ., & Hensel, P . (Eds) (1991). Marketing Scales Handbook, A Compilation of Multi-Item Measures . Chicago : American Marketing Association . Chronbach, L . (1951) . Coefficient alpha and the internal structure of tests . Psvchometrika, (September), 297-334 . Churchill Jr., G . (1979) . A paradigm for developing better measures of marketing constructs . Journal of Marketing Research, 16 (February), 64-73 . Cuccia, A ., & G . McGill . (2000) . The role of decision strategies in understanding professionals' susceptibility to judgment biases . Journal of Accounting Research, 38(2) (Autumn),419-435 . Davis, J . (1995) . A perspective on experimental tax research . Journal of the American Taxation Association, 17 (Supplement), 114-122 . Davis, J ., & Mason, J . D. (1997) . Similarity and Precedent in Tax Authority Judgment . Working Paper, University of Illinois at Urbana-Champaign . Johnson, L. (1993. An empirical investigation of the effects of advocacy on preparers' evaluations of judicial evidence . Journal of the American Taxation Association, 15 (Spring), 1 22 . Levy, L. (1996) . Evidence evaluation and aggressive reporting in ambiguous lax situations . Dissertation, University of Colorado. Lichtenstein, D. R ., Netemeyer, R . G ., & Burton, S . (1990) . Distinguishing coupon proneness from value consciousness : An acquisition-transaction utility theory perspective . Journal of Marketing, 54 (July), 54-67 . Likert, R . (1932) . A technique for the measurement of attitudes . Archives of Psychology (No . 140) . Netemeyer, R ., Burton, S ., & Lichtenstein, D . (1993) . Trail aspects of vanity : measurement and relevance to consumer behavior . Journal of Consumer Research, 21 (March), 612-627 . Nunnally . J . C. (1967) . Psychometric Theory . New York : McGraw-Hill Book Company . Nunnally, J . C ., & Bernstein, I . H . (1994). Psychometric Theory (3rd ed .) . New York McGrawHill Company . Roberts, M . (1998) . Tax accountants' judgment/decision-making research : a review and synthesis . Journal of the American Taxation Association, 20 (Spring), 78-121 . Selltiz, C ., Wrightsman, L . S ., & Cook, S . W. (1976) . Research Methods in Social Relations (3rd ed .) . New York : Holt, Rinehart, and Winston. Shapiro, L . (1987) . IRS Director of Practice Shapiro's comments on proposed changes to Circular 230. Tax Notes, 1150 .1155 . Simon, J . L ., & P . Burstein . (1985) . Basic Research Methods in Social Science (3rd ed .) . New York : Random House. Zaichkowsky, J . L . (1985) . Measuring the Involvement Construct . Journal of Consumer Research, 12 (December), 341-352 .

TAX PRACTITIONERS' WILLINGNESS TO TRUST CLIENTS : EFFECTS OF PRIOR EXPERIENCE, SITUATIONAL AND DISPOSITIONAL VARIABLES William Shafer

ABSTRACT This study investigates professional tax preparers' willingness to include questionable client-provided data in a tax return without verifying the information . Although a recent survey of tax practitioners by Yetmar et al . (1998) found that the issue of reliance on client data is a significant ethical problem in practice, a very limited amount of previous research has investigated the factors that influence tax practitioners' reliance decisions . In the current paper, the issue of client reliance is viewed as a problem of trust and suspicion, and the general model of trust and suspicion proposed by Kee and Knox (1970) is adopted as a conceptual framework for addressing this issue . Based on this model and previous research findings, it was hypothesized that client reliability, the client's year-end tax payment status, the general propensity to trust others, and tax practitioners' attitudes toward risk would influence reliance decisions . The findings, based on a study of CPA tax practitioners, indicate that client reliability and tax payment status each had a significant impact on client reliance . However, contrary to expectations, the general tendency to trust others did

Advances in Taxation, Volume 13, pages 141-167 . Copyright © 2001 by Elsevier Science Ltd . All rights of reproduction in any form reserved. ISBN : 0.7623-0774-9 141

142

WILLIAM SHAFER

not influence client reliance decisions in a tax context . The results provide mixed support for the hypothesized effects of tax preparers' risk attitudes on decisions . The findings of the study also raise questions regarding the appropriateness of tax practitioners' client reliance decisions .

INTRODUCTION Although many studies in the tax literature have investigated professional tax preparers' willingness to advocate aggressive return positions, very little empirical research has examined the issue of reliance on questionable data supplied by clients . However, recent evidence reported by Yetmar et al . (1998) suggests that the latter issue is one of the most significant ethical issues in tax practice . They surveyed a random sample of AICPA Tax Division members to determine the extent to which 54 issues were perceived as significant ethical problems . The results indicated that two of the top three ethical issues related to reliance on client-provided information . The highest rated issue was "accepting a client's deduction amount with partial or no documentation," while the third highest rated issue was "not determining the accuracy of oral or written representations made by the client ." Based on a factor analysis of data obtained from twelve tax managers and partners, Milliron (1988) also found that the perceived quality of client records and client dependability significantly affect tax reporting decisions . Taken together, these findings suggest that perceptions of client (un)reliability can pose a significant dilemma in tax practice . Despite the apparent significance of client reliability in tax reporting decisions, very little empirical evidence relating to this issue has been reported . Helleloid (1989) appears to be the only study that has addressed the effects of client dependability on tax reporting decisions . His study investigated the effects of the ambiguity associated with a client's auto mileage records on tax preparers' recommended mileage deductions, reporting mixed results . The current research extends the study of client reliance to explicitly address the issue of trust and suspicion in a tax setting . In contrast to the Helleloid (1989) study, which investigated the effects of perceived ambiguity of an (apparently) honest client's records, the current study investigates CPAs' willingness to rely on client data when the CPA has reason to question the client's honesty or credibility, and the client has an incentive to cheat on the tax return . To address these issues, the paper adopts the formal model of trust and suspicion proposed by Kee and Knox (1970) . Based on this model, it was hypothesized that perceptions of taxpayer reliability or trustworthiness, based on prior experience in dealing with the client, would have a significant impact on practitioners' willingness to include data in a tax return without

Tax Practitioners' Willingness to Trust Clients

1 43

examining supporting documentation . In addition, it was hypothesized that a client's tax payment status would affect CPA trust and suspicion, through its effect on perceived incentives to cheat . Finally, two dispositional or personality traits : (I) the general predisposition toward trusting behavior, and (2) CPA risk attitude, were hypothesized to influence client reliance decisions . The data were obtained by mailing instruments to a random sample of AICPA members in tax practice, who responded to two cases involving client reliance issues . The results indicate that, as hypothesized, perceptions of taxpayer trustworthiness based on client-specific experience significantly affected reliance decisions . Also, consistent with the research hypotheses, CPAs were more likely to require supporting documentation for deductions claimed by taxpayers who were in a tax due position at year-end, which suggests that tax practitioners recognize the incentive effects created by the client's year-end payment status . Contrary to expectations, the measure of propensity to trust others did not have a significant impact on reliance decisions, and mixed support was obtained for the hypothesized effects of risk attitudes on client trust . The results of this study suggest that the Kee and Knox (1970) model may provide a useful theoretical framework for the investigation of trust and suspicion among tax preparers . The findings also raise questions regarding the appropriateness of CPAs' client reliance decisions . Even in cases where the client had overstated deductions or understated revenues in the past and had current incentives to cheat, participants' mean estimate of the likelihood of relying on questionable data ranged from approximately 30 to 40% . It is suggested that future studies should try to obtain a better understanding of the factors that affect client trust in tax preparation settings, and attempt to assess the appropriateness of CPA tax practitioners' client reliance decisions . The next section provides a review of relevant literature and develops the research hypotheses . This is followed by discussions of the research method and empirical findings . The final section of the paper contains a discussion of the findings and limitations of the study, as well as suggestions for future research .

LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT General Model of Trust and Suspicion

The problem of reliance on questionable client data may be viewed as an issue of trust and suspicion . Rotter (1971 : 444) defined trust as :

144

WILLIAM SHAFER

. ., an expectancy held by an individual or a group that the word, promise, verbal or written statement of another individual or group can be relied upon .

Suspicion, on the other hand, may be viewed as the complement of trust (Shaub, 1996) . Kee and Knox (1970) defined a trust situation, broadly, as a scenario involving two parties who are interdependent with respect to certain outcomes defined by their joint choices . One of the parties (P) is confronted with the choice of either trusting or not trusting the other (0), who in turn has a choice of being either trustworthy or untrustworthy . Kee and Knox's basic conceptualization of trust and suspicion is presented in Fig . 1 . The model makes an explicit distinction between subjective and behavioral trust or suspicion . Subjective trust refers to P's assessment of the probability that 0 will be trustworthy, while behavioral trust refers to the manifest act of trust. According to Kee and Knox, a person may manifest complete trust in another until the assessed probability of trustworthiness falls below a critical threshold, at which point the probability of trusting behavior will decline . The level of this threshold will depend on situational variables, such as the risks and rewards associated with trusting behavior . Subjective trust or suspicion will be determined by P's perception of O's motives and/or competence . The model recognizes three types of independent variables that may influence these perceptions : (1) structural and situational factors, (2) dispositional factors, and (3) previous experience . Structural and situational variables include factors such as incentives, power, and characteristics of the trusted party . Rational choice models of trust recognize the influence of situational variables on trusting behavior . For example, Hardin (1992) suggested that a critical element of a trusting relationship is the incentive of the trustee to honor the trust . According to Hardin, a person will trust another only if he/she believes it will be in that person's best interest to be trustworthy . The influence of incentives on trust also is supported by research indicating that suspicion is often triggered by situational cues that raise questions about the motives of the trustee (Fein, 1996 ; Hilton et al ., 1993) . A rational choice model of trust would also suggest that the relative power of the parties will influence trusting behavior . For example, P will have a greater incentive to trust 0 if 0 is in a position of power relative to P . A number of studies also have found that general characteristics of a trustee, such as their membership in various social categories, influence trust (Kramer, 1999) . Dispositional factors refer to characteristics of P, such as motivational orientation, personality factors, and attitudes . According to Kramer (1999), evidence from both laboratory experiments and field research indicates there are significant individual differences in the predisposition to trust others, and the predisposition toward trusting behavior is correlated with other dispositional

14'

ctitioners' Willingness to Trust Clients

2 F .

0 V

6 sv ò (C

2g

F .

> a

G 0 0 Q

7 C

u

0

N a

o V

ro

146

WILLIAM SHAFER

orientations such as beliefs about human nature . Based on the results of several studies, Rotter (1971, 1980) also concluded that individual differences in the propensity to trust can be reliably measured and used to predict differences in trusting behavior. Dispositional traits also include factors such as a person's attitude toward risk . According to Kee and Knox (1970), the likelihood of trusting behavior will be influenced by perceptions of the relative rewards and risks associated with trust and betrayal . Thus, a person who is more tolerant of the risks associated with betrayal should be more likely to engage in trusting behavior . Kee and Knox (1970) acknowledged the importance of previous experience in the development of trusting relationships by positing that experience affects P's perceptions of O's trustworthiness directly as well as indirectly, through its influence on situational and dispositional factors . Kramer (1999) also concluded that perceptions of others' trustworthiness and the willingness to engage in trusting behavior are largely dependent on past experience . Here a distinction should be made between personalized or knowledge-based trust and generalized trusting behavior . A great deal of research has demonstrated that interactional histories between two parties serve a critical role in establishing personalized trusting relationships (e .g . Deutsch, 1958 ; Boyle & Bonacich, 1970 ; Pilisuk et al., 1971 ; Lindskold, 1978) . In the context of the Kee and Knox model, the interactional history between the two parties should directly affect P's perceptions of O's motives and/or competence, thus serving as a basis for personalized trust between P and O . Previous experience also may influence generalized trusting behavior through its influence on dispositional and situational factors . For example, Rotter (1971, 1980) proposed that a person's early trust-related experiences will serve as a basis for the development of general beliefs about the trustworthiness of others . Such beliefs provide the basis for a person's predisposition toward trusting behavior, which according to Rotter is a relatively stable personality trait. Previous experience also may influence a person's perception of various situational variables . For example, positive experiences with individuals from a certain social category may carry over to future relationships, providing the basis for "category-based trust" (Kramer, 1999) . Applicability to Tax Practitioners The Kee and Knox (1970) model has been applied to the study of trust and suspicion among professional auditors (Shaub, 1996), and also provides a useful framework for the study of client reliance decisions in tax practice . However, due to important differences between auditing and tax practice, the willingness to trust clients, as well as the variables that explain this trust, are likely to differ


147

between the two contexts . For example, auditors have a professional responsibility to be skeptical or suspicious of management assertions (AICPA, 1999, AU 230 .07), while tax preparers have both a right and a responsibility to serve as advocates for their clients (AICPA, 1999, TX 112 .04) . Thus, tax preparers should be more likely than auditors to accept client-provided data at face value .' Although tax preparers are not required to audit or otherwise verify data supplied by their clients, if the preparer suspects that client-provided data are incomplete or incorrect, then the preparer should make reasonable inquiries to substantiate the information (AICPA, 1999, TX 132.02) . Thus, if a client provides questionable information, tax professionals are faced with a choice of either trusting the client and relying on the data, or attempting to verify the propriety of the information . If the practitioner trusts the client but the client betrays that trust by providing fraudulent data, both parties may face potential penalties . The survey results reported by Yetmar et al . (1998) indicate that the issue of client trust and reliance is perceived as a serious ethical dilemma by professional tax practitioners, yet very little empirical research relating to this issue has been reported . A number of variables recognized in the Kee and Knox (1970) model may influence tax practitioners' willingness to rely on client data . For example, prior experience with the taxpayer potentially affects reliance through its influence on perceptions of client dependability . As observed by Roberts (1998) . tax practitioners should consider client dependability when evaluating the risks involved in tax reporting . Based on interview data, client dependability was identified by Milliron (1988) as one of several factors that affect tax preparers' reporting decisions, but Helleloid (1989) is the only study that has attempted to assess the effects of client dependability on tax reporting judgments . Helleloid (1989) examined the effects of the ambiguity of a client's automobile mileage records, operationalized as the frequency with which the auto mileage log was updated, on tax professionals' recommended mileage deductions . The results of an initial experiment indicated that the frequency with which the log was updated had no significant effect on recommended mileage deductions . In a follow-up experiment, the frequency of update had a statistically significant, but rather small effect on recommended deductions . Helleloid (1989) concluded that, overall, his subjects exhibited very limited sensitivity to the quality or dependability of client records . Although Helleloid's (1989) findings provide limited support for the effect of client dependability on reporting decisions, his study did not address the issue of CPAs' willingness to trust clients when they may have reason to suspect client dishonesty, or the effects of the perceived incentives of the client to cheat on the tax return . Based on the Kee and Knox (1970) model, the current study proposes

148

WILLIAM SHAFER

that prior experience with a client will influence subsequent client reliance decisions through its effect on perceptions of the client's trustworthiness . The effects of prior experience on CPAs' willingness to rely on client-provided data is an important issue that has not been addressed by previous research . Although the issue of client credibility or trustworthiness has received very little attention in the tax literature, several studies have investigated the effects of client aggressiveness, or client risk preferences, on CPAs' advice (e.g . Cloyd, 1995 ; Cuccia et al ., 1995 ; Schisler, 1994; Duncan et al ., 1989) . The results of these studies generally have indicated that CPAs attempt to accommodate the risk preferences of their clients ; i .e . they are more (less) likely to make aggressive reporting decisions for aggressive (conservative) clients (Roberts, 1998) . Analogous reasoning might suggest that CPAs will be more likely to rely on questionable data supplied by more aggressive or untrustworthy clients ; however, there are important differences between the two situations . According to the AICPA Statements on Responsibility in Tax Practice, it is permissible for a tax practitioner to advocate an aggressive tax return position for a client, provided the practitioner has a good faith belief that there is a realistic possibility the position will be sustained on its merits if challenged by the IRS (AICPA, 1999, TX 112 .05) . Thus, advocating an aggressive (but not fraudulent) position to accommodate the risk preferences of a client does not violate the CPA's ethical standards, provided the realistic possibility standard has been met . In contrast, if the CPA believes that a client has supplied fraudulent data, completing the return without making reasonable inquiries regarding the data would be a clear violation of professional standards . Accordingly, CPAs should be less willing to accommodate their clients' preferences if they feel they have supplied unreliable or fraudulent data .' This reasoning is consistent with recent studies in the auditing literature, which have shown that client preferences influence auditor judgment for accounting issues that are in the "gray area," but not for issues that represent clear violations of GAAP (e.g. Salterio & Koonce, 1997) . Thus, if past experience with a client raises credibility questions, then tax practitioners should be less willing to rely on the data without verification, as reflected in the following hypothesis . Hypothesis 1 . Tax practitioners will be more (less) willing to rely on questionable data provided by a client whose information has been reliable (unreliable) in the past . An important situational variable recognized in the Kee and Knox (1970) model is the effect of incentives for dishonest behavior . Several studies in the tax literature have suggested that a client's year-end tax payment status affects


149

incentives to adopt aggressive positions (e .g . Duncan et al ., 1989 ; LaRue & Reckers, 1989 ; Schisler, 1994, 1995) . Specifically, an underwithheld client should have a greater incentive to engage in aggressive or fraudulent reporting . This incentive effect may be explained by the potentially disruptive consequences of having to make an unexpected payment, or more formally based on Prospect Theory, which suggests that the potential-loss framing effect for the unexpected payment would result in risk-seeking behavior (Roberts,l998) . It has been suggested that tax practitioners also will make more aggressive decisions for clients who are in an underwithheld position, in an effort to satisfy the client (LaRue & Reckers, 1989) . Based on a review of the empirical findings on this issue, Roberts (1998) concluded that year-end payment status alone does not appear to affect tax preparers' decisions, but payment status does interact with other variables such as client risk preferences to affect the likelihood of aggressive reporting . For example, Schisler (1994, 1995) found that payment status increased the likelihood of aggressive recommendations when the client was described as aggressive, but not when the client was described as conservative . These findings suggest that tax practitioners recognize the incentive effects arising from the client's year-end payment status, and attempt to make reporting decisions that are consistent with the client's preferences (Roberts, 1998) . In situations involving decisions of whether to rely on questionable data, tax practitioners also should recognize the incentive effects of the client's year-end payment status . The current study proposes that recognition of the incentive created by underwithholding will reduce the likelihood that the CPA will rely on questionable client data without verification . This proposal may appear inconsistent with previous findings that CPAs will attempt to satisfy their clients by making aggressive reporting decisions . However, as previously explained, there are important qualitative differences between the two situations . If the CPA suspects that the client has provided incorrect or fraudulent data, accepting such data without documentation will be difficult to rationalize on the basis of "client preferences ." Thus, the following hypothesis is proposed : Hypothesis 2 . Tax practitioners will be less (more) willing to rely on questionable data provided by clients who are in an underwithheld (overwithheld) position . Dispositional factors, or personal characteristics of tax preparers, also should influence their willingness to rely on client data . Kee and Knox (1970) acknowledged the potential effects of variables such as motivational orientation, personality factors, and attitudes on trust . The CPA's propensity to trust others

1 50

WILLIAM SHAFER

is one personality trait that potentially influences their willingness to rely on their clients . Although individual differences in trusting behavior appear to be widely acknowledged, very little research has investigated whether such differences influence behavior in a professional context such as tax preparation . Rotter (1971) suggested that dispositional trust is a relatively stable personality trait, which indicates that its influence may carry over to professional decision making contexts. Shaub (1996) tested the relationship between dispositional trust, measured using the Wrightsman (1974) trustworthiness scale, and auditors' judgments regarding the reliability and accuracy of client data . Subjects' propensity for trusting behavior did not significantly influence their professional judgments in any of the eight cases examined . Shaub (1996) suggested this lack of significant results may have been due to the fact that all his subjects were seniors or managers who worked for the same public accounting firm ; thus, firm training and socialization may have produced a relatively homogeneous set of subjects who were more influenced by firm policy than by personal dispositions . Thus, in a less homogeneous subject group, dispositional trust should be more likely to influence judgments, as reflected in the following hypothesis . Hypothesis 3 . Tax practitioners' who have a higher (lower) propensity

to trust others will be more (less) willing to rely on questionable client data . Another dispositional factor that potentially affects client reliance decisions is the tax preparer's attitude toward risk in tax matters . CPAs who are risk averse should be less willing to risk the potential preparer penalties and other sanctions that could result from improper reliance on questionable client data. Although relatively few studies have investigated the effects of CPAs' risk attitudes on tax reporting decisions, they have generally found support for their influence (Roberts, 1998) . For example, Pei et al. (1990) found that tax practitioners' self-assessments of their aggressiveness affected reporting judgments . Carnes et al . (1996) reported that preparers' risk attitude was the only variable that consistently influenced the aggressiveness of their recommendations across both low and high ambiguity tax cases . Finally, Schisler (1994) found that practitioners' risk preferences had a significant impact on their estimates of the probability that they would recommend a questionable deduction . Based on these research findings, the following hypothesis is proposed . Hypothesis 4 . Tax practitioners' who are more (less) risk averse in tax matters will be less (more) willing to rely on questionable client data .


151

RESEARCH METHOD The data for this study were obtained by mailing instruments to a random sample of AICPA members whose membership information indicated that they were in public accounting practice, and that they had an interest in taxation . This section describes the research instrument and the participants .

Instrument

The research instrument included : (1) a cover letter, (2) two tax cases, (3) the MacDonald et al . (1972) Self-Report Trust Scale, and (4) a supplemental data sheet .' To enhance external validity, two separate tax cases were used . Client reliability and tax payment status were manipulated on a between-subjects basis for both cases, which are illustrated in Appendix 1 . In the Low Reliability conditions, the case indicated that in the past the client had been unable to provide documentation relating to certain deductions (Case 1), or that the CPA had discovered several errors in the client's records in the past (Case 2) . In the High Reliability conditions, the case indicated that all information supplied by the client in the past had proven to be reliable . 4 Tax payment status was manipulated by indicating that the client either had a significant amount of tax due, or should receive a significant tax refund . The cases were reviewed by three CPA firm partners, who indicated that they seemed realistic, and that they had encountered similar situations in the past . Taxpayer aggressiveness and payment status were crossed to create four versions of each case, and each potential subject was randomly assigned to one of the four versions of each case . To assess their willingness to rely on client-provided data, participants were asked to estimate the likelihood that they would prepare the client's tax return without attempting to verify the propriety of reported expenses . Responses were provided on an eleven-point Likert scale anchored on "0%" and "100%" . Several additional questions served as manipulation checks and provided data for supplemental analyses . To assess the effectiveness of the client reliability manipulation, subjects were asked to rate the reliability of the information supplied by the client on an eleven-point scale anchored on "very unreliable" and "very reliable" . To test the manipulation of tax payment status, participants estimated the strength of the client's incentives to overstate deductions on an eleven-point scale anchored on "very weak incentive" and "very strong incentive" . Finally, to assess the realism of the cases, subjects were asked to estimate the likelihood of encountering similar situations in their tax practice . Responses

152

WILLIAM SHAFER

to this question were provided on an eleven-point scale anchored on "0%" and "100%d' . The McDonald et al . (1972) Self-Report Trust scale was used to measure respondents' predisposition toward trusting behavior. This scale was developed based on Rotter's (1967) Interpersonal Trust Scale . Reliabilities for this scale reported in previous studies have ranged from 0 .70 (Lagace & Gassenheimer, 1989) to 0 .84 (McDonald et al ., 1972) . The scale, which is illustrated in Appendix 2, consists of ten items, each of which is scored on a scale of one to four, where four indicates more trusting behavior . A total score for each subject is obtained by summing responses to the ten items . To obtain a measure of risk attitudes, the supplemental data section included a question that asked participants to assess their own attitude toward risk. Following Carnes et al . (1996) and Schisler (1994), risk preferences were obtained using a single measure of risk propensity in tax matters .' Responses were provided on an eleven-point Likert scale anchored on "very conservative" and "very aggressive ." Participants Instruments were mailed to a random sample of 1,000 AICPA members . After a follow up mailing, a total of 256 usable responses were obtained, providing a response rate of approximately 25% . A comparison of the responses to the tax case and demographic data for the early and late respondents indicated no significant differences . A demographic profile of participants is provided in Table 1 . As indicated in the table, 81% of the respondents were male, and approximately 98% were either partners or managers in public accounting firms . Since most respondents were partners or managers, it is not surprising that the sample was predominantly male . Data reported by Doucet and Hooks (1999) indicated that in 1997, males accounted for 84% of all partners in public accounting firms . Thus, the gender composition of the sample appears to be approximately representative of this population . The fact that virtually all respondents were either partners or managers suggests that higher level employees may be more likely to respond to surveys ; however, this was not considered a cause for concern because these are the employees who are most likely to make the ultimate decision regarding reliance on client-provided information . Approximately 95% of the participants were employed by local or regional accounting firms . All respondents indicated a taxation interest in their AICPA membership information, and on average they spent approximately 60% of their time in the tax area . Since both cases dealt with small tax clients, it appears

Tax Practitioners' Willingness to Trust Clients Table 1 .

153

Demographic Profile of Participants . Number

Percent

Sample Size

256

Gender: Male Female

208 48

81 .3 18 .7

Position : Partner Manager Supervisor and below

196 56 4

76 .6 21 .9 1 .5

Firm size : NationalJlnternational Regional/Local

12 244

4.7 95 .3

Highest Degree : Bachelors Masters Other (e.g. JD, Ph .D .)

160 88 8

62 .5 34 .4 3 .1

Age: Mean Standard deviation

46 .3 10 .3

Public accounting experience (years) : Mean Standard deviation

19 .6 10 .8

Percentage of time spent doing tax work : Mean Standard deviation

58% 25

Responses to Self-Report Trust Scale : Mean' Standard deviation

26 .6 2 .9

Attitude toward risk in tax matters : Mean Standard deviation

5 .1 2 .0

' Scores could range from 10 (least trusting) to 40 (most trusting) . 2

Responses were provided on an eleven-point scale where 0 = "Very conservative" and 10="Very aggressive".

154

WILLIAM SHAFER

that they were appropriate for this participant group . However, it should be recognized that the findings of this study may not be generalizable to employees of national or international accounting firms . The majority of respondents had not earned a graduate degree . The average participant was 46 years old and had almost 20 years of public accounting experience . Table 1 also reports information on participants' dispositional trust and tax risk attitude. The reliability of the Self-Report Trust Scale, based on coefficient alpha, was 0 .72, which appears to be acceptable . The mean value for selfreported trust was approximately at the midpoint of the scale, which is comparable to the results reported by Lagace and Gassenheimer (1989) based on a sample of 242 adults . The standard deviation of trust scores was small in relation to the mean, which indicates that participants were relatively homogeneous in terms of dispositional trust . Responses to the tax risk attitude measure indicate that participants in the current study were slightly more conservative than those included in prior studies in the tax literature, and that significant individual differences in context-specific risk attitudes existed . 6

RESULTS Responses to Tax Cases and Preliminary Analysis A summary of responses to the tax cases is provided in Table 2 . As the data indicate, participants' mean estimates of the likelihood they would include the questionable deductions in the client's tax return without examining supporting documentation or otherwise verifying them ranged from approximately 42 (32) to 73 (55) percent for Case 1 (2) . For both cases, the lowest estimates were obtained for the Low Reliability, Tax Due condition, and the highest estimates were for the High Reliability, Refund condition, which is consistent with the research hypotheses . The responses also indicate significant variation in participants' willingness to rely on questionable client data, based on the relatively large standard deviations of the estimates . Responses to the manipulation checks are summarized in Table 3 . The estimated reliability of the client-provided data was used to test the effectiveness of the reliability manipulation . One-way ANOVA models with estimated reliability as the dependent variable and manipulated reliability as the independent variable indicated that the effects of the manipulation were significant at the 0 .0001 level for both cases, although the effect appears to be more pronounced for Case 1 . Participants' estimates of the strength of the hypothetical client's incentives to overstate their reported expenses served as a check for the payment

155

Tax Practitioners' Willingness to Trust Clients Table 2 .

Responses to Tax Cases : Estimated Likelihood of Client Trust. Reliability Low

High

Pooled

4.21 (3 .07) n=68

6.08 (3 .41) n=56

5 .05 (3 .32) n=124

Refund

5 .75 (2 .79) n=64

7 .29 (2 .45) n=68

6 .54 (2 .70) n=132

Pooled

4.96 (3 .04) n=132

6.74 (2 .90) n=124

5 .82 (3 .13) n=256

3 .23 (2.14) n=66

4.20 (3 .05) n=70

3 .73 (2.88) n=136

Refund

4 .20 (2 .62) n=56

5 .53 (2 .83) n=64

4 .91 (279) n=120

Pooled

3.67 (2 .55) n=122

4.84 (2 .97) n=134

4.28 (2.93) u=256

Case l :

Payment Status : Tax Due'

Case 2 : Payment Status : Tax Due

I 2

Reported numbers are mean responses. Numbers in parentheses represent standard deviations . All responses were provided on an eleven-point scale where 0="0%" and 10="100%" .

status manipulation . One-way ANOVA models with estimated incentives as the dependent variable and payment status as the independent variable indicated that the effects of this manipulation also were significant at the 0 .0001 level for both cases . To assess the realism of the cases, respondents were asked to estimate the probability that they could encounter similar situations in their tax practice . The mean estimates reported in Table 3 ranged from 64 (61) to 75 (72) percent for Case 1 (2), which indicates that the hypothetical scenarios were perceived as

156

WILLIAM SHAFER Table 3 .

Manipulation Checks and Supplemental Data . Reliability Low

High

Refund

Tax Due

Refund

Tax Due

Reliability''

4.25 (1 .90)

4 .00 (2.25)

8 .55 (1 .03)

6 .67 (2 .41)

Incentives'

3.89 (1 .96)

8 .21 (1 .43)

1 .82 (1 .18)

6 .29 (1 .63)

Likelihood of similar simations 4

7 .21 (1 .36)

7 .46 (1 .13)

6 .42 (1 .47)

7 .05 (1 .25)

Reliability' • 3

4.33 (2.77)

3 .03 (2 .23)

5 .60 (1 .68)

4 .39 (1 .69)

Incentives 3

4.33 (2 .50)

7 .10 (1 .81)

4 .07 (1 .73)

7 .20 (1 .57)

Likelihood of similar situations 4

6.48 (1 .11)

7 .19 (1 .36)

6 .14 (1 .35)

6 .78 (1 .04)

Case 1 :

Case 2 :

' Reported numbers are mean responses . Numbers in parentheses represent standard deviations. 2 Responses were provided on an eleven-point scale where 0 ="very unreliable" and 10 ="very reliable" . 3 Responses were provided on an eleven-point scale where 0="very weak" and 10="very strong" . 4 Responses were provided on an eleven-point scale where 0 = "0%" and 10 = "100%" .

relatively realistic . Consistent with the findings of Yetmar et al . (1998), these results also suggest that the problem of reliance on questionable client data is a significant ethical issue in tax practice . Univariate ANOVA and regression models were used to test for possible effects of various demographic factors on participants' willingness to rely on client data . Univariate ANOVA models indicated that neither gender, position (partner vs . manager), education level (bachelors vs . masters), nor firm type (nationallinternational vs . regional/local) had a significant influence on reliance decisions . Linear regression models also revealed that age, years of experience,

1 57


and the percentage of time spent doing tax work generally did not affect reliance .7 Based on the lack of significant influence of demographic factors on client reliance decisions, these variables were excluded from subsequent analyses . Hypothesis Tests

The hypotheses were tested using analysis of covariance (ANCOVA) models . The results of the models for both cases are included in Table 4 . Both models include client reliability and year-end tax payment status as between subjects variables, and self-reported trust and risk attitude as covariates s The model for Case I indicates that the effects of both client reliability and payment status were highly significant, while their interaction was not .' These results

Table 4.

ANCOVA Results . Sum of Squares

Case 1 : Between-subjects effects : Reliability Payment status Reliability*Payment status

F-value

Significance Level'

178 .8 108 .4 0 .7

22.3 13 .1 0 .1

0.000 0.000 0.770

0 .1 87 .9

0 .0 10.9

0.947 0.001

Case 2Between-subjects effects : Reliability Payment status Reliability*Payment status

40.1 39 .0 10.2

4 .8 4 .6 1 .2

0.030 0 .032 0 .270

Covariates : Self-reported trust Risk attitude

21 .3 11 .3

2 .5 1 .3

0 .112 .246 0

Covariates : Self-reported trust Risk attitude Model R2

Model R 2

0 .19

0 .06

' Reported significance levels are based on one-tailed tests .

158

WILLIAM SHAFER

provide support for Hypotheses 1 and 2 . Participants' tax risk attitude also had a significant effect on their reliance judgments . As indicated in Hypothesis 4, subjects who rated themselves as more aggressive were more likely to accept questionable data without verification . Contrary to Hypothesis 3, the effects of self-reported trust on reliance decisions did not approach significance . The results for Case 2 also indicate that the effects of reliability and payment status were significant, although to a lesser degree than for Case 1 . These findings also support Hypotheses 1 and 2 . As in Case 1, the interaction of reliability and payment status did not approach significance . For Case 2, neither of the covariates were significant . Thus, the results for Case 2 failed to support Hypotheses 3 and 4 .

DISCUSSION, LIMITATIONS, AND SUGGESTIONS FOR FUTURE RESEARCH The results of this research support the hypothesized effects of client reliability on practitioners' willingness to accept questionable client-provided data . The effects of reliability on CPAs' reliance decisions were significant for both of the cases examined . Thus, CPAs' willingness to trust clients appears to be dependent on prior history in dealing with the client . This result is consistent with prior research on trust and suspicion, which has demonstrated that trust is largely a history-dependent process (Kramer, 1999) . The findings also support the hypothesized effects of the client's year-end tax payment status on CPAs' acceptance of questionable data. Apparently, tax preparers recognize the incentive effects of the client's payment status, and act more cautiously when the client has an incentive to "cheat" on the return . The study failed to support the hypothesized effects of the predisposition toward trusting behavior on client reliance . The lack of significant results may be due to the fact that participants' attitudes toward trust were relatively homogeneous, as reflected in the small standard deviation of responses to the Self-Report Trust Scale . This explanation is consistent with Shaub's (1996) speculation that the lack of significance of dispositional trust in his study was a result of the high degree of homogeneity of his sample . Although the sample in the current study was less homogeneous than that of Shaub (1996) in the sense that participants did not work for the same firm, individual differences in dispositional trust appeared to be relatively small . The lack of significance for dispositional trust also may be attributable to the fact that context-specific information, such as the prior history of dealings with a particular client, overrides the general predisposition toward trusting behavior . Rotter (1971)


1 59

suggested that general tendencies toward trust will be most effective in predicting behavior in unusual or novel situations . This may explain why this personality trait was not an effective predictor of trusting behavior in familiar professional contexts such as those examined in this study and in the Shaub (1996) study . The findings of this study provide mixed support for the hypothesized effects of tax preparers' risk attitudes on their willingness to accept questionable data . The effects of risk attitude were significant for Case 1, but not for Case 2 . The lack of significance for Case 2 appears inconsistent with recent research findings . In a study that employed 18 different tax scenarios, Carnes et al . (1996) found that contextual differences had a significant effect on tax preparers' willingness to support clients' return positions ; however, practitioners' risk propensity was a significant explanatory variable across all the scenarios tested . The primary difference between the two cases is that Case I dealt with the taxpayers' personal tax return, while Case 2 dealt with the corporate tax return for a small business . If most subjects perceived a low level of risk associated with Case 2, this could explain why risk attitude was not a salient determinant of client reliance . However, the data indicate that, compared to Case 1, subjects estimated a lower likelihood of client reliance, rated the reliability of the client's data lower, and generally felt that the client had a greater incentive to cheat in Case 2 . These results suggest that there was a significant amount of risk associated with Case 2 and, accordingly, that risk attitude should have affected judgments . One possibility for the differential results may be that participants perceived the threat of IRS audit or penalties to be lower in Case 2 than in Case 1 . Due to the lack of a clear explanation for the inconsistent results, future research should attempt to clarify the effects of risk attitude on tax practitioners' client reliance decisions . The current study was an attempt to gain a better understanding of CPAs' willingness to rely on questionable client data, and the factors that influence such reliance decisions . The findings of the study are subject to a number of limitations and should be interpreted with caution . For example, most respondents were employed by local CPA firms . If systematic differences exist between the judgments of tax practitioners employed in large and small firms due to differences in such things as selection, training or socialization, then the findings of the current study may not be generalizable to employees of national or international firms . Also, although the response rate was comparable to that typically obtained in this type of study (e .g. Bandy et al ., 1994 ; Schisler, 1994), there is a chance that the results were affected by nonresponse bias . The generalizability of the findings to practice settings also is an open question . An attempt was made to capture some of the key contextual influences on CPAs' reliance decisions in

160

WILLIAM SHAFER

the hypothetical cases ; however, client reliance decisions in practice may be affected by factors not addressed in the current study, such as economic incentives . The findings of this study have implications for tax compliance and professional ethical standards . From a public policy perspective, an important issue is whether professional tax preparers are in effect aiding noncompliance by failing to question client data that appear suspect. Previous research has demonstrated that, given an incentive to do so, tax professionals are willing to exploit the ambiguity provided by either vague professional standards or weak evidential support in order to justify a client's aggressive reporting position (Cuccia et al ., 1995) . The results of the current study suggest that, at least in some cases, practitioners may also choose to ignore suspicions of client improprieties rather than demand supporting documentation. The AICPA Statements on Responsibility in Tax Practice (SRTPs) (AICPA, 1999) stipulate that if a paid preparer has reason to suspect that client data is incorrect or incomplete, then further inquiries should be made to substantiate the validity of the information . In the current study, even when participants felt that the hypothetical taxpayer was relativley unreliable and had a strong incentive to cheat, they estimated a 30 to 40% likelihood that they would accept the client's deduction amounts without asking for any supporting documentation. The results also suggest a general lack of consensus among CPA tax practitioners regarding when client reliance is justified . Thus, future research should attempt to establish a baseline for what constitutes reasonable client reliance, and address the issue of the appropriateness of reliance by CPAs and other professional tax preparers . 10 One possible approach for addressing the issue of whether CPAs' willingness to rely on client data is reasonable would be to compare CPA tax practitioners' reliance judgments with those of IRS agents or other professional groups such as attorneys and non-licensed tax preparers . If future studies conclude that CPA tax practitioners appear too eager to rely on questionable data, then perhaps professional standards should be modified to clarify the tax preparer's professional responsibility to verify such data. Future research also should attempt to obtain a better understanding of the factors that influence tax practitioners' client reliance judgments . The current study addressed the effects of only one situational and two dispositional variables on client trust, and the relatively low RI values for the ANCOVA models suggests that other variables may play an important role in explaining variation in professional judgments . Roberts (1998) recognized the potential influence of numerous situational or environmental variables on tax practitioner judgment . For example, factors such


161

as client importance, client tenure, and client sophistication may influence reliance decisions . Kee and Knox (1970) also recognized the potential influence of variables such as the relative power of the parties involved on trust and suspicion . In a tax preparation context, the relative power of the parties will be influenced by economic considerations such as client importance . Previous studies have shown that economic incentives of this nature may influence CPA tax practitioners' ethical judgments (e .g . Burns & Kiecker, 1995) ; thus, such incentives may play a role in client reliance decisions . Other potential situational influences include factors such as the perceived IRS audit probability and perceived likelihood of monetary penalties or other sanctions (Roberts, 1998) . Other dispositional traits also could be examined . The current study generally found no evidence that demographic variables such as age, years of experience, gender, or education level affect client trust . However, other personality characteristics such as ethical attitudes or ethical development may influence client reliance . For example, studies in the auditing literature have found that individual differences in cognitive moral development can have a significant influence on professional judgment (Louwers et al ., 1997) . The influence of such factors on client reliance in a tax setting also could be investigated .

NOTES I . Of course, one could question whether auditors do actually exercise the level of skepticism toward client assertions that is required by professional standards . For example, recent research suggests that auditors' reporting judgments may be biased by their economic incentives (Hackenbrack & Nelson, 1996) . An inherent bias toward supporting the client's position is likely to be present in both auditing and tax practice . Nevertheless, the fundamental differences in professional responsibilities between the two contexts should produce differences in trusting behavior . 2 . This discussion is not intended to imply that economic considerations, such as the fear of losing the client, will not influence tax practitioners' judgments . Previous research by Bums and Kiecker (1995) indicated that tax practitioners are more likely to encourage their clients to overstate their deductions when their firm benefits from such actions . However, the current study proposes that, ceteris paribus, knowledge that the client is not trustworthy should cause the CPA to be more cautious in relying on data that appear suspect . 3 . The two tax cases were presented first, followed by the Self-Report Trust Scale (STS) and the supplemental data sheet . To test for possible effects of completion of the experimental cases on the general propensity toward trust, responses to the STS were compared for participants who received : (I) the low reliability version of both cases, (2) the high reliability version of both cases, and (3) a mixture of low and high reliability cases . There were no significant differences in the mean responses to the STS among these three groups ; thus, it appears that completion of the experimental cases did not have a significant impact on the measure of general propensity toward trusting behavior.

162

WILLIAM SHAFER

4 . As illustrated in the Appendix, in the low (high) reliability version of Case 1, the taxpayer was described as aggressive (conservative) . These terms were intended to convey the general idea of a taxpayer who is more (less) willing to take risks . As observed by Helleloid (1989), more aggressive taxpayers are more likely to be viewed as possessing low credibility or trustworthiness . 5 . Schisler (1994) also measured risk preferences using the Pettigrew (1958) CategoryWidth instrument . However, the measure of risk preferences provided by this instrument had no significant impact on tax practitioners judgments, while self-reported risk in tax matters did have a significant influence . 6 . On average, Schisler's (1994) subjects rated themselves at 63 on a 100-point scale of tax aggressiveness, where 50 represented the "average tax preparer ." Similarly, the mean response for tax risk preferences in the Carnes et al . (1996) study was 4 .77 on a 7-point scale where 1 (7) represented "very risk averse" ("very risk prone") . The mean risk attitude in the current study was approximately at the midpoint of the scale . The standard deviation of tax risk attitude scores was approximately 40% of the mean response in the current study, compared with 32% in the Schisler (1994) study . Standard deviations were not reported by Carnes et al . (1996) . 7 . Years of experience had a marginally significant effect (0 .05 level) on reliance decisions for Case 1, but not for Case 2 . For Case 1, more experienced subjects estimated slightly higher probabilities of relying on the client-provided data . 8 . The two covariates were not significantly correlated (Pearson r = 0 .03, two-tailed significance level of 0 .627) . Two alternative models that included only one of the two covariates also were run for both cases . All substantive results and conclusions based on these models were the same as those reported in Table 4 (i .e. the self-report trust measure was not significant for either case, and tax risk attitude was significant for Case I but not for Case 2) . 9 . Interaction terms were included in the models to control for possible interaction effects, although interactions were not formally hypothesized. None of the reported results are significantly different when the terms are excluded from the models . 10 . Client trust or the appropriateness of reliance on questionable data should be viewed distinctly from the issue of recommending aggressive tax return positions . CPAs may recommend "aggressive" positions if they have a good faith belief that such positions have a "realistic possibility" of being sustained on their merits if challenged by the IRS . However, intentional overstatements of expense or understatements of revenue reported by a client obviously have no "merits," and would never be sustained if detected by the IRS .

REFERENCES American Institute of Certified Public Accountants (AICPA) (1999) . AICPA Professional Standards (Vol . 2) . New York : AICPA . Bandy, D ., Betancourt, L ., & Kelliher, C. (1994) . An Empirical Study of the Objectivity of CPAs' Tax Work. Advances in Taxation, 6, 1-23. Boyle, R ., & Bonacich, P . (1970). The Development of Trust and Mistrust in Mixed-Motives Games . Sociometry, 33, 123-139 . Bums, J . 0 ., & Kiecker, P. (1995) . Tax Practitioner Ethics : An Empirical Investigation of Organizational Consequences . The Journal of the American Taxation Association, 28(2), 20-49 .


163

Carnes, G . A ., Harwood, G. B ., & Sawyers, R. B . (1996) . The Determinants of Tax Professionals' Aggressiveness in Ambiguous Situations . Advances in Taxation, 8, 1-26 . Cloyd, C . B . (1995) . The Effects of Financial Accounting Conformity on Recommendations of Tax Preparers . The Journal of the American Taxation Association, 17(2), 50-70 . Cuccia, A . D ., Hackenbrack, K ., & Nelson, M . W . (1995) . The Ability of Professional Standards to Mitigate Aggressive Reporting . The Accounting Review, 70(2), 227-248 . Deutsch, M . (1958) . Trust and Suspicion . Journal of Conflict Resolution, 2, 265-279 . Doucet, M . S ., & Hooks, K . L . (1999) . Toward an Equal Future. Journal of Accountancy (June), 71-76 . Duncan, W . A ., LaRue, D . W ., & Reckers, P . M . J . (1989) . An Empirical Examination of the Influence of Selected Economic and Non-economic Variables in Decision Making by Tax Professionals . Advances in Taxation, 2, 91-106 . Fein, 5 . (1996) . Effects of Suspicion on Atributional Thinking and the Correspondence Bias . Journal of Personality and Social Psychology, 70(6), 1164-1184 . Hackenbrack, K ., & Nelson, M . W . (1996) . Auditors' Incentives and Their Application of Financial Accounting Standards . The Accounting Review, 71(I), 43-59. Hardin, R . (1993) . The Street-Level Epistemology of Trust . Politics and Society, 21(4), 505-529 . Helleluid, R . T, (1989) . Ambiguity and the Evaluation of Client Documentation by Tax Professionals . The Journal of the American Taxation Association, 11, 22-36 . Hilton, J . L., Fein, S ., & Miller, D . T . (1993) . Suspicion and Dispositional Inference . Personality and Social Psychology Bulletin, 19, 501-512 . Kee, H . W ., & Knox, R. E. (1970) . Conceptual and Methodological Considerations in the Study of Trust and Suspicion. Journal of Conflict Resolution, 14, 357-366 . Kramer. R . M . (1999) . Trust and Distrust in Organizations : Emerging Perspectives, Enduring Questions . Annual Review of Psychology, 50, 569-598 . Lagace, R . R ., & Gassenheimer, J . B . (1989) . A Measure of Global Trust and Suspicion : Replication . Psychological Reports . 65, 473-474 . LaRue . D ., & Reckers . P . M . J . (1989). An Empirical Examination of the Influence of Selected Factors on Professional Tax Preparers' Decision Process . Advances in Accounting, 7, 37-50 . Lindskold, S . (1978) . Trust Development, the GRIT Proposal, and the Effects of Conciliatory Acts on Conflict and Cooperation . Psychological Bulletin, 85, 772-793 . Louwers, T. J ., Ponemon . L . A., & Radtke, R . R . (1997) . Examining accountants' ethical behavior : A review and implications for future research . In : S . Sutton & V. Arnold (Eds), Behavioral Accounting Research : Foundations and frontiers (pp . 188-221) . Sarasota FL: American Accounting Association, McDonald, A . P., Kessel, V . S ., & Fuller, J . B . (1972) . Self-Disclosure and Two Kinds of Trust . Psychological Reports, 30, 143-148 . Milliron, V . C . (1988) . A conceptual model of factors influencing tax preparers' aggressiveness . In : 5 . Moriarity & .1 . H . Collins (Eds), Contemporary Tax Research (pp . 1-23) . Norman OK : The Center for Economic and Management Research, University of Oklahoma . Pei, B . K. W ., Reckers, P. M . J ., & Wyndelts, R . W . (1990). The Influence of Information Presentation Order on Professional Tax Judgment . Journal of Economic Psychology, l1(I), 119-146 . Pettigrew, T . F. (1958) . The Measurement and Correlates of Category Width as a Cognitive Variable . Journal of Personality, 26, 532-544 . Pilisuk . M ., Kiritz, S ., & Clampitt, S . (1971) . Undoing Deadlocks of Distrust : Hip Berkeley Students and the ROTC . Journal of Conflict Resolution . 15, 81-95 .

164

WILLIAM SHAFER

Roberts, M . L . (1998) . Tax Accountants' Judgment/Decision-Making Research : A Review and Synthesis. The Journal of the American Taxation Association, 20(l), 78-121 . Rotter, J . B . (1967). A New Scale for the Measurement of Interpersonal Trust. Journal of Personality, 35, 651-665 . Better, J. B . (1971) . Generalized Expectancies for Interpersonal Trust . American Psychologist, 26, 443-452 . Rotter, J . B . (1980). Interpersonal Trust. Trustworthiness, and Gullibility . American Psychologist, 35, 1-7 . Salterio, S ., & Koonce, L . (1997) . The Persuasiveness of Audit Evidence : The Case of Accounting Policy Decisions. Accounting, Organizations and Society, 22(6), 573-587 . Schisler, D . L. (1994) . An Experimental Examination of Factors Affecting Tax Preparers' Aggressiveness: A Prospect Theory Approach . The Journal of the American Taxation Association, 16(2), 124-142 . Schisler, D. L. (1995) . Equity, Aggressiveness, Consensus : A Comparison of Taxpayers and Tax Preparers . Accounting Horizons, 9(4), 76-87 . Shaub, M . K . (1996) . Trust and Suspicion: The Effects of Situational and Dispositional Factors on Auditors' Trust of Clients . Behavioral Research in Accounting, 8, 154-174 . Wrightsman, L . 5 . (1974) . Assumptions about human nature : A social-psychological approach . Monterey CA : Brooks/Cole. Wrightsman, L . S . (1991) . Interpersonal trust and attitudes toward human nature . In : J. Robinson, P . Shaver & L. Wrightsman (Eds), Measures of Personality and Social Psychological Attitudes (pp. 373-412) . London : Academic Press . Yetmar, S. A., Cooper, R. W., & Frank, G . L . (1998) . Ethical Issues Facing CPA Tax Practitioners . The CPA Journal (October), 29-33


165

APPENDIX 1 Experimental Cases Case No . 1 Low Reliability, Tax Due : Bob Jenkins and his wife Diane have been tax clients of your firm for several years . Bob is president of a large local bank, and Diane is a partner in a local law firm . Based on prior experience, you have found that the Jenkins are aggressive taxpayers . In fact, on a few occasions they were unable to provide documentation you requested relating to certain deductions they were claiming on their return, such as charitable contributions and medical expenses . Last week, you had a meeting with the Jenkins to discuss the current year's tax return . Based on preliminary analysis of the data they provided, you informed them that they had approximately $10,000 tax due . The Jenkins were noticeably upset that their underpayment was so large. This week, Bob Jenkins called you to say that he had overlooked a number of expenses relating to rental properties they own . He faxed you a list of the additional expenses, which included a variety of repair and maintenance items and other miscellaneous expenses . The additional expenses totaled approximately $12,000 . High Reliability, Tax Refund : Bob Jenkins and his wife Diane have been tax clients of your firm for several years . Bob is president of a large local bank, and Diane is a partner in a local law firm . Based on prior experience, you have found that the Jenkins are very conservative taxpayers, and that the information they have supplied you with has been very reliable . Last week, you had a meeting with the Jenkins to discuss the current year's tax return . Based on preliminary analysis of the data they provided, you informed them that they should receive a substantial refund . This week, Bob Jenkins called you to say that he had overlooked a number of expenses relating to rental properties they own . He faxed you a list of the additional expenses, which included a variety of repair and maintenance items and other miscellaneous expenses . The additional expenses totaled approximately $12,000 . Case No . 2 Low Reliability, Tax Due : For the past five years, you have prepared the personal tax returns for Richard and Mary Cooper, as well as the corporate return for their family-owned advertising

166

WILLIAM SHAFER

agency, Advertising Solutions, Inc . Due to its small size, Advertising Solutions has never been required to have a CPA associated with its financial statements . The company submits its internally prepared statements to the local bank and to your firm for tax return preparation . Not surprisingly, on several occasions in the past you have noticed apparent errors in the Advertising Solutions financial statements, which generally turned out to be either understatements of revenues or overstatements of expenses . Based on recent discussions with Richard Cooper, you expected Advertising Solutions' profitability to increase significantly in the current year, since the company performed a great deal of work for two large new clients during the fourth quarter . Consequently, you informed Mr . Cooper that the company would have a significant amount of tax due . However, when the Coopers submitted the current year financial statements to you, you noticed that net income was approximately the same as in the prior year . Although revenue did increase significantly, the increase was largely offset by increases in travel, advertising and promotion, and miscellaneous expenses . The Coopers were not able to provide you with a specific explanation for the increase in these expenses . High Reliability, Tax Refund : For the past five years, you have prepared the personal tax returns for Richard and Mary Cooper, as well as the corporate return for their family-owned advertising agency, Advertising Solutions, Inc . Due to its small size, Advertising Solutions has never been required to have a CPA associated with its financial statements . The company submits its internally prepared statements to the local bank and to your firm for tax return preparation . Despite the fact that the Advertising Solutions financial statements have not been audited or reviewed, they have appeared to be very reliable in the past . Based on recent discussions with Richard Cooper, you expected Advertising Solutions' profitability to decrease in the current year, due to the loss of two large clients during the fourth quarter . Consequently, you informed Mr . Cooper that the company should receive a substantial tax refund . When the Coopers submitted the current year financial statements to you, you noticed that reported net income was even lower than you anticipated . The decrease in income was due to the loss of the two clients, as well as the fact that travel, advertising and promotion, and miscellaneous expenses significantly increased . The Coopers were not able to provide you with a specific explanation for the increase in these expenses .

167

Tax P ct otters' Willingness to Trust Clients

APPENDIX 2 Self-Report Trust Scale Please respond to each of the following statements by circling the number that corresponds most closely with your personal attitudes : I, 1 expect other people to be honest and open . (reverse scored)

I Strongly Agree

2 Agree

Disagree

2 . I am less trusting than the average person. (reverse scored)

1 Strongly Disagree

2 Disagree

3 Agree

4 Strongly Agree

3 . 1 am more trusting than the average accountant .

I Strongly Disagree

2 Disagree

1 Agree

4 Strongly Agree

4 . 1 am suspicious of other people's intentions .

I Often

2 Sometimes

1 Seldom

4 Never

5 . 1 am less trusting than the average CPA in my area of concentration . (reverse scored)

I Strongly Disagree

2 Disagree

3 Agree

4 Strongly Agree

6. I have faith in human nature .

1 Strongly Disagree

2 Disagree

3 Agree

4 Strongly Agree

7 . 1 feel that other people can be relied upon to do what they say they will do .

1 Nobody

2 A few people

1 some people

4 Most people

8 . 1 feel that other people are out to get as much as they can for themselves .

I Most people

2 Some people

3 A few people

4 Nobody

9 . 1 have faith in the promises or statements of other people . (reverse scored)

I Very much

2 Much

3 Little

4 Very little

1 Strongly Agree

2 Agree

3 Disagree

4 Strongly Disagree

Ip. I am cynical (pessimistic) .

4 Disagree Strongly

MEASURING VARIATION IN TAX LIABILITY AMONG ECONOMIC EQUALS Peter J . Weston

ABSTRACT The coefficient of variation (CV) and coefficient of residual variation (CRV) have been used as measures of horizontal equity. Both, however, are noisy measures in that they overstate the amount of variation due to inequity in the tax system (Grasso & Frischmann 1992) . The purpose of the current study, therefore, is to use the CRV of tax liability and provide an estimate of the portion of this CRV measure that is due to the tax system alone and that portion that is due to specification error. This precision is an improvement over the Grasso and Frischmann model . The first purpose is achieved by computing the difference between two CRV measures . The first CRV measure is derived from a regression of an expanded total income amount (ETI) on tax liability; the second CRV measure is from a regression of several explanatory variables on tax liability . To the extent that the more fullyspecified regression captures the Internal Revenue Code provisions, the reduction in the CRV measure can be attributed to the tax system . A second purpose of this study is to estimate the extent to which the various types of tax code provisions cause this variation . The second purpose is achieved by an iterative process of omitting one explanatory variable from the full regression and determining the change in CRV due to this omitted variable.

Advances in Taxation, Volume 13, pages 169-203 . Copyright m 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved . ISBN : 0-7623 .0774-9 169

170

PETER J . WESTORT

INTRODUCTION Empirical studies of horizontal equity often use the coefficient of variation (CV) of tax liability (or effective tax rate) as the measure of equity . The CV is the standard deviation of a sample divided by its mean and expressed as a percent. Data generally are split into groups based on an expanded income figure . Taxpayers in the same expanded income group are considered economic equals for income tax purposes and thus should pay the same tax . The CV is computed for each group and is interpreted as the measure of inequity for that group . A study might recompute tax liability after a proposed tax change and then recompute the CV after the change . The difference in the CV measures (before and after the tax law change) is interpreted as the effect on horizontal equity . One criticism of this measure is that, since those taxpayers at the higher end of an income range have, by definition, more income than taxpayers at the lower end, then under an "equitable" system they should pay more tax . Thus, the measure of equity is distorted because it also includes variation due to different income levels within the group . In an attempt to address this issue, Grasso and Frischmann (1992) propose the use of the coefficient of residual variation (CRV) . The CRV is the estimate of the standard deviation of a regression's error term divided by the mean of the dependent variable, expressed as a percent . They sort their sample into deciles based on expanded income and, for each decile. regress expanded income on tax liability and compute the CRV for each decile . As they observe, they trade off one type of error for another . While they control for differences in income, they introduce specification error into the measure . Thus, as measures of horizontal equity, both CV and CRV overstate the amount of inequity observed (see Grasso & Frischmann, 1992, 131) . What is not known is the degree to which horizontal inequity is overstated in these measures . This is important because existing studies cannot observe whether changes in the tax law produce large or small changes in equity . They can only state whether or not there is a statistically significant difference . For example, Pierce (1989, p . 61) states that the largest positive weighted-average percentage change in the overall analysis was 5 .70% . Since the CV measure she uses overstates true horizontal inequity, her 5 .70% may be very conservative . There is a need for a more precise measure of horizontal equity . If the effect of a proposed tax law change is perceived to be small, it might lack support ; whereas a larger effect may garner more support and be implemented . From a broader perspective, therefore, a more precise measure is important . The first purpose of the current study is to use CRV of tax liability and provide an estimate of the portion of this CRV measure that is due to the tax

Measuring Variation in Tax Liability Among Economic Equals

171

system alone and the portion that is due to specification error . This is achieved by comparing the CRV measures of two regressions in each expanded income group . The first regression is essentially a replication of Grasso and Frischmann, simply a regression of expanded total income (ETI) on tax liability . The second regression includes the ETI variable plus explanatory variables that account for Internal Revenue Code (IRC) provisions that result in different tax liabilities . Specifically, the second regression includes explanatory variables for exempt income, favorably-taxed income, exemptions, itemized deductions, AGI adjustments, AMT tax preferences, carryover losses, and credits, as well as the expanded income amount . If this more complex regression model captures all the IRC provisions that affect tax liability, then the difference in the CRV measure for the two models is that portion due to the tax system . That is, subtracting the CRV measure for the full model from the CRV measure for the simple model results in the portion of the CRV due to the tax system . Figure I illustrates this concept . Implicitly then, if the full model has no omitted variables, the entire CRV for this model is due to lack of linearity in the data. While the equity measure of earlier studies overstated variation due to the tax system, this method would tend to understate it . Notice in Fig . 1 that the smaller the CRV due to the full model, Actual Tax Liability

Estimated tax Liability Estimated Tax Liability Using Full Model Using Simple Model (Grasso & Frischman)

CRV using simple model

CRV using full model

CRV caused by Specification Error

CRV Caused by the Tax System

Fig. I .

1 72

PETER J. WESTORT

the larger the CRV due to the tax system . To the extent the full model is imperfectly specified, some of the variation due to the tax system will be attributed to specification error . This is an improvement over the Grasso and Frischmann technique because it is more precise . Grasso and Frischmann use the entire CRV as a measure of inequity . The proposed model allocates part of that CRV to specification error and the balance is considered a measure of the inequity in the tax system or, at the least, a measure of variation in tax liability due to the tax system . A related issue to the size of the horizontal equity measure is the cause of the inequity . Currently, many researchers focus on one issue, for example, the alternative minimum tax, and assess the effect on horizontal equity by changes in that one item . Why the horizontal equity measure is large or small at any given income level is an unanswered but interesting question . For example, if the purpose of the alternative minimum tax is to reduce inequity, then it must address the causes of that inequity . It becomes important to know which items give rise to horizontal inequity . Thus, the second purpose of this paper is to decompose the variation attributable to the tax system into the types of IRC provisions causing it. This is achieved by an iterative process of fitting a series of regressions for each income group . Each regression omits one of the explanatory variables . Comparing the CRV from the regression for the omitted variable to the CRV due to the tax system results in a conservative estimate of the change in the CRV due to the omitted variable . The larger the change in CRV due to the omitted variable, the larger its contribution to variation in tax liability .' Results from this type of analysis may have policy implications, especially with respect to the causes of variation in the tax burden . Policy makers or lobbyists concerned about amending the IRC to reduce perceived inequity (at least as measured by variation in tax liability) now can observe what contributes the most to that variation . This could have an impact on what issues they choose to pursue and may force policy makers to make trade-offs . For example, exemptions may add a great deal of variation to tax burdens, yet be perceived as equitable and therefore allowed to persist. Alternatively, the size of the variation they introduce may be deemed too large and thus adjusted. The next section addresses some theoretical issues regarding the concept of horizontal equity and the effect of implicit taxes . The research method, description of the variables, and the data set are described in the third section . The results are then presented . Concluding remarks are in the final section .

HORIZONTAL EQUITY AND IMPLICIT TAXES This section provides background information on two issues generally affecting studies that use CV or CRV to measure equitable distribution of the tax burden .


173

The first issue deals with the use of the term "horizontal equity" and what it implies . The second issue addresses the concept of implicit taxes and its general absence from tax burden studies that use archival data . Horizontal Equity

Taxpayers with the same amount of income may pay different amounts of taxes . Labeling this difference as an inequity is a value judgment . The purpose of this study is to assess the extent to which the tax system causes variation in explicit tax liability among those deemed to be economic equals, and further to assess the extent to which various IRC provisions contribute to this dispersion . Studies that group taxpayers by some expanded income figure imply that taxpayers in the same grouping are equals, or nearly so (Anderson, 1985 ; Pierce, 1989 ; Allan & Iglarsh, 1996) . The further implication is that they should pay the same tax or have the same effective tax rate. This is a normative statement which places a priority on the expanded income figure as a measure of equity . It implies that the only equityrelevant characteristic is the pre-tax income amount used to determine the income groupings . However, items such as the dependency exemptions have long been part of the IRC . The IRC provides tax relief for taxpayers who must provide financial support for other individuals (dependents) . Taxpayers with the same amount of income, regardless of how it is defined, do not pay the same amount of tax if they have a different number of dependency exemptions . These two taxpayers are not treated as equals, yet many taxpayers would consider this provision equitable. Other provisions of the IRC provide other tax benefits or penalties for various reasons ; some are for equity purposes, others are for incentive, punitive, or regulatory purposes . To determine the degree of inequity in the tax system, one would need to control for each equity-relevant provision of the IRC . This requires analyzing the justification for each Code provision. Some studies have controlled for the number of exemptions claimed (Jenkins, 1988 ; Grasso & Frischmann, 1992), but no horizontal equity studies have controlled for other provisions . 2 Further, Grasso and Frischmann are non-committal about whether dependency exemptions are equity relevant. They simply state : . . . allowances for differences in the number of individuals in a taxpaying unit may be viewed as a mechanism promoting horizontal equity or as an example of preferential treatment contributing to horizontal inequity (1992, 127) .

Kaplow observes, Violations of HE [sic] are understood as being measured without regard to the origin orjustification of the initial distribution of income - including, for example, the possibility that it merely reflects the incidental or even capricious effects of previous reforms (1989, 139) .

174

PETER J . WESTORT

Noting this distinction then, the current study observes variation in tax liability and the sources of that variation but does not take a stand as to its impact on equity . Perhaps in the future these studies will be referred to as tax burden distribution studies rather than horizontal equity studies . Implicit Taxes

One criticism of horizontal equity studies using archival data is that they omit implicit taxes . The current study suffers from this deficiency. Implicit tax generally refers to the reduced return that taxpayers accept as the cost of investing in tax-favored assets . Because the amount of implicit tax "paid" is not measured in empirical studies, these studies provide an incomplete picture of the total tax burden. This shortcoming is another reason to avoid the label of inequity and simply assess the extent of the explicit tax burden . However, a few short observations are appropriate . First, while considering implicit taxes in a business decision setting (Scholes & Wolfson, 1992) is generally accepted, it is not at all clear that many individual income tax provisions can be priced in any market ; for example, deductions for blindness, marital status, or casualty loss are unlikely to have market prices . Second, even with municipal bonds where implicit taxes are frequently assumed to eliminate horizontal inequity, there still may be inequities . Economic equals are defined by expanded income, not by marginal tax rates . If municipal bonds are priced to appeal to those with the lower marginal tax rate, then those with higher marginal tax rates reap a benefit . Bittker (1980) refers to this as the "trickle-up phenomenon" . Third, it always has been considered socially desirable to distribute the tax burden according to ability to pay . The extent to which we do that is an empirical question, but a good starting point is how we distribute the explicit tax burden . In summary, while implicit taxes are a phenomenon with which we must deal, studies observing the explicit tax burden still provide important and useful information .

RESEARCH METHOD Grasso and Frischmann (1992) introduce the concept of coefficient of residual variation (CRV) into the horizontal equity literature . The CRV is the estimate of the standard deviation of the regression's error term expressed as a percentage of the mean of the dependent variable . Following CV analysis, they split their data into groups based on expanded income . Observing that tax liability and effective tax rate increase within groups as the expanded income amount increases, they suggest that the expanded income amount be regressed on tax


175

liability (or effective tax rate) for each group . They further suggest that the associated CRV can be used as an alternative expression of horizontal inequity . Grasso and Frischmann also run a set of regressions including the number of personal and dependency exemptions as a second explanatory variable, citing other studies suggesting that this may be equity relevant . Using the regression approach, the measure of tax liability variation (horizontal inequity) is not based on the distance from actual tax liability to the mean for the group, but rather on the distance from the actual tax liability to the estimated tax liability for the observation . Implicit in the grouping by an expanded income amount is that this is the appropriate basis for tax liability and, as income increases, so should tax liability . That is, the tax liability estimated by the regression is what tax liability "should be" if expanded income were the only criterion for determining tax liability . The distance, however, from actual tax liability to estimated tax liability can be attributed to two causes . Part of the variation is due to provisions of the IRC that allow other items to be considered in computing actual tax liability (exemptions, deductions, credits, etc .) and part of the variation is due to specification error. The CRV measure, then, overstates the variation in tax liability due to the tax system, as does the CV measure . The current study proposes to estimate the variation due to the tax system alone by fitting a second regression, computing its CRV and using the difference in CRVs as the measure of variation in tax liability . Specifically, the second regression includes proxies for various IRC provisions that affect tax liability and are grouped into the following categories : exempt income, favorably-taxed income, personal and dependency exemptions, itemized deductions, AGI adjustments, tax preferences, carryover losses, and tax credits . By including a variable to represent each of these categories, plus an expanded income variable, and regressing on tax liability, the resulting estimate should provide a better fit as measured by adjusted R-square . The anticipated better fit results because these additional variables are, in fact, used to determine actual tax liability . That is, the adjusted R-square for a model using all of the above variables should be greater than the adjusted R-square for a model using only an expanded income figure as the explanatory variable . The increase in adjusted R-square will be accompanied by a corresponding decrease in CRV . The difference between the CRV for the simple form of the model (i .e . CRVs using only the expanded income amount as the explanatory variable) and the CRV for the full model, CRV 1, is the change in CRV attributable to the additional explanatory variables . This is an estimate of the amount of variation in tax liability induced by the tax system . The percentage change in CRV explained by this technique is given by the following formula :

176

PETER J . WESTORT CRV Percentage change in CRV

CRV RVI

* 100

(1)

CRV , is from the simple form of the model : TL =13 0 + 13,ETI + e

(2)

CRV I is from the full model : TL = (3 0 + 13,ETI + (3 2EI + 133FTI + 134E + (3 51 + (3 6AGIADJ + (37TAXPREF + (3 8COLAB + 130 CREDITS + e

(3)

where : TL = tax liability, ETI = expanded total income, El = amount of exempt income, FTI = amount of favorably taxed income, E = personal and dependency exemptions, I = total itemized or standard deduction, AGIADJ = total deductions for AGI, TAXPREF = total tax preferences for AMT purposes, COLAB = amount of prior year losses used in the current year, CREDITS = total tax credits taken, e = error term . The contribution of each variable to the portion of CRV induced by the tax system can be estimated by a series of regressions . The full model can be fit omitting one of the explanatory variables . The difference between the CRV for this model (CRVI_ i) and the CRV for the full model (CRV I) is a conservative estimate of the change in CRV caused by the omitted variable . This process is repeated for each explanatory variable except the expanded income variable, which is the basis for comparison. The result is an estimate of the contribution of each variable to the variation in tax liability caused by the system . The formula is as follows : CRV _ - CRV Percentage change in CRV due to variable i = CRY, - CRV fI * 100 (4) DATA AND VARIABLES This study uses the individual model file (IMF) for 1992 from the IRS Statistics of Income Division (SOI) . Variables are defined as follows :


177

TL is the tax liability per return and is the income tax after credits plus the alternative minimum tax . ETI is expanded total income. It equals AGI plus the IRA deduction for taxpayer and spouse, Keogh and SEP deductions, penalty on early withdrawal of savings, alimony paid, tax-exempt interest, tax-exempt social security benefits, and tax preference items (see Allan & Iglarsh, 1996, 34) and excludes the deduction for prior-year losses . It includes all known sources of income, gain, or loss for the current tax year, including exempt income . ETI attempts to approximate taxpayers' economic income during the tax year . Taxpayers in the same ETI range have traditionally been considered economic equals (Anderson, 1985 ; Pierce, 1989 ; Allan & lglarsh, 1996) . El is the amount of exempt income reported on the return . It is the total of tax-exempt interest income plus the exempt portion of Social Security benefits . FTI is the amount of favorably taxed income . Favorably taxed income consists of net long-term capital gain for taxpayers with marginal tax rates higher than 28% . E is the total for personal and dependency exemptions . It is the number of exemptions claimed multiplied by $2,300 (the value of each exemption in 1992), and reduced by the phase-out amount. I is the total of itemized deductions taken or the standard deduction if taxpayers did not itemize . AGIADJ is the total deductions for AGI . It consists of IRA deductions for taxpayer and spouse, one-half self-employment tax, self-employed health insurance deduction, Keogh and SEP deductions, penalty on early withdrawal of savings, and alimony paid . TAXPREF is the total tax preference reported for alternative minimum tax purposes . It is intended to capture the effects of the AMT on total tax liability . COLAB is the amount of capital loss carryover that was actually used during the current year . This amount is added back so that current year economic income is not reduced by prior-year losses . CREDITS is the total of all tax credits taken, including the amount of earned income credit used to offset income tax before credits . It does not include the refundable portion of the earned income credit . The study was restricted to married taxpayers filing joint returns since this provided a better grouping of economic "equals" . The number and width of

178

PETER J . WESTORT

ranges used in prior studies has varied significantly . Anderson (1985) used 28 ranges ; the first 13 corresponding to the tax brackets then in effect, the last 15 subjectively chosen to keep ranges narrow yet observations sufficiently high. Pierce (1989) uses ten ranges ; the lower and upper ones are open-ended . The others are either $5,000, $10,000, $25,000, or $100,000 in width . Grasso and Frischmann simply use deciles . This study uses fixed-width income groups because of the distortion caused by using groups of unequal width (see Westort & Wagner, 2000 for a discussion of this issue) . An income range of $10,000 was used so that the income groups would be narrow enough to consist of taxpayers who may reasonably be considered economic equals . In addition, the data were truncated by deleting observations with ETI less than zero or more than $300,000. Truncating at ETI less than zero avoids difficulties with predicting tax liability when NOLs are involved . Truncating at $300,000 keeps the number of groups to a manageable number (30) . Allan and Iglarsh, (the only other study to use fixed income groups of the same width) truncated at zero and $200,000 to generate 40 groups with an income range of $5,000 each . While one regression for the entire data set could have been fit, Grasso and Frischmann observed substantial differences in the estimated slopes across deciles (1992, 129) . Thus, splitting the date into income groups is appropriate . The initial sample consisted of 35,676 observations? The influence of each observation on the estimates for the full model was measured using the statistics proposed by Belsley, Kuh and Welsch (1980) . Using this technique on each income group resulted in the identification and removal of eight outliers, leaving a final sample of 35,668 observations . Grasso and Frischmann also use a log transformation of expanded income . The current study also fitted regressions with a log transformation of ETI with virtually no difference in results . For this reason, the simpler, untransformed model is presented here .

RESULTS Table 1 provides descriptive statistics . Panel A provides statistics for the entire sample . Of interest are the maximum values of several variables . Exempt income (EI), favorably-taxed income (FTI), itemized deductions (I), and the amount of carryover losses added back (COLAB) all have maximum values greater than the $300,000 ETI cutoff. Some values are far greater ; for example, the highest amount of itemized deductions in the sample total $1,602,000 . This taxpayer had substantial investment income but also substantial losses from schedules E and F . The net result was an AGI of a little over $200,000 . The investment income of over $1,000,000 indicated more than adequate wealth to cover the

179


large itemized deductions, the majority of which was the investment interest paid deduction . Evidently, some taxpayers with very high amounts of income use losses and deductions to significantly reduce their taxable income such that their ETI is less than $300,000 . This gives some hint of the degree of variation existent in the data . Panel B provides the same descriptive statistics for each of the of the 30 income ranges . Table 2 is a correlation matrix for the entire sample . Although many of the correlations are significant, there are no large pair-wise overall correlations in the sample . The largest coefficient is 0 .43807 (ETI and 1), and the significance of the correlations primarily is due to the very large sample size . Correlation coefficients also were computed for each income group . Those results are not presented here in order to conserve space . However, there were only four observations of pair-wise correlations greater than 0 .6 . In the last four income groups ($260,000 through $300,000) exemptions (E) and exempt income (EI) had correlation coefficients ranging from 0 .70 to 0 .85, all with a p-value Table 1 . Descriptive Statistics for Regression Variables Panel A, Entire Sample . n = 35,668 VARIABLE TL ETI EI FTI E 1 AGIADJ TAXPREF COLAB CREDITS

n

MEAN

Std . Dev .

Minimum

35,668 35,668 35,668 35,668 35,668 35,668 35,668 35 .668 35,668 35 .668

11,380 77,931 3,654 3,040 6,632 17,575 1,775 166 350 181

15,789 68,852 17 .821 35,668 3,038 28,289 5,051 2,841 5,174 1,831

0 0 0 0 0 0 0 0 0 0

Maximum 159,600 300,000 691,500 3,206,000 41,400 1,602,000 108,420 139,600 578 .100 78,880

TL = tax liability, ETI = expanded total income, El = exempt income, FTI = favorable-taxed income, E = personal and dependence exemptions, I = itemized or standard deduction, AGIADJ = deductions for adjusted gross income, TAXPREF = total of tax preference items for AMT purposes, COLAB = amount of prior year losses used in the current year, CREDITS = total tax credits taken including amount of earned income credit used to offset income before tax credits .

180

PETER J . WESTORT

Table] .

Continued .

n = 35,668 ETI 0 TO 10,000 VARIABLE

n

MEAN

Std . Dev .

1,615 1,615 1,615 1,615 1,615 1,615 1,615 1,615 1,615 1,615

32 6,085 415 0 6,988 7,983 310 2 160 0

698 2,758 4,221 0 3,077 8,229 1,602 49 1,382 7

ETT 10,001 TO 20,000 VARIABLE n

MEAN

Std . Dev .

TL ETI El FTI E 1 AGIADJ TAXPREF COLAB CREDITS

250 15,255 963 0 6,846 8,228 398 11 95 62

758 2,851 3,098 0 2,871 14,047 1,048 328 549 156

ETI 20,001 TO 30,000 VARIABLE n

MEAN

Std . Dev .

TL ETI EI FTI E 1 AGIADJ TAXPREF COLAB CREDITS

1,214 25,028 2,147 0 6,942 8,690 544 12 103 51

954 2,880 5,422 0 2,891 8,891 1,254 295 621 196


3,392 3,392 3,392 3,392 3,392 3,392 3,392 3,392 3,392 3,392

3,821 3,821 3,821 3,821 3,821 3,821 3,821 3,821 3,821 3,821

Minimum

Maximum

0 0

21,350 10,000 144,900 0 27,600 122,600 45,360 1,471 49,730 248

Minimum

Maximum

0 0 0 0 2,300 0 0

0 10,001 0 0 4,600 0 0 0 0 0

Minimum 0 20,010 0 0 4,600 0 0 0 0 0

36,070 20,000 34,210 0 27,600 502,500 27,498 14,390 11,150 1,181

Maximum 18,290 30,000 129,346 0 25,300 183,300 30,540 10,960 15,540 2,876

Measuring Variation in Tax Liability Among Economic Equals Table 1 . ETI 30,001 TO 40,000 VARIABLE n TL ETI El FTI E I AGIADJ TAXPREF COLAB CREDITS

4,051 4,051 4,051 4,051 4,051 4,051 4,051 4,051 4,051 4,051

MEAN 2,516 35,037 2,170 0 7,057 8,746 695 98 53

Continued. Std . Dev .Minimum 1 .184 2,896 5,053 0 2,832 5,778 1,421 158 655 283


MEAN

Std . Dev .

TL ETI EI FIT E 1 AGIADJ TAXPREF COLAB CREDITS

3,797 44,896 2,144 0 7,004 10,392 807 5 III 77

1,436 2,842 4.942 0 2,844 11,567 1,792 157 930 393


MEAN

Sid . Dev .

TL EL El FTI E 1 AGIADJ TAXPREF COLAS CREDITS

5,395 54,741 1,710 0 7,016 11,784 873 24 148 88

1,965 2,834 4,409 0 2,734 10,424 2,029 609 696 461

3,937 3,937 3,937 3,937 3,937 3,937 3,937 3,937 3,937 3,937

3,085 3,085 3,085 3,085 3,085 3,085 3,085 3,085 3,085 3,085

181

0 30,001 0 0 4,600 0 0 0 0 0

Minimum 0 40,005 0 0 4,600 0 0 0 0 0

Minimum 0 50,010 0 0 4,600 5,774 0 0 0 0

Maximum 25,720 40,000 49,060 0 20,700 82,600 16,280 7,610 23,260 4,249

Maximum 31,560 50,000 64 .590 0 41,400 302,000 34,172 8 .517 47,230 5,336

Maximum 35,850 60,000 79,940 0 23,000 283,800 35,329 31,040 10,120 7,575

182

PETER J . WESTORT

Table 1 .

Continued.


MEAN

Std . Dev .

TL ETI FTI El FTI E I AGIADJ TAXPREF COLAB CREDITS

7,251 64,689 0 1,548 0 7,180 14,265 1,127 61 217 93

2,463 2,829 0 4,900 0 2,810 14,023 2,415 1,769 1,780 483


MEAN

Std . Dev .

TL ETI EI FTI E I AGIADJ TAXPREF COLAB CREDITS

9,283 74,709 1,855 0 7,173 16,336 1,390 57 224 107

3 .135 2,891 6,503 0 2,858 14,390 2,971 992 1,525 684


MEAN

Std . Dev .

TL ETI EI FIT E 1 AGIADJ TAXPREF COLAB CREDITS

11,303 84,720 2,214 0 7,078 18,817 1,978 35 191 101

3,903 2,876 7,761 0 2,768 16,249 3,848 505 983 676

2,383 2,383 2,383 2,383 2,383 2,383 2,383 2,383 2,383 2,383 2,383

1,813 1,813 1,813 1,813 1,813 1,813 1,813 1,813 1,813 1,813

1,269 1,269 1,269 1,269 1,269 1,269 1,269 1,269 1,269 1,269

Minimum 0 60,010 0 0 0 4,600 6,000 0 0 0 0

Minimum 0 70,002 0 0 4,600 6,000 0 0 0 0

Minimum 0 80,010 0 0 4,600 0 0 0 0 0

Maximum 38,790 70,000 0 107,678 0 23,000 372,800 20,982 83,040 51,54 10,310

Maximum 47,400 80,000 143,200 0 25,300 329,500 29,329 32,680 53,900 14,580

Maximum 64,300 89,998 130,100 0 18,400 188,200 39,240 10,490 17,832 13,510

183


Table 1 .

Continued .


MEAN

Std Dev .


13,416 94,863 2 .457 3 7,178 20.323 2,384 217 355 220

4,647 2,858 7,456 76 2,929 18,361 4,284 2,927 2,366 1,303

1 .099 1,099 1,099 1,099 1,099 1,099 1,099 1,099 1 .099 L099

ETI 100,001 TO 110,000 VARIABLE H

MEAN

857 857 857 857 857 857 857 857 857 857

15,522 104.854 3,139 1,533 7,074 21,998 2,729 162 558 202


MEAN

TL ETI El FTI E I AGIADJ TAXPREF COLAB CREDITS

TL FTI EI FTI E 1 AGIADI TAXPREF COLAB CREDITS

818 818 818 818 818 818 818 818 818 818

17,676 115,035 3,223 3,317 7,141 24 .470 3,335 419 14 208

Std Dev . 5,073 2,853 11,760 14,918 2,903 19 .207 4,832 3,065 4.516 1,089

Std . Dev . 5 .852 2,856 9,545 16,762 2,883 24.990 6.913 5,221 7 .489 1,498

Minimum 0 90,007 0 0 4,300 6,000 0 0 0 0

Minimum 0 10,1012 0 0 4,600 6.000 0 0 0 0

Minimum 0 110 .037 0 0 4,300 0 0 0 0 0

Maximum 59,900 99,990 81,428 2,198 23 .000 I9&600 36,144 74,890 68,010 18 .710

Maximum 57,500 110,000 252,166 367,900 29,900 315,500 40.772 82,730 116,270 20,020

Maximum 65,510 120,000 110,705 291,500 20,700 497 .800 100 .567 117,200 205,850 22,560

184

PETER J . WESTORT

Table I . ETI 120,001 TO 130,000 VARIABLE n TL ETI EI FTI E 1 AGIADJ TAXPREF COLAB CREDITS

MEAN

887 887 887 887 887 887 887 887 887 887

20,305 124,881 2,809 3,936 7,183 25,728 2,504 334 432 172


MEAN


727 727 727 727 727 727 727 727 727 727

22,269 134,966 4,407 7,533 6,998 26,493 3,854 204 593 252


MEAN

624 624 624 624 624 624 624 624 624 624

25,133 144,984 3,873 11,705 7,076 28,904 3,814 220 902 276

TL ETI El FIT E 1 AGIADJ TAXPREF COLAB CREDITS

Continued.

Std . Dev . 5,551 2,828 9,559 14,471 2,820 22,560 4,662 3,051 3,739 1,341

Std . Dev . 6,497 2,950 12,191 54,207 2,653 21,348 7,579 2,495 3,312 1,766

Std . Dev . 7,592 2,938 10,823 130,026 2,765 25,678 6,251 2,136 13,316 2,027

Minimum 0 120,024 0 0 4,600 6,000 0 0 0 0

Minimum 0 13,0026 0 0 4,600 6,000 0 0 0 0

Minimum 0 140,013 0 0 4,600 6,000 0 0 0 0

Maximum 55,510 130,000 120,624 142,500 23,000 339,300 33,100 49,430 75,330 25,220

Maximum 42,310 140,000 102,065 1,241,000 18,400 224,000 66,103 0,470 52,470 31,090

Maximum 73,490 150,000 82,823 3,206,000 16,100 233,200 37,226 34,150 331,000 28,590

185


ETI 150,001 TO 160,000 VARIABLE n TL ETI El FTI E I AGIADJ TAXPREF COLAB CREDITS

Table 1 .

Continued .

MEAN

Std . Des .

548 548 548 548 548 548 548 548 548 548

27,009 154,953 4,889 8,053 7,245 33,241 3 .570 272 819 385

.000 ETI 160,001 TO 170 VARIABLE n

MEAN

TL ETI El Fri E 1 AGIADJ TAXPREF COLAB CREDITS

477 477 477 477 477 477 477 477 477 477

28,823 164,797 5,271 10,399 6,866 33,419 4,812 267 1,294 543

ETI 170,001 TO 180 .000 VARIABLE a

MEAN

TL ETI El Fri E I AGIADJ TAXPREF COLAB CREDITS

434 434 434 434 434 434 434 434 434 434

31,721 174,861 5,489 13,170 6,761 35,632 5,712 277 1,217 464

8,015 2,901 14,062 25,716 2,961 41,949 6,031 2,261 7.512 2,351

Std. Dev . 8,607 2,924 23,501 34,663 2,724 32,941 9 .618 3 .123 8,514 3 .107

Std . Des . 10,249 2,849 17,081 67 .485 2,605 49,978 9,287 3,477 9,383 2,408

Minimum 0 150,031 0 (1 4,508 6,000 0 0 0 0

Minimum 0 160,012 0 0 4,140 6,000 0 0 0 0

Minimum 0 170,100 0 1) 3,772 6.000 0 0 0 0

Maximum 59,630 160,000 126,660 262,400 22,540 692,800 33,803 34,530 162,000 30,160

Maximum 41,330 169,970 436,600 398 .200 21,160 411,700 102 .199 55,360 102 .200 34,810

Maximum 103,800 180,000 144,800 980 .800 15 .820 820,000 51,729 69,210 163,542 31 .670

186

PETER J . WESTORT

ETI 180,001 TO 190,000 VARIABLE n TL ETI El FTI E 1 AGIADJ TAXPREF COLAB CREDITS

Table] .

Continued .

MEAN

Std . Dev .

384 384 384 384 384 384 384 384 384 384

33,576 185,223 7,632 15,561 5,850 36,766 5,417 518 796 948

11,700 2,870 22,364 86,332 2,302 32,432 9,848 5,313 4,992 4,490


MEAN

Std . Dev .

TL ETI EI FTI E 1 AGIADJ TAXPREF COLAB CREDITS

361 361 361 361 361 361 361 361 361 361

36,388 195,016 9,667 10,357 5,359 39,003 5,936 907 1,193 464


MEAN


327 327 327 327 327 327 327 327 327 327

36,188 205,306 15,851 14,169 5,056 46,021 5,661 804 1,983 992

11,622 2,910 43,108 37,842 2,249 39,445 9,753 8,406 8,184 3,356

Std . Dev . 15,495 2,789 49,942 58,628 2,020 66,116 11,728 6,876 20,243 5,658

Minimum 0 180,015 0 0 3,404 6,000 0 0 0 0

Minimum 0 190,025 0 0 3,036 6,000 0 0 0 0

Minimum 0 200,062 0 0 2,668 0 0 0 0 0

Maximum 119,200 190,000 217,450 1,492,400 16,150 238,000 63,088 90,940 63,670 40,160

Maximum 94,720 200,000 670,500 383,032 16,700 299,800 51,567 112,600 146,200 45,460

Maximum 97,050 209,977 423,000 884,300 13,250 738,300 83,339 89,130 356,500 54,730



Table l .

Continued .

MEAN

Std . Dev .

307 307 307 307 307 307 307 307 307 307

41,224 215,432 13,409 17,078 4,550 43,484 5,364 902 1,139 680


MEAN

TL ETI El Ff1 E 1 AGIADJ TAXPREF COLAB CREDITS


304 304 304 304 304 304 304 304 304 304

39,375 224,899 22,292 17,564 4,116 51,978 6,423 977 1,374 822


MEAN

TL ETI El FTI E 1 AGIADJ TAXPREF COLAS CREDITS

309 309 309 309 309 309 309 309 309 309

44,742 235,297 16,050 22,749 3,560 51,280 6,847 703 585 558

17,127 2,832 37,591 75 .511 1,898 55,285 9,893 7,893 10,784 4,092

Std. Dev . 18,294 3,022 63,132 71,173 1,851 64,905 11,812 5,485 7,546 5,267

Std . Dev . 16,860 2,768 43,818 129,328 1,632 111,164 13,082 4,039 2,278 3,873

187

Minimum 0 210,168 0 0 2,300 6,000 0 0 0 0

Minimum 0 220,200 0 0 1,932 6,000 0 0 0 0

Minimum 0 230,100 0 0 1,564 6 .000 0 0 0 0

Maximum 159,600 220,000 211,050 1,152,000 13,340 468,800 60,000 124,500 183,500 55,100

Maximum 93,340 230,000 691,500 981,300 14,080 536,400 69,063 57,410 95,970 51,250

Maximum 91,250 240,000 295,296 2,013,000 10,350 1,602,000 108,420 39,680 31,280 63 .370

188

PETER J . WESTORT

Table 1 . ETI 240,001 TO 250,000 VARIABLE n TL ETI El FTI E 1 AGIADJ TAXPREF COLAB CREDITS

MEAN

306 306 306 306 306 306 306 306 306 306

46,142 245,324 19,029 12,982 2,924 51,292 5,395 1,721 774 881


MEAN

TL ETI El FIT E 1 AGIADJ TAXPREF COLAB CREDITS

349 349 349 349 349 349 349 349 349 349

51,008 254,933 17,181 22,969 2,262 50,119 4,545 882 630 596


MEAN


305 305 305 305 305 305 305 305 305 305

53,577 264,942 18,590 21,852 1,903 48,308 5,294 922 1,061 610

Continued .

Std . Dev . 19,260 2,867 51,308 41,636 1,623 60,147 10,217 9,457 3,612 5,893

Std . Dev . 18,822 2,804 49,496 108,627 1,239 61,912 9,043 4,971 2,926 4,800

Std . Dev . 19,509 2,822 53,825 90,239 1,878 52,879 11,428 6,824 5,993 3,710

Minimum 0 240,100 0 0 1,196 6,000 0 0 0 0

Minimum 0 250,018 0 0 828 6,000 0 0 0 0

Minimum 0 260,038 0 0 460 6,000 0 0 0 0

Maximum 97,420 249,900 385,800 334,100 11,500 551,400 42,392 92,480 44,600 65,980

Maximum 104,400 260,000 358,000 1,482,467 9,200 618,900 37,197 47,470 44,810 68,130

Maximum 128,200 270,000 395,100 1,161,450 13,800 645,500 83,873 88,660 69,980 52,370


189

Table 1 . Continued . ETI 280,001 TO 290,000 VARIABLE

n

MEAN

296 296 296 296 296 296 296 296 296 296

57,042 274,879 16,679 19,982 1,263 49,806 4,774 842 658 683


MEAN

319 319 319 319 319 319 319 319 319 319

56,752 284,893 20,558 27,080 1,036 52,530 4,963 1 .772 1,784 1,434


MEAN

264 264 264 264 264 264 264 264 264 264

60,056 294,662 21,445 29,854 851 53,360 5 .814 828 3,115 479

TL

ETI El FTI

E 1 AGIADJ TAXPREF COLAB CREDITS

TL

ETI El FTI E

1 AGIADJ TAXPREF COLAB CREDITS

TL

ETI El FTI E 1 AGIADJ TAXPREF COLAB CREDITS

Std . Dcv .

18,817 2,985 53,238 62,620 1,789 59,446 10,421 7,588 3,080 4,041

Std . Dev .

21,411 2,738 53,784 97,557 1,983 57,073 10,741 12,167 11,069 8,675

Std. Dev.

21,345 2,928 51,179 115,163 1,756 61 .031 11,130 4,663 35,823 1,503

Minimum

0 270 .044 0 0 92 6,000 0 0 0 0

Minimum

0 280,048 0 0 0 6,000 0 0 0 0

Minimum

0 290,005 0 0 0 6,000 0 0 0 0

Maximum

100,800 279,990 512.600 600,500 13,800 652,100 80,771 121,600 34,710 47,930

Maximum

98,910 290,000 306,119 879,800 13,800 343,300 80,429 139,600 169,079 78 .880

Maximum

104,400 300,000 280,700 1 .411,000 9,200 464,200 46,050 50,860 578,100 11,730

1 90

PETER J . WESTORT


191

of 0 .0001 . These correlations indicate that the contribution of El to variation in tax liability in the last four income ranges is somewhat understated, although still quite high. This effect is discussed later . Table 3 provides regression results for each of the two models by income group . Results for the simple model are not inconsistent with the Grasso and Frischmann results, although they did observe high-adjusted R-squares at the highest deciles . This high decile result, however, is likely due to the width of the tenth decile, which is open-ended . For the full model, the adjusted R-square is quite high for all but a few of the income groups and is highly significant for all income groups . Clearly, the fully-specified model explains significantly more of the variation in tax liability . A clearer comparison of the models, however, is provided in Table 4 . Columns 2, 3, and 4 of Table 4 contain the CV, CRV s , and CRV f measures, respectively, for each income group . The next two columns show the improvement of the CRV measure for the simple model (CRV) over the traditional CV measure, both in absolute difference and in percentage change . The last two columns provide the same analysis for the improvement of the CRV measure for the full model over the CRV measure for the simple model (CRV f over CRV) . Visual inspection of the absolute differences indicates that the CRV s measure generally is not much smaller than the CV measure and, in fact, in several of the higher income groups (e.g . 230,001 to 240,000, 240,001 to 250,000) is slightly larger . The CRVf measure, however, generally is much smaller than the corresponding CRV measure . This difference is highlighted in the percentage columns . The percentage change for CRV over CV ranges from a minus 0 .15% for the 240,001 to 250,000 income group to a high of a little over 6% for the 20,001 to 30,000 income group . The percentage change associated with the CRV f measure, however, ranges from slightly less than 5% for the 10,001 to 20,000 income group to more than 73% for the 280,001 to 290,000 income group . Several observations can be made from these results . First, ETI does not explain much of the variation in tax liability, even with wider income ranges, than in most of the deciles observed in Grasso and Frischmann . As noted in the introduction, implicit in the Grasso and Frischmann model is the assumption that, in general, tax liability should increase as income (or ETI) increases . Secondly, the CRV for the simple model does not provide significant improvement over the traditional CV measure and, at some of the higher income groups, is a poorer measure . Thirdly, while the full model reduces the CRV measure substantially, the largest decrease is still only about 75% and for many of the income groups is less than 50% . Clearly, then, a large part of the variation

192

PETER J . WESTORT

U W W K

F

7 w

Q

a

w n a

omoo O'S'0N O'n'SwmN'0< 0 ut d 0 0 0 0 N 0N 00 0od b o 0 0 o c 0

0

000 0 0000 0 do 0 0 0 0 0 0

O

U

b

7

0 6

w

N

M

V

N

0 n

7 W O' GC V 7 V1 V1 V1 V1 N 7 vt O O q [O V7 7 0' 0 'O 7 V' N 0 V M N N~p

V7 O\ N V oo V, en N v1 N OC f4 W 0''0 V7 00 to U M `n i

Un h M N

W

W

bA N

F W

v O' b II O

F

F

N

N m 0 O V1 7 N m y 7 N

..a F

az

w

W

N x CC d V N 0

N O~ p

S O -NN+ 0 0 0 2 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

n O 0 0 0 0 c0 00 00 80 0 0 0 0 0 0 0 0N 0

7 td

V) V) O

7 O\ O N N'0 W 00 N 7 N M N M^ O N N N N 7 - V1 7 to N -

W

u

N -+ N V7 C1 to U O pi~ N OO N N V O N 'O +iInM W OInT (n O NM W N O W W CO N bb Vi V

"O 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ----------------

o-Nen'SU

N000

-NenZU

193


a WW ~+1 N N NT O\ cO cO Mm 0, V7b y V7 d b r b r D W 0 0 0 0 0 0 0 0 0

'O O O\ N W W O O 0 0 0 0

-------

------

8800000

000008

•

d C6666

c0 O 7

h V 0m T

b 0 7 t

b -

d d

0

d d

0

O'0 en b

O~

N Mm r 7 h y O

N N .- N 7 oom

b a) a a a O

U N to r cc 0 0 O O O O O O O O O C O O O C- P O w

0

N N

000 0 00

a c

0 0

00

vrNM-'ONr m v0'r ro0

N r0Mn n~ MV1M-+r W rn f~lh 7Mm NO^ N. mM N r W7 en f- M 7 O r N N 0 0 0 0 0 0 0 0 0 0 0 0 0 V W 7V MO O^ -O m b N 7 r r b vtO MM 'O N 'O 7 •

0 -' -- N O O O O N O O -'

V V G r r O v mm M en

O vl cO1 \O O\ 7 mm m nl

---N NNN N NNN NNm

• 2 2 2 2 0 0 0 0 0 2 3 2

- - - - - - - - - - - - .- - - N N N N N N N N N N

1 94

PETER J . WESTORT

x U v

m> 04

U c

u °0 .t .I .

U

r

u

r N h N

U 7

E

o U

o

r a •

o

0

2

Corn 7 rn N NN-7 NN N000 m 7 00 m0-N e N

I

I

.

W

o

U

.I .

h

Nvl 'C vl 0C W rn

Vl o~

--

el V M O M M r W l0 O O O N O M m O O N ~p h 7 M N N 0 0--- N O O O O O O -~ .

U

• c

~

6 v,

o

E

u

m

o

U ^~

y

. =

A

o N to 7 Me07 W N aN-N O c 0 0 7yenyenN N vl (V m-NCONm N N N N 7rn N -+ - 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 I

U

.~ 00 'O W O (n N 0; 7 MN NN--- vl 7 rn 7 m'07 0N 0' N rn rn NN N N N W tai [V -i [V V N N 0

vl N N N W v, b- N CC N b rnw,e M- O N i l Nmm~nrnmo~v, oa,evleNNOa 0qC Nv -7i Q N en M M to M M m m N N N N N en In M 7 7 N N

N

M W 7COT NN NV7

>

C

W

Wul 7 NN ulv) N ~Ob-W 1~7 '- 7MW- N M N N en N N M 7n to 7 7

enn M M M M ten M

U

NNN p NM7vlcoNW-0 0 u0 0 0 0 0 0 0 0 0 0 0 9 0 0 0 0 v0 O O O - - - - - - - - - - - - - - - - - - - - O

N M 7

N W

-NN

195


UV 0

ou

U

00 > i U c cv e E ^~ o i c V

W

W N N

b b Vl

e N N

7

v

U

0 0 N N N N

U

v

cI.

r V

>U

0 0 0 0 0 0 0 0 I

I

I

I

I

U

C (n O E

7c

Q yC

O

O a V

N

N 'O 100 N 1 0 0 0 0 0 0 0 00000000 I

I

I

I

I

u

a U a d H

a

N

d a ~ ~ U

0 7 fn a

N 0• y to N h u= N 0' 0' N N

d C E o

'0 V17 -+wV N 0 00 O\ m O N h b'0 M M v M a m-,

e 0070-rnr, v v'0N rnv 0'r ~, b N N V) 'O - '0 M M Ol M M 7 MN V

II 1

N N P V N N N N N N fn 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 N N N N N N N N N N N N

196

PETER J . WESTORT

in tax liability using this measure is due to specification error . This has important implications for the interpretation of prior studies . The CV is a noisy measure of the variation in tax liability and much of the variation should not be attributed to the tax system . See, for example, Pierce (1989, 61) who comments that the effect of homeowner preferences on horizontal equity is small, - "[T]he largest positive weighted average percentage change in the overall analysis was 5 .70% . None of these policy options would greatly improve horizontal equity ." The effect may well be much larger than she thinks, but the CV measure is noisy and minimizes the impact . How much of the CRV is attributable to the tax system? It is the difference between the CRV s and the CRV f measures (column 7 of Table 4) . If CRV s is the result of using only ETI as the explanatory variable and CRVf is the result of adding all the other explanatory variables, then the difference in the CRV of the two models is the reduction in CRV caused by the additional explanatory variables . To the extent these variables account for the provisions of the tax system, the reduction in CRV is that portion of the CRV due to the tax system . Thus, it is column 7 from Table 4 that is a more precise measure of variation in tax liability due to the tax system (horizontal inequity) . Comparing the values in column 7 with those in columns 2, 3, and 4 (CV, CRV5, and CRV f) results in a substantially different interpretation of the data. The CV and CRV measures all have large values in the first two income ranges, rapidly decreasing and leveling off. For CRV f, the measure gradually decreases as ETI increases . While the value in column 7 is relatively large for the first ETI range, it decreases over the next three ETI ranges and then gradually increases as ETI increases . Not only is this trend different from the trend of the other measures (CV, CRV s , and CRVf), it also is different from trends observed in prior research . Some prior horizontal equity researchers provide only the percent increase or decrease in CV (e.g . Anderson, 1985, 1988 and Pierce, 1989) . However, others present raw CV or CRV data. In these later cases the trend always is the same : large values in the first one or two income ranges, then leveling off or gradually decreasing as the income measure increases (e .g . Allen & Iglarsh, 1996 ; Grasso & Frischmann, 1992, and Enis & Craig, 1990) . The only exception is where the last income range is open-ended (Grasso & Frischman, 1992) in which case the measure increases . A measure of horizontal inequity that increases as income increases, however, may be more intuitive . In all income ranges, some taxpayers will manage to pay little or no income tax . As the income range increases, the distance from the mean tax liability for that range to zero tax liability also increases . Thus, while it may not be considered too inequitable for taxpayers with relatively low incomes to have a zero tax liability, it generally would be considered very inequitable for a taxpayer with a very high income to have a


197

zero tax liability . Higher measures of inequity at the higher income levels, therefore, should not be too surprising . This result also is consistent with Westort and Wagner (2000) who observe greater variation in the upper income levels using a CV ratio technique . Using the technique described in equation 4, Table 5 presents the percentage change in CRV due to each variable . Columns 2, 3, and 4 of Table 5 provide the CV, CRV s , and CRV 1 for reference purposes . Column 5 is the difference between columns 3 and 4 and is the decrease in CRV caused by using the full model rather than the simple model . It is the same value as column 7 of Table 4 . Columns 6 through 13 of Table 5 indicate the percentage of column 5 attributed to the omitted variable . This is an estimate of the amount each variable contributes to the overall variation . Several observations can be made regarding these results . The lowest ETI range has negative values for four of the variables . This seeming anomaly is simply because the additional explanatory power of the omitted variable is less than the effect on CRV of adding another explanatory variable (see appendix A for an illustration) . For 22 of the 30 income groups, itemized deductions (I) is the greatest cause of variation in tax liability . In many of the groups, greater than half of the percentage change in CRV is attributable to this variable . Exempt income (EI) is the greatest cause of variation in seven of the eight remaining groups . For those groups with itemized deductions as the leading cause of variation, generally the second largest cause of variation is due to the exempt income variable . For two of the lower ETI ranges the exemption amount (E) is the leading or second leading cause of variation in CRV . The effect of exemptions, however, decreases rapidly as ETI increases . Tax credits (CREDITS) and AGI adjustments (AGIADJ) generally are the third and fourth largest causes of variation, but not to the same order of magnitude as itemized deductions and exempt income . The effects of tax preferences (TAXPREF) and the deduction of prior year losses (COLAS) generally are quite small, with one or two exceptions . Finally, the favorable tax treatment afforded capital gains has virtually no impact (as measured by the amount of favorably-taxed income, FTI) . For 1992, the highest tax bracket was 31% and the maximum tax on capital gain was 28% . Thus, there is only favorably-taxed income for taxpayers in the 31% bracket who have capital gains . Since the tax rate difference was only 3%, this tax benefit has very little effect on variation in tax liability . As stated earlier, E and El are highly correlated in the last four income groups . Collinearity of two explanatory variables means that it is difficult to infer the separate influence of such explanatory variables on the response variable (Belsley, Kuh & Welsch 1980, 86) . However, given the trends observed in Table 5, it is unlikely that this collinearity would cause a different interpretation of the result . That is,

198

PETER J . WESTORT

N 7- VAQ WN'0 m 7 CO m Vl m 7 ~°~OVm my ab Ovvtm tm O N^ N W O, m vt n O -+ -- N^ -+ O O N N ~O vl V 'O O n 7 0 v, n 'tNmmn N vl N m W N nmn N O 0 ~O ot v, T enU)N'0N N V vt ~D m N Omn O ooci N nooo-: o000 00000000

I

'"

I

I

I

g g o o g o 0 0 0 0 0 0 0 0 0 9 0 0 0 0 o

Nmv vienoo

^ - ^ ^ - ^ ^ - ^ N N


- - - - - - - - - - - N N N N N N N N N N

199

200

PETER J . WESTORT

exempt income (EI) is consistently the greatest or second greatest cause of variation in tax liability . In the last four income groups, its contribution is still large even if it is somewhat understated . Similarly, the contribution of exemptions (E) is very small in the last 15 income groups . Thus, even if its contribution is understated, it is at most the third largest cause of variation in tax liability in these income groups The effect of itemized deductions and exempt income is so pervasive that it clearly has policy implications . Concerns about the distribution of the explicit tax burden clearly need to focus on these two areas . Additionally, while the phase-out policy for exemptions appears to be effective, the phase-out policy for itemized deductions does not . Note that the percent of variation due to exemptions decreases rapidly and is no more than one or two percent for the upper half of the income groups . The effect of itemized deductions on tax liability variation, however, is large for almost all income groups . In general, the contribution of this technique for horizontal equity studies and for policy analysis is that it allows the researcher to decompose the horizontal equity measure to determine the sources of that inequity . The decomposition can be done both before and after any proposed or actual tax law change .

CONCLUSIONS This study provides a technique for measuring more precisely the variation in tax liability among economic equals . It is a more conservative measure in that it tends to understate variation rather than overstate it as prior studies have done . This is important because it allows those concerned with tax burdens to make more accurate statements about the magnitude of various IRC provisions on the variation in that tax burden . Additionally, this study provides a means to decompose the variation in tax liability into the types of IRC provisions that cause it . This decomposition allows researchers to observe the relative size of the variation in tax liability caused by various IRC provisions . This more precise measure and the ability to decompose taken together have important tax policy implications . Those concerned with the distribution of tax liability now can observe what contributes the most to variation in that liability and how that contribution changes as income changes . This should be of concern in assessing any proposed or actual tax law change . Trends observed in the current study are the large impact of itemized deductions, exempt income, and exemptions, respectively, and the lack of impact of favorably-taxed income . The phase-out policy for exemptions can be observed in the higher income brackets . Future studies might focus on specific items such


201

as tax-exempt interest, not just total exempt income, or the appropriate number and width of income brackets for observing the effects of tax policy . Finally, an unanticipated but significant result of this study is the observation that variation in tax liability increases as income increases . While this has intuitive appeal, it is in contrast to some prior studies .

NOTES 1 . Another method is to use step-wise regression and measure the change in CRV as each variable is added to the model . Both methods were observed to provide similar results . Since the SAS program used to generate the statistics does not readily provide CRV in a step-wise regression, they were computed manually . For simplicity of presentation and because step-wise is not theoretically superior, the iterative approach was used in this paper . 2 . Exemptions, deductions and credits have been studied with respect to their impact on progressivity and vertical equity (see, e .g . Seetharaman, 1994) . Also Minarik (1980) illustrates the extent to which various provisions are used to go from nominal tax rate to actual tax rate at various income levels, but provides no measure of dispersion . 3 . Outliers also are a representative part of the sample . The justification for removing them is that they obscure underlying relationships . Only those outliers that, by themselves, had a significant effect on the CRV of their income group were removed . While leaving outliers in the sample would have resulted in a few extreme parameter estimates for some variables in a few income levels, it would not have changed the overall results and conclusions . 4. Observing the parameter estimates and t-statistics for the regressions confirms this result . That is, in the regression where El is omitted, both the parameter estimate and t-statistic for E become very large, but not larger than that for itemized deductions (I) . Similarly, in the regression where E is omitted, the parameter estimate and t-statistic for El become even larger, but not larger than that for l .

ACKNOWLEDGMENT I would like to thank Tim Rupert, Janet Wagner, and two anonymous reviewers for their insightful comments and recommendations . An earlier version of this paper has benefited from comments of participants at the Northeast Regional Meeting of the American Accounting Association .

REFERENCES Allen, R . G ., & Iglarsh, H . J . (1996) . A cluster analysis of horizontal tax equity. The Journal of the American Taxation Association, IR (Spring), 31-46 . Anderson, K . E . (1985) . A horizontal equity analysis of the minimum tax provisions : An empirical study . The Accounting Review, 60 (July) 8, 357-371 .

202

PETER J. WESTORT

Anderson, K. E. (1988) . A horizontal equity analysis of the minimum tax provisions : 1976-1968 tax acts . The Journal of the American Taxation Association, 10 (Fall), 6-25 . Anderson, K . E., Hill, Jr., T . J., & Murphy, D . P. (1995) . Horizontal equity and implicit taxes . The Journal of the American Taxation Association, 17 (Fall), 6-25 . Belsley, D . A ., Koh . E ., & Welsch, R . E . (1980). Regression Diagnostics . New York : John Wiley & Sons. Bittker, B . 1 . (1980). Equity, efficiency, and income tax theory : Do misallocations drive out inequities? In : H. J . Aaron & M . J . Buskin (Eds), The Economics of Taxation . Washington, D .C. : The Brookings Institution . Enis, C . R ., & Craig, D . L. (1990) . An empirical analysis of equity and efficiency attributes of degressive forms of a flat tax . The Journal of the American Taxation Association, 11 (Spring), 17-33 . Grasso, L . P ., & Frischmann, P . J . (1992) . Measuring horizontal equity : A regression approach . The Journal of the American Taxation Association, 14 (Fall), 123-133. Jenkins, S . P. (1988) . Empirical measurement of horizontal inequity . Journal of Public Economics, 37, 307-329. Kaplow, L. (1989) . Horizontal equity : Measures in search of a principle . National Tax Journal, 42 (June), 139-154 . Minarik, J. J . (1980) . Who doesn't bear the burden? In : H . J . Aaron . & M. J . Boskin (Eds), The Economics of Taxation (pp . 55-68) . Washington, D . C . : The Brookings Institution . Musgrave, R . A . (1959) . The Theory of Public Finance . New York : McGraw-Hill . Pierce, B . J . (1989) . Homeowner preferences : the equity and revenue effects of proposed changes in the status quo . The Journal of the American Taxation Association, 10 (Spring), 54-67 . Seetharaman, A. (1994) . An isolation of the effects of personal deductions, tax credits, and the tax rate schedule on U.S . individual income tax progressivity and income inequality . The Journal of the American Taxation Association, 16 (Spring), 101-121 . Scholes, M ., & Wolfson, M . (1992) . Taxes and Business Strategy: A Planning Approach . Englewood Cliffs, NJ : Prentice Hall, Inc . Weston & Wagner . (2000) . Toward a better measure of horizontal equity . Working paper . University of Massachusetts Boston .


203

APPENDIX A The explanation for the apparent anomaly is as follows . Assume three variables : A, B, and C where : A equals the sum of the squared error terms for a regression, B equals the number of observations, and C equals the mean of the dependent variable . Then the CRV is computed as follows :

CRV =

V A / B B-k-1

C

* 100

where k is the number of explanatory variables . Assume further that all the explanatory power of the model is contained in the first two variables and that adding six additional explanatory variables adds no explanatory power . The sum of the squared error terms for both versions of the model would be almost identical, and would be represented by variable A . The CRV for the two models would be as follows : CRV =

CRV .=

VA/B C

1 * 100

VA /B-9* C

100

In such a case, the CRV for the full model would be larger than the CRV for the reduced model .

THE USE OF GENERALLY ACCEPTED ACCOUNTING PRINCIPLES (GAAP) IN TAX LITIGATION G. E . Whittenburg, William A . Raabe and James R . Lackritz

ABSTRACT Over the years, Generally Accepted Accounting Principles (GAAP) issues have been raised in tax litigation by both taxpayers and the IRS . By studying the tax cases in which GAAP has been used, the authors were able to identify which issues the parties raised, which issues the courts used in making decisions, and whether raising the GAAP issue helped the taxpayer or the IRS prevail . The information developed in this study is useful to tax advisors and the IRS, in deciding when and how to use GAAP as part of a tax litigation strategy, and to the courts, in assessing the proper use of accounting standards in determining taxable income .

INTRODUCTION This paper examines the use of Generally Accepted Accounting Principles (GAAP) issues raised in tax litigation . GAAP is the basis for financial accounting reporting, taught in major universities and used by all businesses to some extent, but required only of publicly-traded companies . The calculation

Advances in Taxation, Volume 13, pages 205-218 . Copyright © 2001 by Elsevier Science Ltd. All rights of reproduction in any form reserved . ISBN : 0-7623-0774-9 205

206

G . E . WHITTENBURG, W . A . RAABE AND J . R . LACKRITZ

of taxable income is based on the provisions of the Internal Revenue Code . The tax law necessarily entails an accounting system in itself, one that is charged with deriving the taxpayer's annual taxable income, but matters of accounting periods and methods are not central to the language of the tax law . The Internal Revenue Code (IRC) states that the method of accounting to be used in calculating taxable income is ascertained as follows . SECTION 446. GENERAL RULE FOR METHODS OF ACCOUNTING (a) General Rule Taxable income shall be computed under the method of accounting on the basis of which the taxpayer regularly computes his income in keeping his books (i .e., financial income) . (b) Exceptions If no method of accounting has been regularly used by the taxpayer, or if the method used does not clearly reflect income, the computation of taxable income shall be made under such method as, in the opinion of the Secretary, does clearly reflect income. Over the years, the calculation of taxable income has evolved to the point that there are so many exceptions from typical book income computations, either because of requirements of the IRC and regulations or as authorized by §446(b) above, that taxable income often bears only a slight resemblance to financial accounting income for some taxpayers . GAAP concepts still greatly influence the calculation of taxable income in certain areas, such as inventories, discharge of indebtedness, cost of goods sold, accounting methods, depletion, and others . This study examines to what extent the federal courts use GAAP to calculate taxable income in decisions controlled by the Code .

Tax Accounting vs. Financial Accounting

The language of the U .S . Supreme Court in Thor Power Tool' points out that financial and tax accounting systems have different purposes and therefore almost always produce different income numbers for an accounting period . Financial accounting standards are designed to protect investor interests and typically tend toward more conservative (meaning smaller) income amounts, whereas protection of the fisc (and specification by Congress or the Treasury) often requires that taxable income is recognized earlier and in a greater magnitude than corresponding book income items .

The Use of Generally Accepted Accounting Principles (GAAP) in Tax Litigation

207

The plain language of §446 seems to mean that the IRS can impose a tax accounting method upon the taxpayer only when the taxpayer uses no corresponding method for book purposes or when the taxpayer's method does not clearly reflect income . Two observations can be made at this point . Very few taxpayers with publicly-traded equity, and very few others who are financially in a position to litigate a tax dispute, lack pertinent accounting methods? Perhaps the only situation where this clause might be used is where the tax law imposes a computation that exists only for purposes of completing a tax form, like that used in computing a base-period amount to be used in deriving a tax credit . Almost certainly the IRC language requires that the taxpayer's method clearly reflect taxable income, as the tax law has no apparent authority to affect a financial accounting income calculation . Logically, because nearly all businesses apply GAAP principles in keeping their financial records, one might conclude that any GAAP-approved method must be allowed by the IRS, as tax accounting follows financial accounting under the terms of §446(a), and GAAP procedures, by consensus of the US business community anyway, produce a clear reflection of income .; Thus, by using GAAP, the taxpayer would automatically meet the clear-reflection standard . More broadly, the courts have followed the language of the regulations and refused IRS efforts to impose a tax accounting inventory method on the taxpayer where the method in use clearly reflects income, follows the best accounting practices of the taxpayer's trade or business, and is applied consistently from year to year .4 Also improper, where the taxpayer's tax accounting method does not clearly reflect income, is the substitution of an IRS-selected method that also does not clearly reflect income .= In Wal-Mart Stores, the opinion found conformance with GAAP due to the unqualified audit opinion issued by the taxpayer's financial accounting auditors . The court also held that when a taxpayer follows a GAAP method, it necessarily met the "best accounting practices" tesL 7 The Tax Court in Wal-Mart failed to recognize that: an unqualified opinion does not mean that the auditor asserts that the taxpayer's computation of income is "correct" according to some absolute standard, as is the purpose of the tax laws, only when there is found a material departure from GAAP is an unqualified opinion unavailable to the business, and GAAP need not be followed in any context where the amounts involved are immaterial in the auditor's judgment.

208

G . E. WHITTENBURG . W. A . RAABE AND J . R . LACKRITZ

But other courts have rejected the idea that all GAAP methods meet the requirements of §446(a) . For instance, the Eleventh Circuit has struck down a consistently applied GAAP-approved inventory method because it did not clearly reflect income .' Thor Power Tool itself states that there is no overriding presumption that an accounting practice conformable to GAAP thereby is valid for tax purposes,' but Reg . §1 .446-1 can be read to conclude that "ordinarily" GAAP methods meet the clear reflection requirement . 1 ° Nonetheless, the IRS at least once twisted the statutory language so as to require of the taxpayer proof that the book method in use clearly reflects income ." While this construction cannot be supported, one could agree with the courts as to the "missing" word in the statute, such that a tax accounting method must clearly reflect taxable income, which is not strictly a designed purpose of GAAP . A taxpayer who follows GAAP requirements has met a necessary, but not always sufficient, test of §446(a) . This allows the IRS, or perhaps the courts, the last word on the use of GAAP as an effective defense of a tax return position . It is appropriate, then, that we examine who brings about the "GAAP defense" in tax litigation, and whether such an approach is convincing to the courts as a result .

Research Method

Kleinrock's Tax Expert was used to identify the tax cases from U .S . federal courts decided during an 11-year period (1988-98) in which GAAP methods were cited by the court . The cases were reviewed and the following data were collected for each case . • The tax issue for which GAAP was discussed (e.g ., inventories) . • Whether one of the parties introduced a GAAP argument into the litigation, and who did so (taxpayer or IRS) . • Whether the court relied on GAAP in crafting its opinion . • Descriptive data (e .g ., which court, dates) . • Who prevailed in the decision (taxpayer or IRS) .

Seventy-two cases that discussed GAAP in a tax setting were found . A list of these cases is shown in Table 1 . The data collected were analyzed for nominal level association using the chi-squared test. A logit regression model was then developed to aid the parties to a tax dispute in predicting the probability of winning a tax case where a GAAP issue was raised .


Table l .

209

Cases Used in Study .

Amerco & Subsidiaries, 96 TC 18 American Telephone & Telegraph Co ., TC Memo 1988-35 AmSouth Bancorporation v . US, KTC 1988-142 ; 88-I USTC 19232 Ansley-Sheppard-Burgess Co ., 104 TC 367 Apple Computer, Inc ., 98 TC 232 Applied Communications, Inc ., TC Memo 1989-469 Boecking, TC Memo 1993-497 Buckley, TC Memo 1994-470 Buyers Home Warranty Co., TC Memo 1998-98 Carlstedt Associates, Inc ., TC Memo 1989-27 Centennial Savings Bank FSB v . US, KTC 1988-202; 88-I USTC 19153 CF Industries, Inc ., TC Memo 1991-568 Challenge Publications, Inc . v . Commissioner, KTC 1988-107 ; 88-1 USTC 99327 Citizens & Southern Corporation ., 91 TC 463 Clark Equipment Co ., TC Memo 1988-111 Colorado National Bankshares, Inc ., TC Memo 1990-495 Connecticut Yankee Atomic Power Co. v . US, KTC 1997-418 ; 97-2 USTC 950,693 Cook, TC Memo 1997-378 Cottage Savings Assoc . . 90 TC 372 CSX Corp . v . US, KTC 1996-206 Dayton Hudson Corporation ., TC Memo 1997-260 Doubleday & Co ., Inc . v . US, KTC 1989-4 : 89-2 USTC 19549 Fidelity Associates, Inc ., TC Memo 1992-142 First Federal Savings & Loan Assoc . of Temple v . US, KTC 1988-188 ; 89-1 USTC 19321 Ford Motor Co ., 102 TC 87 Gamac Grain Co ., 95 TC 7 Gamac Grain Co., TC Memo 1991-363 General Dynamics Corporation ., TC Memo 1997-420 Gold Kist, Inc ., 104 TC 696 Hachette USA Inc., 105 TC 234 Hallmark Cards, Inc ., 90 TC 26 Hamilton Industries, 97 TC 120 Hermes Consolidated, Inc ., KTC 1988-147 ; 88-I USTC 19220 Highland Farms, Inc ., 106 TC 237 Hitachi Sales Corp. of America, TC Memo 1992-504 Honeywell, Inc ., TC Memo 1992-453 Hospital Corp . of America, TC Memo 1996-105 IT&S of Iowa, Inc ., 97 TC 496 J .E . Seagram Corporation ., 104 TC 75 Kohler Co . v . US, KTC 1997-414; 97-2USTC 150,673 Kohler Co . v . US, KTC 1995-545 ; 95-2 USTC 150,600 Kraettli, TC Memo 1988-413 Kraft, Inc . v . US, KTC 1994-708 ; 94-1 USTC 150,080 Kroger Co., TC Memo 1997-2 continued overread

210

G. E . WHITTENBURG, W. A . RAABE AND J . R . LACKRITZ

Table] .

Continued .

Lacrosse Footwear, Inc . v. US, KTC 1997-39 ; 97-1 USTC 150,439 Lacrosse Footwear, Inc . v. US, KTC 1998-198 ; 98-I USTC 150,436 Merit Life Insurance Co . v . Commissioner, KTC 1988-95 ; 88-2 USTC 19457 Merkel, 109 TC 463 Mulholland v . US, KTC 1993-417 ; 93-1 USTC 150,286 Knoll Cellar, TC Memo 1994-396 Oglethorpe Power Corporation ., TC Memo 1990-505 Ohio Periodical Distributors, Inc., TC Memo 1995-496 Pacific Enterprises, 101 TC I Pierce, TC Memo 1997-411 Prabel v . Commissioner, KTC 1989-72 ; 89-2 USTC 19488 RLC Industries Co ., 98 TC 457 Robertson, TC Memo 1990-275 Seagate Technology, Inc ., 102 TC 149 Sears Imported Autos, Inc., TC Memo 1992-307 Sears, Roebuck & Co ., 96 TC 61 Snap-Drape, Inc . v . Commissioner, KTC 1996-486 ; 96-2 USTC 150,564 Straight, TC Memo 1997-569 Swanson, 106 TC 76 Texas Instruments, Inc ., TC Memo 1992-306 Thomas Nelson, Inc, v . US, KTC 1988-189 ; 88-1 USTC 19339 Thomas, 92 TC 206 Transamerica Corp . v . US, KTC 1993-349 ; 93-2 USTC 950,388 Travelers Insurance Co . v . US, KTC 1996-61 ; 96-1 USTC 150,231 TSI Inc. v. US, KTC 1991-92; 91-2 USTC 150,521 UFE, Inc ., 92 TC 1314 United Hardware Distributing Co ., KTC 1988-203 ; 88-2 USTC 19444 Wal-Mart Stores, Inc ., TC Memo 1997-1 Note : "KTC" stands for Kleinrocks Tax Cases

Occurrences by Court Table 2 presents the number of court decisions in which GAAP issues were raised by the court . 12 The Tax Court is by far the federal court that most often addressed GAAP/tax issues . Sixty-nine percent of the cases in this study were litigated in the Tax Court - there were 22 Tax Court Regular decisions and 28 Tax Court Memo decisions in the period under review ." The GAAP/tax issue was addressed 22 times (31%) in the other non-Tax Court Federal Courts : eight times in the District Court, nine times in the U .S . Court of Federal Claims, and five times in a Court of Appeals . The GAAP/tax issue was not addressed in any U .S . Supreme Court case during the period addressed by this study . 14


Table 2,

GAAP/Tax Occurrences by Court .

Court

Occurrence

Tax Court - Regular decision Tax Court - Memorandum decision Court of Federal Claims District Court Court of Appeals

22 28 9 8 5

Totals

72 (100%)

Table 3 .

(30 .5%) (39%) (12 .5%) (11%) (7%)

GAAP/Tax Occurrences by Issue .

Disputed Tax Issue Inventory Matters Operating Financial Issues Miscellaneous Issues

211

Occurrence

Issues

22

Totals

23 (32%) (31%) 16 (22%) I I (15%) 72 (100%)

Issues Raised

We grouped the disputed tax issues in the identified cases into four categories . The frequency with which the courts addressed these issues in a GAAP/tax context is summarized in Table 3 . Inventory Matters - selection and application of specific methods Operating Issues - interest expense, depreciation, methods of accounting, income inclusion/exclusion Financial Issues - acquisitions, net operating losses, book/tax differences, allocations of purchase price, treatment of debt Miscellaneous Issues - matters that did not fit the other categories, such as foreign currency translations, profit sharing plans, earnings and profits computations, and insurance applications . As Table 3 indicates, inventory issues represent the topical area in which GAAP is most often raised in tax litigation ." The second most frequently encountered category was Operating Issues, followed by Financial and Miscellaneous Issues .

212

G. E . WHITFENBURG, W . A. RAABE AND J. R . LACKRITZ

Table 4 .

GAAP/Tax Argument Used in Court's Decision .

Disputed Tax Issue

Used in the Decision

Inventory Matters Operating Issues Financial Issues Miscellaneous Issue

22/23 19/22 9/16 9/11

(96%) (86%) (56%) (82%)

Overall

59/72 (82%)

Issues Used by the Court Table 4 indicates the frequency with which the courts directly relied on an analysis of GAAP in reaching a published decision, when the GAAP issue was raised by one of the parties to the dispute. The courts adopted a GAAP argument 80% or more of the time in three of the categories : Inventory Matters, Operating Issues, and Miscellaneous Issues . The courts seem less willing to follow the GAAP defense as raised in disputes involving Financial Issues, although the GAAP argument still apparently swayed the court in 56% of those cases .

Does a GAAP Argument Help the Taxpayer or the IRS? Table 5 indicates the percentage of "wins" for the taxpayer and the IRS, for cases in which a GAAP issue was ruled on by the court . The only categories where the taxpayer prevailed at least half of the time were in the Financial and Operating areas . In the other categories, Inventory and Miscellaneous, the IRS position dominated the taxpayer's position, when a GAAP issue was introduced into the court proceedings .

Table 5 . Disputed Tax Issue

Winner by Issue Disputed . Taxpayer Prevailed

IRS Prevailed

Inventory Matters Operating Issues Financial Issues Miscellaneous Issues

9/23 11/22 9/16 3/11

(39%) (50%) (56%) (27%)

14/23 (61%) 11/22 (50%) 7/16(44%) 8/11 (73%)

Overall

32/72 (44%)

40(72(56%)


213

Nominal Measure of Associations The Chi-Squared (X 2) test for independence, a nominal level measure of statistical relationships, was applied to the data gathered in the study . The test is designed to determine whether the distribution of outcomes for one variable are consistent across the groups of a second variable . For this test, the null hypothesis is that the distribution is consistent across the groups of the second variable (i .e . independence is presumed) . If the test is rejected, then a significant relationship (i.e dependence) exists between the two variables . Thus, the chances of occurrence of one variable change as one moves across the categories of the second variable . The results are presented in Tables 6A and 6B . Three of the tabulations produced a statistically significant relationship . The strongest relationship was

Table 6A .

Variables Tested Issue by Used

Issue by Court Used by Winner Winner by Court Used by Court Winner by Issue

Table 6B .

Chi-Square Tabulations .

Relationship Tested Is the type of issue related to whether the GAAP defense was used? Is the type of issue related to the court writing the decision? Did the use of a GAAP defense affect the chances of winning? Did the chance of winning vary across courts? Did the use of the GAAP defense vary across courts? Did the chance of winning vary by type of issue?

Chi-Square

Degrees of Freedom

10 .244

P

0.017

15 .131

6

0.019

3 .498

I

0.061

3 .302

2

0.192

2 .283

2

0.319

2.813

3

0 .421

Detail of Chi-Square Tabulations - Used by Winner . Winner

Used

Government

Taxpayer

Totals

No - Actual Expected

4 .00 7 .04

9 .00 5 .96

13 .00

Yes - Actual Expected

35 .00 31 .96

24.00 27 .04

59 .00

214

G . E . WHII IENBURG, W. A . RAABE AND J . R . LACKRITZ

found for Issue by Used, i .e. if either party introduced to the court a GAAP issue, the court admitted on record to using the GAAP argument in arriving at its decision . The relationship was especially strong in the Operating, Inventory, and Miscellaneous categories, as would be expected from the results in Table 4 . The Issue by Court tabulation also evidenced a statistically significant relationship . The Tax Court used the GAAP issue more than the other pertinent federal courts . This might be expected, as the Tax Court judges are technical specialists in the tax law, while the judges in the other courts tend not to be so technically conversant with tax provisions . GAAP issues were used more often in Tax Court Regular decisions, and the other three categories of tax issues (Operations, Inventory, and Miscellaneous) were used more often in the Memorandum decisions . A moderately significant statistical relationship was found in the Used by Winner tabulation ; specifically, if a GAAP issue was used in the court's decision, it tended to help the IRS, not the taxpayer. Data supporting this observation are summarized in Table 6B . This relationship was not necessarily expected . Most tax professionals likely would anticipate that if a court used GAAP, it would benefit the taxpayer - the taxpayer after all deliberated before adopting the accounting method, and then adopted the method due to its projected benefit . The lack of significance in three of the pairwise relationships further prompts some interesting conclusions . For instance, the chances of winning the case appear to be not significantly related to the issue raised (the Winner by Issue tabulation) . Further, the chance of winning does not initially appear to vary by court type (Winner by Court) . If such a relationship did exist, then a strategic decision in choosing one's trial level court could occur . Logit Model To develop a prediction model for the win/lose outcome, it is common to use either Discriminant Analysis or Logit Regression . Discriminant analysis is a classification procedure which identifies the weighting of the individual factors (in this work, the factors might be used, court, or issue) on the final outcome (win) . Discriminant analysis produces an equation in which the outcome is predicted as a function of the specific factors and each of their perceived effects on the win/loss outcome . However, discriminant analysis is a binary procedure, which forces classification into one of two outcomes, with little or no middle ground . Thus, in the current context, unless the model is 90-95% accurate in making its predictions, it may have little or no value in constructing a strategy for carrying out one's tax dispute in the judicial system .


215

In contrast, Logit Regression creates a probability structure for the desired outcome of winning the case, according to the effects of the individual factors . Thus, for any case, the procedure forecasts a probability of winning (from 0 to 1) and gives the tax professional a more reasonable assessment of the chances of prevailing in the case . The standard form of the Logit Regression procedure is : In(p/I-p)=a+b l x l +6 2x2 + . . .+b k x k

(I)

where each xl is the numerical value of one of the predictor variables [e .g ., was GAAP used (I = yes, 0 = no), type of court, type of issue], and the b, terms are the relative weights of the regression coefficients to be used for predictive purposes . Exponentiating the predicted value, then solving for p, gives a prediction of the likelihood of winning a specific case . The application of the logit regression is summarized in Table 7, examining the predictive value of whether a GAAP argument was used by the court in crafting its decision, and also of the publication decision made by the Tax Court Chief Judge [i .e. regular ("law issue") or memo ("fact issue") decision] . We see that the Tax Court Regular v Memo Decision variable was not significant in the final model, nor were any interaction combinations of the predictor variables . Therefore, our predictive model is : In(p/l -p) = 0 .1781-1.1833 (if GAAP used) + 1 .0858 (if T . C . Regular) + 0 .7406(if TC Memo)

(2)

Thus, to predict the outcome of a case in which GAAP is cited in a Tax Court regular opinion : ln(p/l-p)=0 .1781-1 .1833 (1) + 1 .0858 (I)+0 .7406 (0) = 0 .0806 and

(full -p) = eo-o806 = 1 .0830, and p = 0 .5201

Table 7 . Predictor Constant

Logit Regression . Coefficient

Standard Deviation

0.1781

0 .7103

-1 .1833

0 .6820

Tax Court Regular

1 .0858

0 .6483

Tax Court Memorandum

0 .7406

0 .6139

Used

Log-Likelihood = -46.364 Test that all slopes are zero : G=6 .584, Up =3, P-Value = 0 .086 (< a=0.10)

216

G . E. WHITTENBURG, W. A . RAABE AND J . R . LACKRITZ Table 8.

Logit Regression Predictors of Taxpayer Success in Court Decision . In Which Court?

GAAP Argument Used? Yes No Difference

Tax Court Regular

Tax Court Memorandum

Other than the Tax Court

0 .5201 0.7797 0 .2596

0 .4342 0 .7145 0 .2803

0 .2679 0 .5444 0 .2765

A complete table forecasting the results for the significant relationships is found in Table 8 . The introduction of a GAAP argument by the taxpayer reduces the chances of winning by about 27 percentage points . Regardless of whether GAAP is used, the taxpayer's best chance of winning an argument is clearly in the Tax Court (Regular or Memo) . Taxpayer expectations should be set properly in reviewing these data . Budget limitations on the Treasury Department lead to skewed litigation decisions by the Chief Counsel, such that the IRS tends to litigate only those cases that : (1) it expects to win and (2) it expects will make good legal precedent and discourage other taxpayers from trying similar tax return filing positions . Thus, the taxpayer position is upheld completely by the court only rarely . 16 Relative to these broad results, whether to use a GAAP defense become a more complex decision .

CONCLUSIONS AND IMPLICATIONS When a GAAP issue was raised in the context of tax litigation, the courts tended to use it in the published opinion instead of rejecting the argument outright (x2, p = 0.017) . In addition, when a GAAP issue was used, it generally tended to help the IRS prevail over a taxpayer (x2, p=0 .061) . However, a Logit Regression Model refined the outcome prediction by indicating when GAAP was used, it slightly helped the taxpayer prevail in Regular decisions of the Tax Court (52 .01% chance of winning), but that a GAAP defense was not helpful in other forums (43 .42% in Tax Court Memo decisions and 26 .79% in decisions outside the Tax Court) . These conclusions are strongly convincing, in that all occurrences of a GAAP defense during the prior decade were examined . Sometimes the results were contrary to the expectations of many tax professionals . For example, many tax


217

practitioners would probably not expect that using GAAP in tax litigation helped the IRS more than taxpayers . In tax planning, the results of this study can be used in two ways . Taxpayers can use the results of this study as a practical tool for selecting an accounting method for tax purposes that reduces future IRS challenges . Second, the results of this study can be used to evaluate the probability of a favorable court decision, with respect to a legal challenge by the IRS to a GAAP accounting treatment . This could aid taxpayers and tax advisors in making decisions regarding the contesting of a GAAP issue through the courts . The final observation relates to suggestions for further research . The Logit Regression Model utilized in this study has potential in other areas of tax (and legal) research . It provides a appropriate quantitative basis for evaluating areas that have previously been subjected to predominantly qualitative judgements .

NOTES 1 . 99 SO 773 (1979) . 2 . Publicly-traded entities are required to be audited, and thus they must follow GAAP to receive an unqualified opinion as to their financial statements . GAAP in the form of depreciation and inventory computations are required of businesses of substantial size under IRC §448 . 3 . Statement on Auditing Standards No . 69 defines GAAP as "a technical accounting term that encompasses the conventions, rules, and procedures necessary to define accounting practice at a particular time . It includes not only broad guidelines of general application, but also detailed practices and procedures . . . Those convention, rules and procedures provide a standard by which to measure financial presentation ." By strength of authority, five tiers of GAAP elements are identified in SAS 69, all of which are products of the Financial Accounting Standards Board or the American Institute of CPAs, and none of which are government pronouncements, including those of the IRS, the Treasury, and the SEC . 4 . Special weight is given to the consistency of application . See Reg . 1 .471-2(a) and (b), and 1 .446-1 . 5 . Dayton Hudson, TC Memo 1997-260, rev'd 153 F3d 660 (CA-8, 1998) . 6 . TC Memo 1997-1, affd 153 F3d 650 (CA-8, 1998) . 7 . Because audit standards commonly apply materiality thresholds in implementing required standards and tests, Gaffney and Franz [1997] maintain that materiality exceptions now also apply in the tax law . This conclusion is not the subject of the current work, but the present authors do not believe that the Wal-Mart decisions introduce any new materiality exceptions into the application of §446 . 8 . The Tog Shop Inc ., 721 FSupp 300 (DGa, 1989), aff'd 916 F2d 720 (CA-11, 1990) . 9 . At p . 523 . 10 . "A method of accounting which reflects the consistent application of generally accepted accounting principles in a particular trade or business in accordance with accepted conditions or practices in that trade or business will ordinarily be regarded as

218

G . E. WHITTENBURG, W. A . RAABE AND J . R . LACKRITZ

clearly reflecting income, provided all items of gross income and expense are treated consistently from year to year ." Reg. §1 .446-1(a)(2) . 11 . PLR 7920008 (2-12-79) . "If a taxpayer has chosen a method which is not clearly described in the regulations, then the District Director has the authority to recommend changes in the taxpayer's method . . . The burden of proving that the taxpayer's current method is appropriate rests with the taxpayer ." The inventory method commented on in this ruling is one that Carnes and Englebrecht [1992] characterize as approved by GAAP and the SEC, authorized by the regulations and taught as acceptable in IRS training manuals . Yet, the ruling found that the method did not clearly reflect income . 12. Taxpayers choose the trial-level court for a dispute, either the Tax Court (a forum that rules on tax issues only, it is chosen 95% of the time), the pertinent U .S . District Court (4%), or the Court of Federal Claims (1%) . Strategies for making this choice are discussed in Raabe et al . [2000], at pp . 128-139 . 13 . The difference between a Tax Court regular and memorandum opinion is merely a publication decision that is made by the court's Chief Judge . Presumably, a regular decision addresses a new interpretation or application of tax law (known as a "law issue"), while a memorandum decision applies a set of facts to existing law (a "fact issue") . Raabe et al . [2000], p . 130 . 14. The key Supreme Court case on point, Thor Power Tool, was issued in 1979 . 15 . A string of taxpayer judicial successes in this area led to Gaffney and Smith's [1998] comment that "Thor has become or is likely to become nothing more than a memory" in some situations . 16 . For the 1990 litigation year, a total taxpayer victory was attained in the (predecessor to) the Court of Federal Claims about 15% of the time, 23% in the pertinent District Court, and 3% in published Tax Court cases . Raabe et al . (2000], p . 407 .

REFERENCES Carnes, G ., & Englebrecht, T . (1992) . The IRS' Increasing Power with the Clear Reflection of Income Standard . The Tax Executive (November 1992), 456 . Gaffney, D ., & Franz, D. (1997) . Tax Court Introduces Financial Statement Materiality into the Tax Law . The Tru Adviser (December 1997), 750 . Gaffney, D ., & Smith, M. (1998). The Shrinking Influence of Thor Power Tool Co. ; How Manufacturers and Retailers Circumvent Supreme Court Ruling on Case Involving Inventory Write-down. Management Accounting (October 1998), 32 . Raabe, W . A ., Whittenburg, G. E ., & Sanders, D. L . (2000) . Federal Tax Research, 5e. Cincinnati : South-Western College Publishing .

Advances in Taxation, Volume 13

Advances in Taxation, Volume 15

Advances in Taxation, Volume 16

Advances in Taxation, Volume 17

Advances in Taxation, Volume 14

Advances in Taxation, Volume 18

Advances in Immunology Volume 13

Advances in Computers, Volume 13

Advances in Immunology Volume 13

Advances in Agronomy, Volume 13

Advances in Genetics Volume 13

Advances in Geophysics, Volume 13

Advances in Microbial Physiology, Volume 13

Advances in Chemical Engineering, Volume 13

ADVANCES IN HETEROCYCLIC CHEMISTRY V13, Volume 13

Advances in Heat Transfer, Volume 13

Advances in Pharmacology and Chemotherapy Volume 13

Advances in Mathematical Economics Volume 13

Advances in Environmental Research, Volume 13

Advances in applied mechanics. volume 13

Advances in Management Accounting, Volume 13 (Advances in Management Accounting)

Advances in Geosciences: Volume 13: Solid Earth (Se)

Advances in Electronics and Electron Physics, Volume 13

Advances in Industrial and Labor Relations, Volume 13

Advances in Catalysis and Related Subjects, Volume 13

Advances in Atomic and Molecular Physics, Volume 13

Experimental and Behavioral Economics, Volume 13 (Advances in Applied Microeconomics)

Topics in Stereochemistry, Volume 13

Dislocations in Solids, Volume 13

Methods in Microbiology: Volume 13

Advances in Planar Lipid Bilayers and Liposomes (Advances in Planar Lipid Bilayers & Liposomes, Volume 13)

Advances in Taxation, Volume 13

Advances in Taxation, Volume 15

Advances in Taxation, Volume 16

Advances in Taxation, Volume 17

Advances in Taxation, Volume 14

Advances in Taxation, Volume 18

Advances in Immunology Volume 13

Advances in Computers, Volume 13

Advances in Immunology Volume 13

Advances in Agronomy, Volume 13

Advances in Genetics Volume 13

Advances in Geophysics, Volume 13

Advances in Microbial Physiology, Volume 13

Advances in Chemical Engineering, Volume 13

ADVANCES IN HETEROCYCLIC CHEMISTRY V13, Volume 13

Advances in Heat Transfer, Volume 13

Advances in Pharmacology and Chemotherapy Volume 13

Advances in Mathematical Economics Volume 13

Advances in Environmental Research, Volume 13

Advances in applied mechanics. volume 13

Advances in Management Accounting, Volume 13 (Advances in Management Accounting)

Advances in Geosciences: Volume 13: Solid Earth (Se)

Advances in Electronics and Electron Physics, Volume 13

Advances in Industrial and Labor Relations, Volume 13

Advances in Catalysis and Related Subjects, Volume 13

Advances in Atomic and Molecular Physics, Volume 13

Experimental and Behavioral Economics, Volume 13 (Advances in Applied Microeconomics)

Topics in Stereochemistry, Volume 13

Dislocations in Solids, Volume 13

Methods in Microbiology: Volume 13

Advances in Planar Lipid Bilayers and Liposomes (Advances in Planar Lipid Bilayers & Liposomes, Volume 13)

Recommend Documents