Persuasive Communications: Tax Compliance Enforcement Strategies for Sole Proprietors*

Transcription

1 Persuasive Communications: Tax Compliance Enforcement Strategies for Sole Proprietors* 1. Introduction JOHN HASSELDINE, University of Nottingham PEGGY HITE, Indiana University SIMON JAMES, University of Exeter MARIKA TOUMI, University of Nottingham Tax systems such as those found in Canada, the United States, and the United Kingdom primarily rely on the notion of voluntary compliance by taxpayers, but, importantly, the threat of possible audit also serves to encourage voluntary tax compliance (Braithwaite 2003). Audits, however, are not without cost to tax administrations, and they may not be an optimal use of government resources for all taxpayers. 1 Given limited resources, the tax agency will ultimately be constrained by the extent of audit enforcement strategies, and, as a result, alternative strategies need to be considered. Prior behavioral research using hypothetical cases and taxpayer self-reports has reported mixed results regarding the success of using persuasive communications to encourage compliant tax reporting (Hasseldine and Kaplan 1992; Hite 1989, 1997; Violette 1989). Critiques of that research often focus on the lack of generalizability to real taxpayer data. Few behavioral studies have been able to use actual taxpayer data (Schwartz and Orleans 1967; McGraw and Scholz 1988, 1991; Blumenthal, Christian, and Slemrod 2001). The results from these studies are inconsistent and could be a result of the differing research designs. The current study addresses some of those design issues by using brief, straightforward normative appeals (positive reasons why compliance is advantageous) and sanction appeals (appeals emphasizing the negative consequences of noncompliance) to test for behavioral responses on actual sole proprietors tax reports to the UK Inland * Accepted by Kenneth Klassen. We thank Kenneth Klassen (associate editor) and two anonymous reviewers for their detailed and helpful comments on earlier drafts. We also thank Marsha Blumenthal, Anne Ehinger, Darius Fatemi, Joel Slemrod, and David Paton for their comments, as well as participants at the 2005 IRS Research Conference. In addition, we appreciate the help we received from the UK Revenue and Customs in particular, John Connor, Nicola Fletcher, Ed Hagger, David Stephens, and Susannah Wright who executed the experiment described in this paper. The views expressed here, however, are not necessarily shared by the department or these individuals. We also gratefully acknowledge Economic and Social Research Council funding under the Future Governance program. Contemporary Accounting Research Vol. 24 No. 1 (Spring 2007) pp CAAA

2 172 Contemporary Accounting Research Revenue (IR), which is the equivalent of the Internal Revenue Service (IRS) in the United States. Taxpayers frequently engage the services of paid preparers, who play a key role in the tax environment with responsibilities to both their clients and to the state (Erard 1993; Nichols and Ellis Price 2004). Indeed, there now is a substantial literature on the role of paid preparers in the United States (Roberts 1998). One key aspect of this literature has been to examine compliance levels associated with the use of a paid preparer by analyzing data from IRS tax audits. The direction of the results, however, has been inconclusive (Hite and Hasseldine 2003; Christian, Gupta, Webber, and Willis 1994). Nonetheless, most agree that the paid preparer does affect the process, one way or the other. The current study is the first to distinguish between the effects of normative and sanction-based appeals on actual self-prepared and paid-preparer returns. In order to analyze the role of paid preparers and the effect of persuasive communications, we examine three research questions: (a) whether targeted written communications are effective strategies for a tax agency, (b) which of five strategies tend to be more effective, and (c) whether the effectiveness of these strategies varies by type of preparer. This paper makes its contribution through an innovative, controlled field experiment that analyzes how actual tax-reporting behavior is affected by exposure to one of five treatments: simple offer of assistance, citizenship appeal, threat of audit, threat of audit with possible penalties, and virtual guarantee of forthcoming audit once the return is filed. Sales and net profit data are analyzed for a nationwide sample of over 7,300 UK sole proprietors who, for two consecutive tax years, reported a turnover (sales) level just below a fixed threshold of 15,000. Taxpayers with turnover below the threshold qualify for a simplified format for tax reporting, allowing them to reduce their costs of compliance. Thus, temptation to reduce tax liability, as well as costs of compliance, makes the sole proprietors an intriguing and appropriate population to target for this research. Consequently, we focus on two dependent variables: (a) whether turnover exceeded the reporting threshold and (b) the monetary increase in net profit reported on the tax returns filed after the treatment letters were received. Our results show evidence of a significant overall treatment effect for both normative and sanction groups. Specifically, when we measure whether taxpayers reported increases in turnover (over the 15,000 threshold for simplified reporting) after receipt of the letter, there are significant differences for each treatment group on one or both of the dependent variables. We use ordinary least squares (OLS) and probit regressions to examine the impact of the treatment letters contingent on preparation mode, and we find that the communications are particularly effective for self-preparers when reporting turnover. We also find that the sanction letters are generally more effective than the normative citizenship letter for reported turnover. In addition, the sanction letters are more effective than the normative letter for change in net profit but only for the self-prepared returns, not for the paid-preparer returns. This study contributes to the literature by providing evidence that written communication strategies can be effective tools for improving actual taxpayer compliance. Prior research reported mixed results and did not demonstrate specific

3 Tax Compliance Enforcement Strategies for Sole Proprietors 173 effects on sales and net profit for sole proprietorships, nor did prior studies elaborate on the distinction between the effects on self-prepared and paid-preparer returns. This study also contributes to the literature by documenting results that are derived from the actual tax-reporting behavior of sole proprietors in the United Kingdom. Prior research has examined the total tax liability of taxpayers, which included income from all sources (investment, salary, and some self-employment). To the extent that many of the non-self-employment items have little opportunity for noncompliance, they create additional noise in the data, reducing the likelihood of finding statistical significance. In contrast, the current study focuses on the proprietorship s turnover and net profit, an area where opportunity for noncompliance is high. Overall, we suggest that communications from a national tax agency have the potential to be an effective tool for encouraging compliant tax reporting, despite prior research that has concluded otherwise. Another contribution of this study is that our method and findings are likely to be applicable in other non-tax areas of accounting. Although persuasive communications have been examined in some accounting-related studies (Kadous, Koonce, and Towry 2005), to date none of those studies has used persuasive communications to encourage compliant reporting behavior and to deter noncompliant reporting. Areas such as compliance with regulations and standards, management control systems, monitoring systems in the internal and external audit function, and other areas where ethics and accountability are important could be used to test normative and sanction frameworks. Although norms tend to be context-specific, there is growing evidence that social preferences such as honesty and fairness can have predictable effects in accounting contexts (e.g., Evans, Hannan, Krishnan, and Moser 2001; Hannan 2005). The remainder of this paper is organized as follows. Section 2 discusses prior literature. Section 3 describes the design and implementation of the field experiment and the rationale for the treatment groups. Section 4 presents the empirical tests for treatment and preparer effects, and section 5 concludes. 2. Prior research and development of hypotheses Auditing taxpayers may not always be cost-effective because the tax revenue ultimately collected could be less than the cost to the state of auditing the taxpayer. Moreover, excessive enforcement actions could be interpreted as unduly oppressive. Thus, as the UK Better Regulation Task Force (BRTF) recently noted, alternative approaches to securing compliance with regulations must be considered (BRTF 2003). Researchers have begun to explore persuasive communications as an alternative approach for encouraging compliant tax reporting. The research methods used have included personal interviews (Grasmick and Scott 1982; Schwartz and Orleans 1967), telephone surveys (Hite 1997), mail questionnaires (Hasseldine and Kaplan 1992; Hite 1989; Porcano and Price 1993; Wartick 1994), and experiments (Jackson and Jaouen 1989; Kaplan, Newberry, and Reckers 1997; McGraw and Scholz 1988, 1991; Roberts 1994; Violette 1989). Most of these studies have used hypothetical or self-reported behavior as dependent variables. In contrast, the current study uses actual taxpayer data that were reported to the national tax agency.

4 174 Contemporary Accounting Research The first significant study using actual taxpayer returns was Schwartz and Orleans Using field interviews, the researchers asked subjects several openended questions, which they categorized as either sanction questions or conscience questions. The questions indirectly asked each subject to reflect on his or her own attitudes and values. After comparing actual tax-return information for the treatment groups with tax-return information for a control group, they found that, compared with a control group, the conscience appeal manipulation had a significant effect, increasing both the amount of reported adjusted gross income (AGI) and the number of returns with an above-average increase in AGI. Results for the sanctionoriented interview questions did not significantly differ from those for the control group on the dependent variables for AGI. Although the results seem promising, the end result was based on a costly and time-consuming interview process. Nearly 20 years after the Schwartz and Orleans 1967 study, McGraw and Scholz (1988, 1991) also analyzed federal tax returns to compare the taxable income and liability of a control group with that of a group of taxpayers who had agreed, months before the tax return was due, to watch normative-based videos. This time lag as well as a self-selection bias to attend the video briefing may have negated any potential treatment effect, because no significant effects were documented. More recently, Blumenthal et al. (2001) mailed one of two normative appeals to separate groups of Minnesota state income taxpayers. The first was structured around a support valuable services theme, while the second was based on a join the compliant majority approach. After statistical comparison with a control group, they concluded that neither normative appeal had a significant impact on reported total taxable income or state tax liability. In a complementary study, Slemrod, Blumenthal, and Christian (2000) analyzed the effect of a letter to a sample of taxpayers stating that their 1994 income tax returns would be part of a study in which some of the returns would be closely examined by the Minnesota Department of Revenue. Their analysis of low- and middle-income taxpayers revealed significant increases, from 1993 to 1994, in total taxable income and state tax liability for high opportunity taxpayers defined as those filing a Schedule C or F in 1993 (indicating business or farm income) and who paid Minnesota estimated tax. In sum, written normative communications from a state tax agency had no apparent effect on overall tax liability to that state, but letters suggesting an increased risk of audit led to reports of higher state tax liability for low- and middle-income taxpayers whose taxable income included some self-employment income. An empirical question remains as to why appeals to conscience were effective in the Schwartz and Orleans 1967 study but not in the Blumenthal et al study. In addition, it is unclear why the sanctions were relatively effective in the Slemrod et al study but not in the Schwartz and Orleans 1967 study. Given the differences between the two studies, potential explanations include the following: changes in taxpayer receptivity over time, direct and indirect wording differences, use of mixed wage and sole proprietor data, communications from a national tax agency versus a state tax agency, use of total federal taxable income and state tax liability, and communications via an interview with repeated inferences versus a single

5 Tax Compliance Enforcement Strategies for Sole Proprietors 175 letter mailed to the taxpayer. Our study, which is nearly 40 years after the original Schwartz and Orleans study, controls for some of these differences. Given that Slemrod et al. (2000) documented sanction-based effects for lowand middle-income sole proprietors, our study targeted the low- to middle-income group of sole proprietors to test normative and sanction-based letters. Accordingly, we test the first hypothesis by using five brief letters that include varying degrees of encouragement. As shown in the appendix, these letters range from an offer of assistance to notification of an audit. HYPOTHESIS 1. Written communications from a national tax agency will positively affect the sales and net profit reporting behavior of small business proprietors. Returns prepared by a paid preparer Use of paid tax preparers has been increasing steadily (IRS 2001). Recent reports indicate that nearly 60 percent of U.S. returns are prepared by a paid tax preparer, and Christian, Gupta, and Lin (1993) documented that having a sole proprietorship significantly increased the likelihood of using a paid preparer. 2 Exactly how paid preparers affect taxpayer compliance is inconclusive. Some behavioral experiments using hypothetical cases conclude that paid preparers may be promoting aggressive reporting behavior (Ayres, Jackson, and Hite 1989; Klepper and Nagin 1989b; Johnson 1993), while others conclude that using a paid preparer results in more compliant reporting (Brody and Masselli 1996; Jackson and Milliron 1989; Cuccia 1994). Empirical evidence is also mixed. Christian et al. (1994) and Erard (1993) used audit data from the Taxpayer Compliance Measurement Program (TCMP) to examine the impact of paid-preparer usage on taxpayer compliance. The TCMP program was designed by the IRS to provide a stratified random sample of individual tax returns. The IRS conducted detailed audits on the returns for the purpose of measuring the extent of noncompliance. On the basis of that data base, Christian et al. (1994) reported that paid-preparer returns had significantly lower tax liabilities than self-prepared tax returns in a comparison group. Erard (1993) found that noncompliance on TCMP audits was greater on paid-preparer returns than on self-prepared returns. Moreover, he found that certified public accountant (CPA) noncompliance was greater than non-cpa preparer noncompliance. Using data from IRS office audits (those returns selected for audit because of a suspected misreporting issue), Hite and Hasseldine (2003) found that there were fewer IRS adjustments on audits with paid-preparer assistance than there were on audits of self-prepared returns. These seemingly contradictory studies may reflect the theme suggested by Klepper and Nagin 1989b that practitioners improve compliance on unambiguous items but exploit noncompliance on ambiguous ones. Alternatively, Kidder and McEwen (1989) assert that there are basically two types of tax practitioners, those who offer brokered compliance and those who offer brokered noncompliance. The implication is that taxpayers could seek out which type of tax assistance they would like to acquire.

6 176 Contemporary Accounting Research Prior research by Hite 2002 found that taxpayers using paid-preparer assistance during an IRS office audit had more negative attitudes toward the tax system than did those who did not use a paid preparer. A study by Stalans, Kinsey, and Smith 1991 also posited and found that involvement of a third party can exacerbate negative attributes. To the extent that paid preparers may affect tax-reporting behavior, they may also affect taxpayer attitudes toward the system. In those studies, the flow of the information about the tax agency was generally going from the paid preparer to the taxpayer-client. In the present study, the information from the tax agency is sent to the taxpayer, who is presumed to share the information with his or her paid preparer. This direction of information could be a signal to the taxpayer s representative that the taxpayer believes that the content of the IR letter is important to the client. Prior research has not documented differential effects of preparation mode on persuasive communications. Sanction letters, however, should be effective for selfprepared and paid-preparer returns because increased audit and penalty likelihoods would have repercussions for the taxpayer and the paid preparer (reputation effects as well as potential preparer penalties). Consequently, this study posits and tests the proposition that persuasive communications uniformly affect self-prepared and paid-preparer returns. That is, the present study posits a null hypothesis to test the effects of normative and sanction appeals by preparation mode: HYPOTHESIS 2. Written communications from a national tax agency will affect the extent of increased levels of reported sales and net profit for both self-prepared and paid-preparer returns. 3. Research design One experimental approach that has been used to encourage change is the use of persuasion-based laboratory experiments (Eagly and Chaiken 1993; Kadous, Koonce, and Towry 2005). Persuasive communications in the tax arena could rely on grounds of morality, educational outreach (e.g., Kaplan et al. 1997), or the use of formal/informal sanctions (e.g., Violette 1989). Regardless of the nature of the persuasive communication, characteristics of the targeted participants could affect their receptivity to the communication. Consequently, an important issue that researchers have had to face in this area is the difficulty of obtaining access to real taxpayers. Most prior studies have used vignettes and hypothetical reporting scenarios. The current study builds on prior research using experimental designs that analyzed actual taxpayer reports subsequent to exposure to manipulations for normative and sanction appeals (Schwartz and Orleans 1967; McGraw and Scholz 1988, 1991; Slemrod et al. 2000; Blumenthal et al. 2001). Opportunity for tax evasion by sole proprietors The tax administration in the United Kingdom has developed quite differently from those in Canada and the United States (James and Nobes 2003). For example, the United Kingdom relies extensively on withholding tax, which is so comprehensive that only about a third of taxpayers are required to file a return each year.

7 Tax Compliance Enforcement Strategies for Sole Proprietors 177 Moreover, for those who do have to file returns, self-assessment is a relatively new phenomenon. In 1990 the tax-reporting requirements of small businesses were simplified in order to ease the compliance costs. Previously, even the smallest business was expected to itemize all revenues and expenses in a formalized income statement. At present, individuals and partnerships trading with an annual turnover of less than 15,000 are required to provide only three figures with respect to their business: turnover (sales), allowable deductions, and net profit. It is this feature of the UK tax system that forms the basis of the experimental work reported here. Although the threshold has remained at 15,000 since 1992, the proportion of sole proprietors reporting turnover below that level has risen substantially. According to internal sources at the Inland Revenue, there were 3,100,310 self-employment returns, with 1,159,070 (37.4 percent) below the 15,000 threshold for By , there were only 2,790,800 self-employment returns, yet the number reporting below the turnover threshold and using the three-line account had grown to 1,161,420 (41.6 percent). Why did the percentage of self-employment returns with revenue below 15,000 increase? One obvious possibility is that large numbers of these taxpayers were understating turnover to reduce their tax bills and compliance costs. Reporting a low level of sales results in a reprieve from the otherwise complex filing requirements, because taxpayers need only provide three lines of summary accounting detail. Although the original purpose of the three-line, simplified filing requirement was to reduce compliance costs for very small businesses, the threshold has led to a behavioral response that may be indicative of tax evasion for these high opportunity taxpayers. Slemrod (2004) recently reviewed the literature on tax compliance costs and observed that while small businesses tend to exhibit high compliance costs relative to larger businesses, they are also among the least compliant, with particularly low voluntary compliance levels. Such a finding makes sole proprietors with sales of 14,000 to 15,000 an ideal UK sample for a controlled field experiment. Given that sole proprietors in the United Kingdom may be especially motivated to report turnover below the threshold level, the primary dependent variable in our study is the proportion of taxpayers who in 2001 report turnover in excess of the threshold. In addition, we examine another dependent variable magnitude of the monetary increase in net profit. Prior research has indirectly examined net profit because of its inclusion in their analysis of taxable income, the net of all types of income and allowable deductions (Schwartz and Orleans 1967; Slemrod et al. 2000). Moreover, by analyzing net profit, we are in effect analyzing a joint reporting decision. Klepper and Nagin (1989a) used TCMP data to conclude that the endogeneity of detection risk significantly affects the level of compliance. That is, taxpayers appear to allocate their noncompliance among a combination of line items so as to minimize the potential detection and penalty risks. Often they avoid extreme noncompliance on one item in order to avoid detection on another line item. This ex ante joint decision implies that net profit is the result of endogenous tax-reporting behavior.

8 178 Contemporary Accounting Research Rationale for experimental treatments Letter 1 Over the last decade, tax agencies have become more concerned about their customer service (IRS 2001; Braithwaite 2003). The same is true for the United Kingdom (Inland Revenue 2004), and one of its key underlying philosophies is to promote an enabling culture in the organization. 3 In effect, this means giving attention to helping their customers (still referred to as taxpayers hereafter) file their tax returns correctly. Letter 1 in the present study tests for positive effects from reminding taxpayers that assistance is available from the national tax agency. Hite (1989) found that taxpayer attitudes about the tax agency were significantly more positive after having received information about tax help that is available from the IRS. That work is supported by theoretical arguments on reciprocity, legitimacy, and procedural fairness, which Smith (1992) discusses, tests, and statistically documents. These prior studies, however, did not examine actual taxpayer data as the current study does. Accordingly, Letter 1 acknowledged that the taxpayer had reported turnover in the 14,000 to 15,000 bracket for the consecutive years of 1999 and In addition, the letter stated that if the turnover for 2001 rose above 15,000, the taxpayer would be required to file extra forms with the tax return. Next, the enabling treatment letter (see the appendix) offered assistance by stating: The Inland Revenue wants to make it easier for you to make a complete and accurate return. We are here to give you advice and support if you need it. Given that approximately two-thirds of sole proprietors in the United Kingdom use a paid preparer, these taxpayers do not need the IR assistance that was offered in the letter. The assumption is that their paid preparer will take care of filing a complete and accurate return on their behalf. Consequently, it was not expected that the enabling letter would affect behavioral change among this group. Letter 2 Although prior empirical evidence is inconclusive on the efficacy of normative appeals, Bardach (1989) notes that even small increases in tax compliance resulting from a conscience appeal may well net a tax gain running into the billions. This, coupled with an interest by the UK Blair government in the notion of citizenship, prompted the second treatment letter. The normative appeal letter combined elements of the two letters used by Blumenthal et al That is, although the treatment in this study was very brief, compared with the Blumenthal letter, it contained both a support valuable services element and a join the compliant majority element. The citizenship treatment (Letter 2, see appendix) contained the following paragraph: Most people in this country pay their proper taxes. But even small mistakes by a lot of people can add up to a lot of lost tax and, therefore, less money available for public spending on things like hospitals, schools and pensions.

9 Letters 3, 4, and 5 Tax Compliance Enforcement Strategies for Sole Proprietors 179 Prior theory and empirical results suggest that an increased audit threat will decrease evasion and if p equals one, any rational taxpayer will report his or her true income (emphasis in original, Slemrod et al. 2001, 458). 4 Accordingly, three further letters were designed to vary the audit manipulation received by the taxpayers. For exampe, Letter 3 was an increased audit treatment (see appendix) that stated: The Inland Revenue is substantially increasing the number of enquiries into the tax returns of people who have previously reported a turnover just below 15,000 in successive years. Your 2001 return may be one of those chosen for enquiry. 5 Letter 4 included a statement on penalties, which added, If we find that a return is incorrect, we may charge financial penalties as well as collect any unpaid tax. For Letter 5 (the preselected audit treatment), the last sentence in the paragraph from Letter 3 (quoted above) was replaced with the sentence: Your 2001 return has already been chosen for enquiry. This treatment was designed so that taxpayers would realize they were facing a certain audit, giving them a strong incentive to report their income and deductions accurately. 6 Prior research using hypothetical cases has reported contradictory results, presumably because of differing ways to describe the persuasive communications (Grasmick and Scott 1982; Jackson and Jaouen 1989; Violette 1989; Roberts 1994; Hite 1997). Moreover, the length of the communications also could have affected the results. For example, much of the results of persuasion tactics reported by Roberts 1994 as well as McGraw and Scholz 1988 was either mixed or nonsignificant in its effects on tax attitudes and behavioral intentions. Both studies used lengthy videos that included a variety of information. Subjects could have agreed with parts of the information and disagreed with other parts. This phenomenon is consistent with the research on behavioral survey techniques indicating that compound questions should not be asked (Sudman and Bradburn 1986). Responses by some subjects could be driven by the first part of the question while responses by others could be influenced by the latter part of the question. Compared with prior research, the wording in the five manipulations for this study is more concise. Selection of actual taxpayer sample As most studies correctly acknowledge, a limitation of using a laboratory or student samples is that the effects may not generalize from the experiment to the target setting. Typically, the concern is the lack of evidence demonstrating that tax attitudes and hypothetical intuitions map into actual tax-reporting behavior (Webley, Robben, Elffers, and Hessing 1991). Our study uses a stratified random sample of adult taxpayers who are sole proprietors. Personnel from the UK Inland Revenue selected the sample, issued the letters, and provided anonymous tax-return data to the authors. The initial sample selection identified the population of taxpayers who had reported turnover on their tax returns between 14,000 and 15,000 for the two

10 180 Contemporary Accounting Research consecutive years of 1999 and A random sample was selected from that population and then randomly assigned to one of six groups (five treatment groups and a control group). In addition, screening criteria were imposed after assignment to a group but before a letter was sent. First, a case was deleted if a completed 2001 return had already been received (very early filers). Second, a case was deleted if the taxpayer was deceased. Third, a case was deleted if the taxpayer was currently subject to an audit inquiry. Finally, a case was deleted if the taxpayer was subject to a foot-and-mouth exclusion. 7 This selection process resulted in 7,307 taxpayers spread across the control group and five treatment groups as follows: control 2,485; enabling 1,013; citizenship 1,040; increased audit 1,082; audit/penalties 1,019; and preselected audit 668. The use of sole proprietors is analogous to the highopportunity taxpayers (Schedule C for trade or business income and Schedule F for farming) used in the Minnesota state experiment (Blumenthal et al. 2001; Slemrod et al. 2000). In Slemrod et al there were 57 low-income/highopportunity taxpayers and 429 medium-income/high-opportunity taxpayers in their final sample. In Blumenthal et al the authors added a dummy variable to capture the presence or absence of Schedule C/F income. 4. Results Impact of treatments on turnover exceeding the threshold A unique feature underlying this study is the existence of the 15,000 threshold. Accordingly, a key dependent variable is the proportion of sole proprietors who increased their 2000 reported turnover (by definition less than the threshold) to report 2001 turnover exceeding the 15,000 cutoff for simplified reporting. This information is shown in panel A of Table 1. The baseline control group did not receive a treatment letter, and Table 1 indicates that only 39.7 percent of the control group self-report above the threshold. As outlined earlier, the enabling treatment (Letter 1) is not hypothesized to be effective for those taxpayers employing a paid preparer, because they are already receiving paid assistance and advice. The offer is simply not meaningful to those who have a paid preparer. The intention was to remind self-preparers that the tax agency is willing to provide advice and support. When hypothesized treatment effects are not symmetrical across cells, a conventional analysis of variance (ANOVA) for interactions effects does not provide an appropriate or powerful test (Buckless and Ravenscroft 1990). In this situation, a more powerful test is to use contrast coding to test the specific hypothesis. Consequently, results for the enabling letter, as presented in the tables that follow, specifically test only the cells for self-prepared returns. The results in Table 1, panel A show that Letter 1, the enabling letter, did not result in a significantly higher number of returns (34.0 percent) with turnover exceeding 15,000 than were made by the control group (29.9 percent). The chi-square statistic was (p 0.05). Hence, the enabling letter was not effective for increasing the probability that self-prepared returns would report turnover in excess of the threshold for simplified reporting.

11 Tax Compliance Enforcement Strategies for Sole Proprietors 181 Using the entire sample for Letters 2 through 5, chi-square tests showed significant differences (p 0.05) between the control group and each of the four hypothesized treatment groups. The respective percentages ( 2 -statistics) for those four groups were 44.8 percent ( ), 53.8 percent ( ), 56.4 percent ( ), and 53.7 percent ( ). Clearly, relative to the control TABLE 1 Treatment versus control group effects on turnover exceeding threshold Panel A: Chi-squares on turnover exceeding threshold Treatment group No [n] Self-prepared Yes [n] All [n] % Use paid preparer Control [1,648] [837] [2,485] Enabling (L1) * 70.7 [716] [297] [1,013] Citizenship (L2) [699] [341] [1,040] Increased audit (L3) [768] [314] [1,082] Audit/penalties (L4) [678] [341] [1,019] Preselected audit (L5) [465] [203] [668] Total n [4,974] [2,333] [7,307] 68.1 Panel B: Probit regressions on turnover exceeding threshold # Treatment group [n] 1 D ltr ** 2 D mode D ltr *D mode Wald Enabling (L1) [3,498] (0.95) ( 7.15) (0.60) Citizenship (L2) [3,525] (1.13) ( 7.15) (2.07) Increased audit (L3) [3,567] (4.89) ( 7.15) (2.58) Audit/penalties (L4) [3,504] (6.02) ( 7.15) (2.44) Preselected audit (L5) [3,153] (3.39) ( 7.15) (3.44) (The table is continued on the next page.)

12 182 Contemporary Accounting Research TABLE 1 (Continued) Notes: For panels A and B, the dichotomous dependent variable is coded 1 when reported turnover exceeded the 15,000 turnover threshold and 0 otherwise. * Significant at the 0.10 level for chi-square results in panel A cross-tabulations and Wald chi-square, probit regression results in panel B. Significant at the 0.05 level for chi-square results in panel A cross-tabulations and Wald chi-square, probit regression results in panel B. Significant at the 0.01 level for chi-square results in panel A cross-tabulations and Wald chi-square, probit regression results in panel B. The enabling letter (L1) is not hypothesized to affect reporting behavior for paidpreparer returns. For informational purposes, those results are presented here. # In panel B the probit regression coefficients are presented with the z-statistics in parentheses. ** Coded 1 for respective treatment letter and 0 for control group (no letter). Coded 1 for returns that were self prepared, coded 0 for paid-preparer returns. Interaction term represents respective treatment (D ltr ) times preparer mode (D mode ). group with 39.7 percent, each of the four letters had a significant impact on increasing reported sales beyond the threshold level. Turnover effects split by preparation mode Following the work of several authors (Stalans et al. 1991; Christian et al. 1994; Erard 1993; Hite and Hasseldine 2003), an important issue is whether the treatment effect is equally effective on self-prepared and paid-preparer returns. The results in panel A of Table 1 show that within the control group the percentage of tax returns reporting turnover greater than the threshold level significantly differed by preparer use ( , p 0.05). Only 29.9 percent of the self-preparers in the control group reported turnover levels exceeding the limit for simplified reporting. In contrast, 44.7 percent of the paid-preparer returns in the control group reported turnover levels that exceeded the threshold. The significant difference between self-prepared and paid-preparer returns in the control group reinforces the need to analyze the treatment effects by type of preparer, self or paid. To the extent that paid-preparer returns are inherently distinct from self-prepared returns, a separate analysis for each category mitigates the effects of unmeasured, extraneous variables (such as complexity and motivation for using a preparer). Those results are shown in panel A. Among the self-prepared returns, each treatment group for Letters 2 through 5 resulted in a significantly higher percentage of returns with turnover exceeding 15,000 than in the control group (29.9 percent). The respective percentages ( 2 -statistics) for the treatment groups were 39.9 percent ( ), 50.0 percent ( ), 52.5 percent ( ), and 54.2 percent

13 Tax Compliance Enforcement Strategies for Sole Proprietors 183 ( ). The implication is that each letter led to a higher likelihood of reporting turnover exceeding the simplified reporting level. Among the paid-preparer returns, all of the sanction treatments (Letters 3 through 5) were significant. The sanction letters resulted in significantly higher proportions of returns with turnover above the threshold. Respectively, these percentages ( 2 -statistics) were 55.3 percent ( ), 58.4 percent ( ), and 53.5 percent ( ), compared with 44.7 percent for the control group. The results for the citizenship letter (47.2 percent) did not significantly differ from those for the control group. The higher percentage of returns with increased turnover for paid-preparer returns does not necessarily signal more compliant reporting relative to the selfprepared returns. It is possible that a true increase in revenue led those taxpayers to seek out professional tax assistance. As noted earlier, the data base provided to the researchers did not include a measure for change from a self-prepared to paidpreparer return. However, if the treatment letters led to a greater likelihood of using a paid preparer, then preparer use in the treatment groups would be higher than preparer use in the control group. Sample sizes for each cell are shown in Table 1, panel A. In the control group, 66.3 percent of the returns were prepared by a paid preparer. For Letters 1 through 5, the percentages for paid-preparer usage were 70.7 percent, 67.2 percent, 71.0 percent, 66.5 percent, and 69.6 percent, respectively. Thus, use of a paid preparer was slightly higher in the treatment groups but only statistically significant (p 0.05) for the enabling letter (Letter 1, ) and increased audit probability (Letter 3, ). The variable of interest for the present study is not whether the treatments increased preparer use but whether the relative number of returns reporting turnover greater than 15,000 is significantly higher for each treatment group than for the respective control group. Since paid-preparer returns may differ for extraneous reasons that are not captured in the available data, the univariate tests in panel A provide statistical comparisons within each category of preparer. Regression analysis controlling for preparer mode and treatment effects While the preceding univariate analysis reveals effects within each preparation mode, a probit regression model is used on the full sample to examine whether preparation mode differentially affects the four treatment effects. We employ the following multiple regression model: Y 0 1 D ltr 2 D mode 3 D ltr *D mode (1), where Y is a dichotomous variable coded as 1 for reporting turnover greater than the 15,000 threshold and 0 otherwise; D ltr is a dummy variable coded 0 for those taxpayers in the control group and 1 for those receiving a letter (a separate regression for each letter);

14 184 Contemporary Accounting Research D mode is a dummy variable coded 0 for those taxpayers using a paid preparer and 1 for those self-preparing; and D ltr *D mode is an interaction term coded 1 for those taxpayers who received a letter and were self-preparing and 0 otherwise. In panel B of Table 1 we report coefficients for five separate probit regressions, where 1 D ltr represents the overall impact of the relevant letter, 2 D mode represents the overall impact of preparation mode, and 3 D ltr *D mode represents the differential impact of preparation mode on the treatment effect of that letter. So, a significantly positive coefficient would lend support to the hypothesis that the communication is more effective among self-prepared returns. Across all treatments, preparation mode was highly significant, indicating that a higher percentage of returns reported turnover in excess of the threshold when they were paid-preparer returns. The interaction terms were significant for Letters 2 through 5, and the positive coefficient in all four regressions indicates that each treatment was significantly more effective for self-preparers than it was for those employing a paid preparer. The citizenship letter was not significant as a main effect, suggesting that this communication was effective only for self-preparers. This is supported by the chi-square results reported in panel A. In contrast, for the sanction groups, the main effect, although strongly significant, was statistically more effective for the self-preparers. Monetary increase in net profit from 2000 to 2001 Panel A of Table 2 reports mean differences in amount of change in net profit from 2000 to The mean increase for the control group was 206. Analysis using one-way ANOVAs shows significantly higher increases in the changes in net profit for Letters 2 through 5 with respective mean changes (F-statistics) of 382 (2.392), 438 (4.303), 731 (19.431), and 976 (31.156). The sanction letters were significant at p 0.05, while the citizenship letter was significant at p In short, the sanction letters were effective for increasing reported net profits, and subjects receiving the normative citizenship letter also reported net profit increases that were higher than those in the control group. The implication is that, overall, the sanction letters and the citizenship letter significantly affected the amount of increased net profit reporting. In respect of self-preparers, separate ANOVAs show significant increases (p 0.05) in mean change for the sanction groups ( 319, F-test 1.987; 690, F-test ; and 1,569, F-test ) compared with the control group ( 35). With regard to the magnitude of the change in net profit, Letters 1 and 2 did not differ from the mean change for the control group. Only the sanction letters resulted in significantly higher changes in net profit. This result implies that a sanction-based letter could be effective, but (as noted earlier) its long-term effectiveness would require significant resources to fund the subsequent audits. For those employing a paid preparer, all four treatments had a significant effect on the size of the monetary increase (panel A, Table 2). The mean increase in

15 Tax Compliance Enforcement Strategies for Sole Proprietors 185 net profit for the control group was 293, while the respective means for the treatments were 581 (F-test 4.417), 487 (F-test 2.073), 751 (F-test 9.584), and 718 (F-test 6.785). Note that the enabling letter (Letter 1) was not significant, nor was it hypothesized to be, because those using a paid preparer no longer needed IR assistance. TABLE 2 Treatment effects on monetary changes in net profit from 2000 to 2001 Panel A: ANOVAs on change in net profit by group (mean changes in ) Treatment group No [n] Self-prepared Yes [n] All [n] Control [1,648] [837] [2,485] Enabling (L1) # [716] [297] [1,013] Citizenship (L2) * [699] [341] [1,040] Increased audit (L3) 487 * 319 * 438 [768] [314] [1,082] Audit/penalties (L4) [678] [341] [1,019] Preselected audit (L5) 718 1, [465] [203] [668] Panel B: OLS regressions on monetary change in net profit ** Treatment group [n] 1 D ltr 2 D mode D ltr *D mode F Enabling (L1) # [3,498] (0.10) ( 1.99) (0.35) Citizenship (L2) [3,525] (2.08) ( 1.98) ( 1.45) Increased audit (L3) [3,567] (1.45) ( 1.98) (0.37) Audit/penalties (L4) * [3,504] (3.14) ( 1.90) (0.78) Preselected audit (L5) * [3,153] (2.56) ( 1.93) (3.73) (The table is continued on the next page.)

16 186 Contemporary Accounting Research TABLE 2 (Continued) Notes: * Significant at the 0.10 level for ANOVA results in panel A and OLS regression results in panel B. Significant at the 0.05 level for ANOVA results in panel A and OLS regression results in panel B. Significant at the 0.01 level for ANOVA results in panel A and OLS regression results in panel B. Panel A presents the ANOVA results for five separate tests per preparer mode (paid, self, and all) to compare each treatment with its control group. In the first column of data, the amount of increased net profit is reported for paid-preparer returns by treatment group. In the second column, it is the amount for self-prepared returns. In the third column, it is the amount for all subjects combined. # The enabling letter is not hypothesized to affect reporting behavior for paidpreparer returns. For informational purposes, those results are presented here. ** In panel B the standardized beta coefficients are presented with the t-statistics in parentheses. Regression analysis controlling for preparer mode and treatment effects The preceding univariate reveals effects by preparer mode. In order to determine whether preparation mode differentially affects the four treatment effects, we employ the following multiple regression model: Y 0 1 D ltr 2 D mode 3 D ltr *D mode (2), where Y is the change in net profit for each taxpayer. As with the preceding analyses, panel B of Table 2 shows a consistent preparer effect. However, the only significant interaction effect is for Letter 5, indicating that the preselected audit letter was significantly stronger for the self-prepared returns than for the paid-preparer ones. The regressions also reveal main effects for Letters 2, 4, and 5. Letter 3 for increased audit threat did not have a significant overall effect on the amount of monetary increase in net profit from 2000 to Thus, the mention of an increased audit rate did affect the percentage of returns reporting turnover above the threshold, but it did not have as strong an effect on increasing net profit reports. 8 Comparison between treatment groups Normative (citizenship) versus sanction groups No specific ex ante predictions on whether and how the treatment groups would differ from each other were hypothesized. Such comparisons, however, have practical significance. When we compared differences between the normative citizenship group and an average of the three sanction groups for the entire sample, our

17 Tax Compliance Enforcement Strategies for Sole Proprietors 187 regression analysis (Table 3, panel B) showed that sanction letters were significantly more effective on the dichotomous variable for reporting turnover in excess of the threshold. In contrast, the effectiveness of the citizenship and sanction letters did not statistically differ on amount of increase in net profit for the paid-preparer group. However, per panel A, the sanction letters led to significantly higher monetary increases in net profit than did the citizenship letter for the self-prepared returns. In sum, while nearly every letter was associated with more returns reporting turnover in excess of the threshold, and higher amounts of net profit, the sanction letters did not consistently result in higher reports than the citizenship letter for the paid-preparer returns. Nonetheless, for the self-prepared group, the sanction letters tended to be significantly more effective than the citizenship letter. TABLE 3 Comparison of normative (L2) versus sanctions appeals (L3 5) Panel A: Univariate comparisons on turnover and change in net profit Dependent variable [n] Self-prepared Paid-preparer All L2 [341] L3 5 [858] L2 [699] Turnover 15, Change in net profit Panel B: Regressions on turnover and change in net profit # L3 5 [1,911] L2 [1,040] L3 5 [2,769] Dependent variable [n] 1 D ltr 2 D mode D ltr *D mode Wald/F Turnover 15, [3,809] (3.98) ( 2.23) (0.87) Change in net profit [3,809] (0.40) ( 2.92) (2.99) Notes: * Significant at the 0.10 level. Significant at the 0.05 level. Significant at the 0.01 level. In panel A, turnover (exceeding the 15,000 reporting threshold) is coded 0 for no and 1 for yes. Thus, results from chi-square cross-tabulations are reported. Change in net profit is a linear variable. Thus, the ANOVA F-statistics are computed. Comparisons are made between sanction and normative appeals within each category of preparation mode. # Because turnover is a dichotomous variable, the Wald chi-square statistic and z-statistics (in parentheses) from a probit regression are reported. For change in net profit, the F-statistic and t-statistics (in parentheses) from an OLS regression are presented.

18 188 Contemporary Accounting Research Comparison between sanction groups A series of post hoc Bonferroni multiple comparison tests on between-groups differences for the sanction letters were computed. For turnover exceeding the threshold, there were no significant differences among the three sanction letters for the total sample. This result suggests that there may not have been any more room for improvement. In other words, it is possible that the true percentage of returns having turnover that exceeds 15,000 is approximately 50 percent. This assertion is based on the fact that Letter 5 guaranteed a forthcoming audit, which should have led to a high compliance rate. For example, 54.2 percent of the self-preparers receiving Letter 5 reported an increase in turnover exceeding the threshold; 53.5 percent of the corresponding paid-preparer group reported turnover above the threshold level. Even when segregated by preparation mode, the relative effectiveness of the sanction letters did not significantly differ for the self-preparers ( percent) or the paid preparers ( percent). Monetary change in net profit between sanction groups On monetary increase in net profit for the total sample, the increased audit letter resulted in significantly lower reports ( 438) than the audit/penalties letter ( 731, F-test 4.457) and the preselected audit letter ( 976, F-test ). For taxpayers using a paid preparer, there were no significant differences among the three sanction groups. In contrast, for taxpayers who self-prepared, the differences were highly significant between the effects of the increased audit letter and the preselected audit letter (Letters 3 and 5). Taxpayers who were preselected for audit reported significantly higher increases in net profit than taxpayers who simply received an increased audit letter ( 1,569 versus 319; F-test ). The net profit reported by the preselected audit group was also significantly higher than that reported by the audit/penalties group ( 690, F-test 9.091). The implication is that greater compliance is induced when a self-preparer is preselected for an audit, but realistically governments cannot afford to audit a high percentage of returns. Consequently, it could be argued that the most economical solution among the three sanction choices is a combined audit and penalty reminder because Letter 4 led to a marginally higher net profit than did the increased audit letter ( 690 versus 319, F-test 2.724, p 0.10). 5. Discussion and concluding comments This paper presents the results of a national controlled field experiment examining the effects of alternative written communications on actual tax-reporting behavior while documenting the independent effects on self-prepared and paid-preparer returns. Apart from the use of a field experiment, we extend the prior literature by testing normative and sanction appeals. Five treatment conditions are examined: an offer of assistance (Letter 1), a citizenship appeal approach (Letter 2), and three sanction-based approaches (Letters 3 through 5). Increasing the compliance of sole proprietors is no easy task, and there is a dearth of large-scale controlled field experiments in this area (Slemrod et al. 2000).