United States General Accounting Office GAO Report to the Chairman, Committee on Ways and Means; and the Chairman, Subcommittee on Oversight, Committee on Ways and Means, House of Representatives February 1999 TAX ADMINISTRATION IRS’ Return Selection Process GAO/GGD-99-30 GAO United States General Accounting Office Washington, D.C. 20548 General Government Division B-279036 February 22, 1999 The Honorable Bill Archer Chairman, Committee on Ways and Means House of Representatives The Honorable Amo Houghton Chairman, Subcommittee on Oversight Committee on Ways and Means House of Representatives For 1992, according to the most recent estimate of the Internal Revenue Service (IRS), individual taxpayers, at the time they submitted their tax returns, did not pay about $100 billion in income tax liabilities. An audit is one of the enforcement tools IRS uses to try to reduce this figure. Recently, Congress has voiced concerns about the methods IRS uses to select returns for audit and the techniques used to conduct audits. These concerns contributed to the passage of the IRS Restructuring and Reform Act of 1998 (Public Law 105-206, July 22, 1998). Among other things, this act limits IRS’ ability to use certain audit techniques and requires that IRS include in information it sends to taxpayers an explanation of the reasons why it selects returns for audit. IRS has many sources from which to select tax returns for audit. IRS’ intent is to select returns with audit potential—returns for which an audit is most likely to find errors and recommend changes to the reported tax. One source is the discrimnant function (DIF), an automated system for scoring individual tax returns according to their audit potential. IRS’ policy also allows returns to be selected through non-DIF sources (e.g., referrals from other federal and state government agencies) if these returns can be shown to have a greater audit potential than a DIF-selected return. In an April 1996 letter, the committee asked that, among other things, we explore how effectively IRS selects individual income tax returns for 1 audit. However, IRS does not have information on all aspects of audit effectiveness. Consequently, we agreed, on the basis of discussions with your office, to focus on audits in which auditors in IRS district offices hold 1 This is the third in a series of reports you requested on IRS’ audit process. The other two were: Tax Administration: More Criteria Needed on IRS’ Use of Financial Status Audit Techniques, (GAO/GGD- 98-38, December 30, 1997); and IRS Audits: Workpapers Lack Documentation of Supervisory Review, (GAO/GGD-98-98, April 15, 1998). Page 1 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 face-to-face meetings with individual taxpayers to review their books and 2 records. For these books and records audits, we agreed to (1) determine the extent to which IRS district offices have used various sources to select these individual returns for audit; and (2) compare the results of audits selected using these sources in terms of the rate at which audits recommended no-change to the tax reported (no-change audits), amount of additional taxes recommended per return audited, and rates at which 3 IRS assessed and collected such recommended taxes after the audit. For our analyses, we examined books and records audits of returns received 4 by IRS in 1992, 1993, and 1994 that were closed within a 4-year period. Of the 1.1 million closed books and records audits of returns received in Results in Brief 1992, 1993, and 1994, our analysis showed that IRS selected 59 percent of the returns using its DIF source. The other 41 percent were selected using non-DIF sources. When we compared the results from DIF and non-DIF audits of returns received in 1992, 1993, and 1994, the non-DIF audits generally resulted in lower no-change rates and higher recommended additional taxes than DIF audits. These results are consistent with IRS’ policy to use non-DIF sources if the audit potential appears to be higher than it would be from a DIF audit. In contrast, we estimated that IRS collected a greater proportion of the additional taxes recommended in DIF audits than for non-DIF audits, based on a sample of returns received in 1992. An estimated 57 percent of the recommended additional taxes were collected for DIF audits versus 35 percent for non-DIF. Several IRS operations affect collections and we were unable to determine from IRS’ data which of these caused the non-DIF collection rate to be lower. Caution is needed if one uses the three results we analyzed to compare the effectiveness of DIF and non-DIF sources. The no-change rate, the recommended additional tax amounts, and the collection rate on these recommended amounts do not present a complete picture of audit 2 As discussed in the Scope and Methodology section and appendix I of this report, we excluded some audits because they are not typical books and records audits. 3 In this report, when we refer to the recommended additional tax, this includes refunds and penalty amounts, if applicable. IRS combines tax and penalties into one figure reported as “audit results.” This figure cannot be separated in IRS’ data files. 4 Between 2 and 5 percent of the returns selected for audit remained open at the end of our study period. See the scope and methodology section for a discussion of these open audits. Page 2 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 effectiveness. For example, data are not readily available on how audits affect voluntary compliance and taxpayer burden. Nor are data readily available on how other factors, such as the quality of the audits, affected the results across the selection sources. IRS conducts various compliance checks to determine whether taxpayers Background reported the correct tax liability on their returns. Initially, taxpayers may be contacted as IRS processes their filed returns if IRS identifies a math error or missing information on the return. IRS also uses computers to match information on the tax return with information obtained from third parties, such as employers and financial institutions, that report payments such as salaries and interest paid to individuals. During fiscal year 1996, IRS used computer matching to identify about 3.2 million taxpayers who may have underreported income on returns or not filed required returns. Additionally, IRS audited about 2.1 million tax returns to determine whether taxpayers reported the correct tax liability. Audits are performed at IRS’ 10 service centers and 33 district offices. Service centers and district offices perform different types of audits on the basis of complexity of the audit, training and pay grade of auditors, and audit techniques used. Service centers conduct correspondence audits, generally of a single issue such as earned income credit claims. These audits rely on correspondence to ask taxpayers to provide information about an item on the tax return. Face-to-face audits with taxpayers to review their books and records occur in IRS’ 33 district offices. District office audits are more complex, with multiple issues and often more than 1 tax year being audited. IRS has developed many sources, or reasons, for selecting returns for audit, which can be segregated into those using DIF sources and those 5 using non-DIF sources. DIF scores are automatically calculated for all filed individual tax returns. This DIF calculation is based on a series of formulas developed by IRS that are designed to indicate the returns that have the highest probability of a tax change if audited. The higher the DIF score the greater this probability. Before implementing DIF, IRS had no systematic way to evaluate which among all filed returns had the greatest potential for changes to the reported tax if audited. Instead, IRS relied on its auditors across the 5 The exact number of source codes used varies by year because IRS periodically combines some and develops new ones. See appendix I for list of all source codes used by IRS during the years covered in this review. Page 3 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 country to identify which returns to audit by using their experience and judgment in reviewing returns. This process was time consuming and required large amounts of resources. Starting in 1969, IRS began using DIF as part of a multistage process to narrow down the number of returns to be reviewed by the auditors. This process change was intended not only to improve the effectiveness of return selection, but also to reduce the amount of judgment involved in selecting returns for audit. Calculating the DIF score for all filed returns is the first stage of this process. Then, returns with DIF scores above a national cutoff are to be placed in the DIF inventory, from which they may be selected for audit by a district office. In the next stage, the district offices are to indicate how many DIF-scored returns will be needed by their various suboffices. Returns with the highest DIF scores are to be pulled from the DIF inventory and reviewed by experienced auditors to determine whether the return will be accepted as filed or sent to the district for audit. In this process, known as classification, reviewers also are to indicate which portions of the return should be audited. Once in the district, returns are generally reviewed at multiple levels before being sent to an audit group for distribution to the auditors. Districts also use numerous non-DIF sources for identifying specific returns for audit. As with DIF returns, these non-DIF returns are to be reviewed in the district office at multiple levels and either accepted as filed or forwarded to audit groups. Unlike audits selected using DIF, these returns need not be selected by DIF score or solely from the DIF inventory. For the non-DIF sources we include in our review, we grouped 6 these into five categories. • IRS projects, which include audits that focus on specific types of issues, such as unreported income, or types of taxpayers, such as those operating cash businesses, whose returns have specific characteristics that indicate potential noncompliance; • referrals of potentially noncompliant returns that come from both inside and outside IRS; for example, returns are referred by IRS’ Collection Division and by state tax agencies; 6 In this report we use the term “non-DIF sources” when referring to the five categories. IRS has many other non-DIF sources for audit selection at the district office. Because this review focused on the sources associated with the selection process for audits of taxpayers’ books and records, we did not analyze these other non-DIF sources. A complete list of source codes and the reasons for excluding them can be found in appendix I. Page 4 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 • individuals whose returns were completed by a preparer that IRS has identified as questionable; • regular classification that is a catchall group for various kinds of audits that had to be manually selected, such as those involving bankruptcy; and • all other reasons for selecting a return for books and records audits, such as those involving income from illegal narcotics, tax fraud, or abusive tax shelters. The DIF formulas, together with the multiple levels of review in the selection process, are designed to ensure that returns having the greatest potential for a change to the reported tax are most likely to be audited. If this audit potential can be justified, IRS districts can use non-DIF sources to select returns for audit. Specifically, the districts should show that using a non-DIF source would generally offer more potential for changing the 7 reported tax than continuing to audit returns identified by DIF. The steps for selecting DIF and non-DIF returns for audit are shown graphically in appendix II. To determine the extent to which IRS district offices have used various Scope and sources to select returns for audit, we first reviewed return selection Methodology procedures and sources as outlined in the Internal Revenue Manual (IRM) and in attachments to the instructions for IRS’ Audit Information Management System (AIMS). We collected data on these sources from IRS’ closed case data files. We talked to IRS’ Examination and Research Division staffs in the National Office as well as Examination staff in the Northern California District and Western Regional offices to understand the information about the procedures and sources. Finally, we collected data from the AIMS data files on closed audits on the extent to which IRS used each source in accordance with the criteria outlined below. To compare the no-change rate and additional tax recommended per audited return for the various audit selection sources, we used data from AIMS files on closed audits. Our primary interest was books and records audits of individual returns. We focused on such audits at district offices because service center audits were being evaluated in another assignment. We eliminated audits such as those coded as claims for refund, nonfilers, research and reference, and taxpayer requests because these are generally not books and records audits. In addition, we eliminated audits of related 7 Certain types of returns, such as those claiming large refunds, are considered mandatory and may or may not be selected for audit solely because of the potential for a tax change. Page 5 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 and multiyear returns because of the difficulty of associating these audits 8 to 1 of the 3 years we analyzed. We analyzed returns received by IRS in 1992, 1993, and 1994. Because the criteria for selecting returns may change from year to year, we wanted to organize our data in a manner that would reflect the selection source. We discussed this with IRS officials and concluded that our analysis would be based on the year IRS received the return (filing year) rather than on the year the audit was closed (as presented in the AIMS files). Using the closed case data files for fiscal years 1992 through 1997, we collected data for audits of returns • received in 1992 with audits closed during 1992 through 1995, • received in 1993 with audits closed during 1993 through 1996, and • received in 1994 with audits closed during 1994 through 1997. We used these years because they were the most recent filing years for which IRS had 4 years of audit closures. Tracking 4 years allowed us to account for almost all of the returns selected for audit. Even so, some audits of returns received in 1992, 1993, and 1994 remained opened at the end of our 4 year test period. Using the AIMS files, we determined that • 4 percent of the audits for 1992 returns were open at the end of fiscal year 1995, • 5 percent of the audits for 1993 returns were open at the end of fiscal year 1996, and • 2 percent of the audits for 1994 returns were open at the end of fiscal year 1997. To compare how much of the additional tax recommended was assessed and collected for DIF and non-DIF audits, we selected a sample of audits from the 1992 study population. We selected 1992 instead of 1993 or 1994 to allow as much time as possible for collection activities. This sample was divided into 4 strata: (1) audits selected using the DIF scores and closed with the taxpayer agreeing to the additional tax recommended (agreed audits), (2) audits selected using the DIF scores and closed without the taxpayer agreeing to all of the additional tax recommended (unagreed audits), (3) agreed audits selected using non-DIF sources, and (4) unagreed audits selected using non-DIF sources. We used transcripts from IRS’ master file to compare the amount actually collected with the 8 See appendix I for a more complete description of the source codes eliminated and those used in the study population. Page 6 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 amounts recommended. A more complete explanation of our sample selection strategy and sampling errors can be found in appendix III. There are some limitations on our data analysis. We did not validate AIMS data files on the closed audits. During our analysis of the master file transcripts for a selected sample, however, we noted some data inconsistencies between the master file and the AIMS files. We found source codes on the AIMS files that differed from those on the master file. According to IRS, the source code recorded on the AIMS files is correct, and there is no mechanism to make these changes on the master file accounts. We also found substitute for return (SFR) audits even though we 9 had removed SFR source codes from our analysis. IRS officials indicated that it was not possible to identify all SFRs by source code. We excluded only those SFRs that could be identified using the AIMS coding. Moreover, we did not adjust the dollar amounts for inflation, thus the dollars reported may be conservative. We performed our audit at IRS headquarters offices in Washington, D.C.; the Western Region; the Northern California District Office; and IRS’ Fresno Service Center. Our work was done between January 1998 and September 1998 in accordance with generally accepted government auditing standards. DIF-selected returns accounted for 59 percent of the books and records Sources of IRS Audits audits conducted at IRS’ district offices of returns received in 1992, 1993, 10 and 1994. The remaining 41 percent of the audits were selected from the 5 non-DIF sources. Table 1 shows the number of books and records audits conducted for each of the six sources for returns received by IRS between 1992 and 1994. 9 IRS files a substitute return for certain individuals who do not appear to have filed required tax returns. These returns are prepared from information documents provided by third parties, such as banks and employers. According to IRS officials, these SFRs may have been selected for other audit purposes and thus were not coded as SFRs. 10 Audits of taxpayer books and records averaged less than one-half of 1 percent of the returns received in 1992, 1993, and 1994. During this same period, staffing and other workload limited all district office audits to only about 1 percent of returns received. Page 7 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 Table 1: Number of Books and Records a Audits Conducted for Returns Received Category 1992 1993 1994 Total Between 1992 and 1994 DIF 178,419 142,142 310,557 631,118 Non-DIF-selected sources 135,564 128,993 174,490 439,047 Projects 41,173 40,804 50,797 132,774 Referrals 40,353 31,173 32,102 103,628 Preparers 26,578 25,799 37,561 89,938 Regular classification 16,563 20,008 24,795 61,366 Other sources 10,897 11,209 29,235 51,341 Total 313,983 271,135 485,047 1,070,165 a See appendix I for a summary of the specific source codes included in each of the categories. Source: Summarized from IRS’ closed case data files for fiscal years 1992 through 1997. Most returns are selected for audit using DIF because, according to IRS officials, it is IRS’ primary indicator of noncompliance and enables them to select returns that have the greatest probability of error on a nationwide basis. In addition, IRS officials also told us that districts might be unable to identify any additional returns with a higher audit potential through non- DIF sources than through those selected by DIF. Moreover, it may be easier to obtain large numbers of returns with potential noncompliance through DIF than through non-DIF sources. For DIF, districts simply request that service centers send them a certain number of returns. The service center selects the highest DIF-scored returns meeting the district’s geographic requirements. For non-DIF, selection is more complex. District office officials must identify specific returns that appear to have a greater potential for tax change than DIF-selected returns. These returns must then be ordered individually by taxpayer name and Social Security number. As part of our analysis of IRS’ audit selection sources, we separated the audits into returns filed by nonbusiness individuals with income under $50,000; nonbusiness individuals with income of $50,000 and above; and individual taxpayers whose primary source of income was from self- 11 employment (labeled as business). To look at how IRS distributed audits by sources and these types of individual taxpayers, we combined the 3 years covered in this review. Figure 1 shows that IRS conducted a larger percentage of audits of nonbusiness individuals with income under $50,000 than of individuals with higher incomes or business income, regardless of the selection 11 Individuals report business income on the Schedule C of Form 1040. Business income cannot be broken out the same way nonbusiness income is because IRS does not organize the income data in the same manner. For example, business income is grouped in three categories as below $25,000, $25,000 but less than $100,000, and over $100,000. Nonbusiness income is grouped into five categories that are not comparable. Page 8 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 source. For example, over half the audits with DIF as the source were of individuals with income under $50,000; 30 percent were of higher income individuals; and less than 20 percent were of individuals with business income. It is important to recognize, however, that over 75 percent of the returns received by IRS during 1992, 1993, and 1994 reported income under $50,000. Although a higher absolute number of audits are done on individuals with nonbusiness income under $50,000, IRS audits a greater percentage of individuals with business income and individuals with nonbusiness income over $50,000. Figure 1: Percent of Audits by Type of Taxpayer and Source of Audit Source: GAO analysis of IRS data for 1992, 1993, and 1994. Page 9 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 For audits of returns that were received in 1992, 1993, and 1994, the no- Comparison of DIF change rates were generally lower and recommended tax amounts were and Non-DIF Audits generally higher for the five non-DIF sources compared to the results from DIF sources. This is consistent with IRS’ policy to use non-DIF sources if the audit potential appears to be higher than for DIF sources. However, on the basis of a sample of audits of returns received in 1992, we found that IRS collected a greater proportion of the recommended tax amount from audits selected using DIF than from those using non-DIF sources. The no-change rate varied depending on the selection source and the type of taxpayer being audited. Overall, for the 3 years of audits covered in our 12 analysis, the no-change rate was 22 percent. For the six sources we identified as being books and records audits, the overall no-change rate varied from about 15 percent for audits selected due to a questionable preparer to about 28 percent for audits selected using regular classification. Returns selected using DIF were closed with no change in about 26 percent of the audits. Detailed information on the no-change rate by source and year can be found in appendix IV. Table 2 shows the no- change rate by year and audit selection source. Table 1: No-change Rate by Filing Year and Selection Source Percent of audits closed with no change to tax Selection source 1992 1993 1994 All years DIF 27 27 25 26 Projects 19 19 20 19 Referrals 14 17 17 16 Preparers 15 17 14 15 Regular classification 24 28 30 28 Other sources 22 19 12 16 Source: GAO analysis of IRS data for 1992, 1993, and 1994. For all sources, audits of nonbusiness individuals with income of over 13 $50,000 had the highest no-change rate. Figure 2 summarizes these results for each source and type of taxpayer. 12 IRS reported no-change rates of 15 percent, 12 percent, and 8 percent for all district office audits of individual returns closed in fiscal years 1992, 1993, and 1994, respectively. IRS’ no-change rates are lower than ours because our calculation includes only audits of taxpayer books and records. IRS’ rates include other kinds of audits, such as claims and nonfilers, that often have very low no-change rates. 13 Regardless of source, the lowest no-change rate occurred in the group of taxpayers with the lowest income and the highest no-change rate occurred in audits of taxpayers with the highest income. Appendix IV describes these results in more detail. Page 10 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 Figure 2: No-change Rate by Type of Taxpayer and Source of Audit Source: GAO analysis of IRS data for 1992, 1993, and 1994. The average amount of additional tax recommended per audit covered in our analysis declined from over $4,000 for closed audits of returns received in 1992 to about $3,300 for closed audits of returns received in 1994 (a decrease of about 18 percent). By audit selection source, our analysis of 1992 compared to 1994 showed that the recommended additional tax increased only in audits of returns selected using DIF and preparer sources. The average additional tax recommended in DIF audits increased from about $2,500 in 1992 to almost $3,100 in 1994 (an increase of about 24 percent). The average additional tax recommended in audits selected due to a questionable preparer increased from about $1,950 in 1992 to almost $2,100 in 1994 (an increase of about 7 percent). Our analysis by selection source also showed that the average amount of additional taxes recommended per audit over the 3 years varied. The Page 11 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 average amounts ranged from less than $2,000 for returns selected because of questionable tax preparers to over $10,000 for referrals. However, the average recommended additional tax for referrals was always at least double the averages for other selection sources. Table 3 shows the average additional tax recommended by selection source for each of the years in our analysis. Table 2: Average Additional Recommended Tax by Filing Year and Average additional recommended tax Selection Source Selection source 1992 1993 1994 Overall DIF $2,487 $2,945 $3,082 $2,883 Projects 4,286 3,771 3,357 3,772 Referrals 12,542 8,804 8,329 10,091 Preparers 1,949 1,912 2,094 1,999 Regular classification 3,358 2,507 2,969 2,924 Other sources 3,974 3,373 2,218 2,843 Source: GAO analysis of IRS data for 1992, 1993, and 1994. As figure 3 shows, the average amount of additional tax recommended per 14 audit varied by type of taxpayer and source. It shows lower average tax amounts in audits of nonbusiness individuals with income under $50,000 compared to the remaining two groups for all sources. It also shows higher average amounts in audits of business individuals compared to audits of nonbusiness individuals—except for referrals and other sources. 14 Regardless of the source, the averages ranged from $2,100 for lower income taxpayers to almost $7,100 for individuals with business returns. Page 12 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 Figure 3: Average Additional Tax Recommended by Taxpayer Type and Source of Audit Source: GAO analysis of IRS data for 1992, 1993, and 1994. IRS does not have a system that allows it to readily track taxes assessed and collected from audits by the source of the audit. IRS can track collections through its Enforcement Revenue Information System (ERIS). However, IRS has not analyzed the information in ERIS to determine the collection rate by audit source. Although an analysis by source is possible using information from ERIS, it would be difficult. To obtain information on the amount of recommended additional tax that is assessed and collected, we selected a sample of audits of returns received in 1992 that were included in our earlier analyses. Using information from these sampled audits and IRS’ master file, we estimated assessments and collections for audited returns in this population. Page 13 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 We found no statistically significant difference in the proportion of recommended tax assessed between audits selected using DIF and audits selected using non-DIF sources. However, we found a statistically significant difference in the proportion collected. We estimated that IRS collected about 57 percent of the recommended tax for DIF returns and 15 about 35 percent for non-DIF returns as of April 1998. It was not clear from the data we had why collections were greater for DIF-selected returns. Our estimated 57 percent collection rate for DIF sources is high compared to the collection rate for all individual taxpayer audits conducted at district offices. In a June 1998 report to the House Ways and Means Committee, we found that for all types of district office audits closed in 1992, IRS had 16 collected about 31 percent of the amounts recommended. Caution is needed if one uses the three results we analyzed to compare the Caution Is Needed effectiveness of DIF and non-DIF sources. The no-change rate, the When Comparing the recommended additional tax amounts, and the collection rate on these Results of DIF and recommended amounts do not present a complete picture of audit effectiveness. Nor is it easy to control for other factors that may influence Non-DIF Audits the three types of results we analyzed. To evaluate the overall effectiveness of selection sources, these audit results are not the only ones that could be used. As noted in our June 1998 report, the impacts of audits on voluntary compliance and taxpayer burden are other results that could help IRS develop a more balanced picture of 17 audit effectiveness, but IRS does not collect such data. That report recognized that gathering such data was difficult because of limitations in data sources and research methods. Beyond these possible results, we concluded that IRS could do more to collect data on its direct and indirect costs to do audits and then assess or collect any additional taxes. We recommended that IRS begin to collect such data, and IRS agreed to do so. In addition, IRS does not have data available that would allow it to control for other factors that may affect the three audit results we analyzed. For example, data on whether the results across the selection sources vary by 15 Because of the sample size and variability of the data, we were unable to make statistically valid estimates of the assessments or collections for specific selection sources. Point estimates and sampling errors can be found in appendix III. 16 Tax Administration: IRS Measures Could Provide a More Balanced Picture of Audit Results and Costs (GAO/GGD 98-128, June 23, 1998). 17 GAO/GGD 98-128, June 23, 1998. Page 14 GAO/GGD-99-30 IRS’ Return Selection Process B-279036 the type of tax issue being audited are not available. Nor does IRS have data on how the quality of the audit affects the results across the selection sources. Even with these cautions, comparing DIF and non-DIF audit sources by the audit results we analyzed can be useful. For example, the comparisons we did by source and selection year could provide a baseline that would allow IRS to gauge some of the impacts of changes to the audit selection process over time. We requested comments on a draft of this report from the Commissioner Agency Comments of Internal Revenue. Officials representing the Assistant Commissioners for the Examination and Research Divisions, as well as a representative from the Commissioner’s Office of Legislative Affairs, provided IRS’ comments in a January 13, 1999, meeting. IRS also provided written comments in a January 20, 1999, letter, which is reprinted in appendix V. IRS was in overall agreement with the draft report and said it fairly describes IRS’ return selection program. IRS also provided some additional information to elaborate on issues we raised and technical comments; where appropriate, we made changes to this report on the basis of these technical comments. We are sending copies of this report to the Committee’s Ranking Minority Member, the Chairman and Ranking Minority Member of the Senate Committee on Finance, various other congressional committees, the Director of the Office of Management and Budget, the Secretary of the Treasury, the Commissioner of Internal Revenue, and other interested parties. The major contributors to this report are listed in appendix VI. If you or your staffs have any questions concerning this report, please contact me or Thomas D. Short, Assistant Director, on (202) 512-9110. James R. White Director, Tax Policy and Administration Issues Page 15 GAO/GGD-99-30 IRS’ Return Selection Process Contents Letter 1 Appendix I 20 Definition of Source Codes for IRS Audits Appendix II 23 Flow Chart of IRS' Audit Selection Process Appendix III 24 Study Population 24 Statistical Sample Selection and Weighting 24 Methodology for Sampling Errors and Confidence Intervals of Estimates 25 Controlling for Nonsampling Errors 25 Analyzing Assessments and Collections of Recommended Tax Appendix IV 26 Tables of Audit Results for Filing Years 1992, 1993, and 1994 Appendix V 31 Comments From the Internal Revenue Service Appendix VI 34 Major Contributors to This Report Page 16 GAO/GGD-99-30 IRS’ Return Selection Process Contents Tables Table 1: Number of Books and Records Audits 8 Conducted for Returns Received Between 1992 and 1994 Table 1: No-change Rate by Filing Year and Selection 10 Source Table 2: Average Additional Recommended Tax by Filing 12 Year and Selection Source Table I.1: Number of Audits by Year the Return Was 20 Received Table I.2: Source Code Table—Inclusions and Exclusions 21 Table III.1: Distribution of Audits in the AIMS Database 24 by Sample Strata Table III.2: Confidence Interval for Estimate of difference 25 in Assessments and Collections for DIF-Selected and Non-DIF-Selected Audits Table IV.1: Number of Returns by Activity Code and 26 Source Group--Returns Received in 1992 Table IV.2: Number of Returns by Activity Code and 27 Source Group—Returns Received in 1993 Table IV.3: Number of Returns by Activity Code and 27 Source Group--Returns Received in 1994 Table IV.4: No-change Rate by Source of Audit and 28 Activity Code--Returns Received in 1992 Table IV.5 No-change Rate by Source of Audit and 28 Activity Code--Returns Received in 1993 Table IV.6: No-change Rate by Source of Audit and 29 Activity Code--Returns Received in 1994 Table IV.7 Average Audit Results by Source of Audit and 29 Activity Code--Returns Received in 1992 Table IV.8: Average Audit Results by Source of Audit and 30 Activity Code--Returns Received in 1993 Table IV.9: Average Audit Results by Source of Audit and 30 Activity Code--Returns Received in 1994 Figures Figure 1: Percent of Audits by Type of Taxpayer and 9 Source of Audit Figure 2: No-change Rate by Type of Taxpayer and 11 Source of Audit Figure 3: Average Additional Tax Recommended by 13 Taxpayer Type and Source of Audit Figure II.1: Summary of IRS’ process for selecting 23 returns to audit. Page 17 GAO/GGD-99-30 IRS’ Return Selection Process Contents Abbreviations AIMS Audit Information Management System DCI data collection instrument DIF discriminant function IRM Internal Revenue Manual IRS Internal Revenue Service SFR substitute for return Page 18 GAO/GGD-99-30 IRS’ Return Selection Process Page 19 GAO/GGD-99-30 IRS’ Return Selection Process Appendix I Definition of Source Codes for IRS Audits This appendix shows the types of audits included in our study population. We included sources that appeared to be for audits of taxpayer books and records conducted at one of the Internal Revenue Service’s (IRS) district offices. We excluded audits for two reasons: • Reason 1: We excluded audits that did not deal with the taxpayers’ books and records or that were not conducted at an IRS’ district office. These audits included claims for refund, nonfilers, substitute for return, service center correspondence audits, and research studies. • Reason 2: We excluded related and multiyear audits because we were unable to relate them to specific returns in one of our target years. Also, some of these returns were selected on the basis of audits of partnership or corporation returns Table I.1 shows the total number of audits closed by IRS for the 3-year period of our analysis and the number included in the analysis. Table I.2 shows the individual source codes and indicates which are included in the study population and which are not. This table also provides the reason a source was excluded or the source code group for those included. Table I.1: Number of Audits by Year the a Return Was Received Description Year return was received by IRS Type of audit 1992 1993 1994 Total recorded audits 806,344 1,129,956 1,593,912 Less (excluded audits) Service center auditsb 246,602 305,711 688,441 Nonfilers 115,108 434,238 296,527 Multiple returns 70,826 61,467 60,144 Related returns 47,046 41,148 50,600 Claims 10,309 11,529 10,175 Other excluded returns 2,470 4,728 2,978 Total audits excluded from analysis 492,361 858,821 1,108,865 Returns included in the analysis DIF-selected returns 178,419 142,142 310,557 IRS projects 41,173 40,804 50,797 Referrals 40,353 31,173 32,102 Preparers 26,578 25,799 37,561 Regular classification 16,563 20,008 24,795 All other selection reasons 10,897 11,209 29,235 Total audits included in analysis 313,983 271,135 485,047 a Each of the years contains audit information from 4 years of IRS’ Audit Information Management System (AIMS) closed case database. b Service center audits are those with an organization code of 5000 or above on the AIMS database. Source: IRS’ AIMS database. Page 20 GAO/GGD-99-30 IRS’ Return Selection Process Appendix I Definition of Source Codes for IRS Audits Table I.2: Source Code Table— Inclusions and Exclusions Source Reason code Definition Status excluded/group 01 Tax shelters and automatics Included Other 02 DIF source returns Included DIF 03 Unallowable items Included Other 04 Multiple filers Included Other 05 DIF-related pickup Excluded Reason 2 06 Correspondence examination Included Other 07a Claim—medium examination potential Excluded Reason 1 08 Self-employment examination Included Other 09a All other claims for refund Excluded Reason 1 10 DIF multiyear Excluded Reason 2 11 Studies, tests, and research programs Excluded Reason 1 12 DIF related delinquent return Excluded Reason 2 13a Married taxpayers filing separately Included Other 14 Information Return Program—high underreporter Included Other 15a Math/clerical error abatement Excluded Reason 1 16a Returns with an illegal narcotics issue Included Other 17 Tax shelter program Included Other 20 Regular classification Included Reg. Class. 23 IRA recovery Included Other 24 Nonfiler/refusal to file Tax Delinquency Excluded Reason 1 Investigation 25b Substitute for return Excluded Reason 1 26 Alternative minimum tax program Included Other 30 Claims for refund Excluded Reason 1 31b Paid claims for refund Excluded Reason 1 32 Carry-back refund Excluded Reason 1 35b Administrative adjustment request (claim) Excluded Reason 1 37a Stockholder returns Included Other 39 Tax shelter—related pickup Excluded Reason 2 40 Multiyear examination—non-DIF-related Excluded Reason 2 44 Delinquent return—non-DIF-related Excluded Reason 1 45 Reference and research Excluded Reason 1 46 IRS employee returns Included preparer 48 Service center—unallowable related Excluded Reason 2 49 Return preparer program Included Preparer 50 Related pickup—non-DIF Excluded Reason 2 60 Information reports Included Referral 62 Compliance projects—national office Included IRS projects 63a Referrals from appeals Included Referral 64 Pickup related to form 1065 (partnership), 1041 Excluded Reason 2 (estate), or 1120S (S-Corp). 65 Collection referral Included Referral 66a Referrals from criminal investigation Included Referral 67a Compliance projects—regional office Included IRS projects 68a Underreported income program—DIF Included DIF 70 Referrals Included Referrals 71a Referrals from Social Security Included Referral 73 Taxpayer request and reconsideration Excluded Reason 2 75a Referrals from Justice Included Referral 76a Referrals from other U.S. agencies Included Referral 77 Referrals from state tax agencies Included Referral Page 21 GAO/GGD-99-30 IRS’ Return Selection Process Appendix I Definition of Source Codes for IRS Audits Source Reason code Definition Status excluded/group 80 Taxpayer Compliance Measurement Program Included Other 85 Information return program match Included Referral 86a All other information return program methods Included Referral 88 Special enforcement Included Referral 90 Fraud Included Referral 91 TCMP related returns Excluded Reason 2 95a IRS racketeer Included Referral 96a Strike force Included Referral 97 a Wagering Included Referral 98 a Illegal income Included Referral a These codes used for returns received in 1992 and 1993 only. b These source codes effective for posting years 1994 and after only. Source: IRS’ AIMS closed case data files and AIMS Coding Manual. Page 22 GAO/GGD-99-30 IRS’ Return Selection Process Appendix II Flow Chart of IRS' Audit Selection Process Figure II.1: Summary of IRS’ process for selecting returns to audit. Source: Review of IRS’ written procedures and discussions with IRS officials. Page 23 GAO/GGD-99-30 IRS’ Return Selection Process Appendix III Statistical Methodology for Analyzing Assessments and Collections of Recommended Tax This appendix describes the methodology we used to determine how much of the recommended additional tax the IRS actually assesses and collects. We used information from IRS’ master file and a sample of audits of returns received in 1992 that were audited during the period 1992 through 1995. We used this sample to compare the amount of assessment and collection for audits selected using the DIF score and those where the DIF score was not the primary reason the return was selected for audit. IRS does not maintain information on the source for selecting a return for Study Population audit, the additional tax recommended, and the amount assessed and collected in the same database. To determine what proportion of additional tax was assessed and collected, we looked at information on IRS’ individual master file for amount assessed and collected and the AIMS closed audit database for the selection source and the amount of additional tax recommended in the audit. As noted in our report, we limited our study population to audits of books and records done at IRS’ district offices. To compare how much of the additional tax recommended was assessed and collected for DIF and non- DIF audits, we selected a sample of audits from our 1992 study population. We selected 1992 instead of 1993 or 1994 to allow as much time as possible for collection activities. We also eliminated audits where the AIMS database indicated that a refund had been made or no additional tax had been recommended during the audit. This sample population contained records of 229,550 audits. To obtain the sample of books and records audits, we selected a stratified Sample Selection and probability sample of 1,083 audits of taxpayer books and records. The Weighting sample was stratified by whether or not the taxpayer agreed or disagreed with the adjustments to tax that IRS recommended and whether or not the audit was selected because of its DIF score or for some non-DIF reason. The number of audits sampled in each stratum was based on the number of returns in the population and the total additional taxes recommended in the audits. The division of the population and sample of audits between different strata is shown in table III.1. Table III.1: Distribution of Audits in the AIMS Database by Sample Strata Audits in Results of Strata population examination Audits in sample Agreed DIF-selected 112,512 $330,957,253 332 Agreed non-DIF-selected 100,254 586,909,138 250 Unagreed DIF-selected 9,583 128,733,806 251 Unagreed non-DIF-selected 7,201 258,077,075 250 Total 229,550 $1,304,677,272 1,083 Source: IRS AIMS database and GAO sampling data. Page 24 GAO/GGD-99-30 IRS’ Return Selection Process Appendix III Statistical Methodology for Analyzing Assessments and Collections of Recommended Tax We obtained copies of the master file transcript for 1,082 of the audits we selected. IRS could find no record for one of the taxpayers. For each of the sample audits, we obtained information on the amount of additional tax recommended (from the AIMS database) and the amount assessed and the amount collected (from the master file transcript). The results of how much of recommended taxes was assessed and Sampling Errors and collected shown in this report are estimates because they are based on the Confidence Intervals of sample of audits drawn from the total population of all eligible audits of Estimates returns received in 1992. The accuracy of these estimates is quantified by their sampling errors, expressed as 95 percent confidence intervals. In table III.2 for example, we estimate that the difference between DIF and non-DIF collections is 22 percent. This estimate is surrounded by a confidence interval of + 13 percentage points, indicating that we are 95- percent confident that the actual percentage difference in the population of all audits of returns received in 1992 lies between 8 percent and 36 percent. Table III.2: Confidence Interval for Estimate of difference in Assessments Percent of audit Percent of audit and Collections for DIF-Selected and Description results assessed results collected Non-DIF-Selected Audits Point estimate DIF-selected returns 76% 57% Confidence interval DIF-selected returns + 8% + 9% Point estimate non-DIF-selected returns 83% 35% Confidence interval non-DIF-selected returns + 7% + 9% Percentage difference 7% 22% Confidence interval at the 95 percent + 11% + 13% confidence level for the percentage difference Source: GAO analysis of sampled data for 1992. In addition to the reported sampling errors, various obstacles can occur Controlling for when conducting this type of review and may cause other types of errors, Nonsampling Errors commonly referred to as nonsampling errors. For example, differences in how questions are interpreted and errors in entering data could affect the results. We included steps in both the data collection and data analysis stages for the purpose of minimizing such nonsampling errors. These steps involved the 100-percent review of completed data collection instruments (DCI) and data entry of those DCIs. We also had a second analyst check all computer analyses and programming. Page 25 GAO/GGD-99-30 IRS’ Return Selection Process Appendix IV Tables of Audit Results for Filing Years 1992, 1993, and 1994 We grouped the audit results of our analyses by IRS’ activity codes. These are codes IRS uses to categorize individual tax returns relative to the various tax forms filed and the taxpayer’s income level. The following are definitions of the activity codes: • 530 - Form 1040A with total positive income less than $25,000; • 531 - Non-1040A with total positive income less than $25,000; • 532 - Total positive income of $25,000 or more but less than $50,000; • 533 - Total positive income of $50,000 or more but less than $100,000; • 534 - Total positive income of $100,000 or more; • 535 - Form 1040C with total gross receipts under $25,000; • 536 - Form 1040C with total gross receipts of $25,000 or more but less than $100,000; • 537 - Form 1040C with total gross receipts of $100,000 or more; • 538 - Form 1040F with total gross receipts under $100,000; and • 539 - Form 1040F with total gross receipts of $100,000 or more. For the grouping of IRS audits included in our analyses, the following sets of tables present the results of these audits by source of audit and activity codes for each of the 3 years reviewed—1992, 1993, and 1994. The first set of tables shows the total number of audits, the second set shows the no- change rate, and the third set shows the average audit results. Each of the tables summarizes information from IRS’ AIMS closed case data files for fiscal years 1992 through 1997. Also, a description of the source codes included in each of the source categories we used can be found in table I.2. Table IV.1: Number of Returns by Activity Code and Source Group—Returns Received in 1992 a Number of Audits by Source Regular Compliance Activity code DIF classification Preparers projects Referrals All other Total 530 27,712 6,413 10,950 16,191 11,955 482 73,703 531 23,800 1,151 3,940 6,196 5,584 1,629 42,300 532 37,987 2,067 5,456 6,094 6,459 2,294 60,357 533 36,298 1,921 2,470 3,380 3,171 2,385 49,625 534 20,557 1,441 290 1,694 2,652 1,592 28,226 535 8,414 502 1,078 1,972 3,349 506 15,821 536 9,845 1,394 1,656 2,880 3,614 1,177 20,566 537 10,757 1,388 644 2,347 3,124 715 18,975 538 2,177 101 59 161 249 77 2,824 539 872 185 35 258 196 40 1,586 Total 178,419 16,563 26,578 41,173 40,353 10,897 313,983 Page 26 GAO/GGD-99-30 IRS’ Return Selection Process Appendix IV Tables of Audit Results for Filing Years 1992, 1993, and 1994 Table IV.2: Number of Returns by Activity Code and Source Group—Returns Received in 1993 a Number of audits by source Regular Compliance Activity code DIF classification Preparers projects Referrals All other Total 530 20,871 6,789 11,523 18,143 12,292 2,658 72,276 531 18,107 1,273 3,425 5,195 3,287 1,228 32,515 532 38,913 3,224 5,200 5,370 4,315 1,542 58,564 533 25,533 2,780 2,418 3,125 2,184 1,860 37,900 534 13,643 1,894 274 2,002 2,062 1,083 20,958 535 6,610 445 1,340 1,294 1,638 770 12,097 536 8,501 1,362 1,106 2,138 2,380 1,248 16,735 537 7,752 1,775 434 2,847 2,492 645 15,945 538 1,548 111 59 209 208 94 2,229 539 664 355 20 481 315 81 1,916 Total 142,142 20,008 25,799 40,804 31,173 11,209 271,135 Table IV.3: Number of Returns by Activity Code and Source Group—Returns Received in 1994 a Number of audits by source Regular Compliance Activity code DIF classification Preparers projects Referrals All other Total 530 39,557 7,688 16,136 19,928 12,061 2,077 97,447 531 37,258 1,758 4,981 5,100 3,298 9,817 62,212 532 84,521 3,453 8,167 6,027 4,689 4,857 111,714 533 62,866 3,686 4,290 5,000 2,733 2,822 81,397 534 30,526 2,967 413 3,755 2,078 1,346 41,085 535 14,264 714 1,578 1,571 1,682 3,383 23,192 536 19,358 1,812 1,329 3,390 2,208 3,095 31,192 537 19,488 2,467 538 4,694 2,950 1,311 31,448 538 1,497 94 97 363 179 443 2,673 539 1,222 156 32 969 224 84 2,687 Total 310,557 24,795 37,561 50,797 32,102 29,235 485,047 Page 27 GAO/GGD-99-30 IRS’ Return Selection Process Appendix IV Tables of Audit Results for Filing Years 1992, 1993, and 1994 Table IV.4: No-change Rate by Source of Audit and Activity Code—Returns Received in 1992 a Percent by source of audit Regular Compliance Activity code DIF classification Preparers projects Referrals All other Total 530 17 17 16 15 10 14 15 531 27 30 12 18 12 18 22 532 24 22 12 22 14 22 21 533 31 22 15 26 20 20 28 534 37 32 31 27 23 29 35 535 27 30 14 12 9 15 20 536 20 29 15 14 14 20 18 537 26 32 14 23 20 22 24 538 39 32 23 28 37 32 38 539 38 29 14 27 34 25 34 Total 27 23 15 18 14 21 22 Table IV.5 No-change Rate by Source of Audit and Activity Code—Returns Received in 1993 a Percent by source of audit Regular Compliance Activity code DIF classification Preparers projects Referrals All other Total 530 16 17 20 12 11 26 15 531 27 31 13 20 18 15 23 532 26 33 13 23 17 14 24 533 27 35 17 27 24 11 26 534 43 40 31 29 26 18 38 535 26 25 12 13 14 16 21 536 21 24 13 22 17 18 20 537 31 34 16 27 21 21 28 538 41 36 20 32 38 38 39 539 36 31 25 22 38 25 31 Total 27 28 16 18 16 18 23 Page 28 GAO/GGD-99-30 IRS’ Return Selection Process Appendix IV Tables of Audit Results for Filing Years 1992, 1993, and 1994 Table IV.6: No-change Rate by Source of Audit and Activity Code—Returns Received in 1994 a Activity code Percent by source of audit Regular Compliance DIF classification Preparers projects Referrals All other Total 530 16 20 18 12 11 16 15 531 25 34 10 17 20 5 20 532 23 28 10 21 15 13 21 533 25 36 12 27 21 15 25 534 43 42 24 37 28 24 41 535 23 27 14 23 17 9 20 536 16 29 10 22 17 15 17 537 27 32 17 28 19 19 26 538 38 38 15 23 30 37 35 539 35 30 9 18 29 41 28 Total 25 29 14 20 16 12 22 Table IV.7 Average Audit Results by Source of Audit and Activity Code—Returns Received in 1992 a Average audit results by source of audit Regular Compliance Average audit Activity code DIF classification Preparers projects Referrals All other result—all audits 530 $1,269 $1,976 $1,278 $2,637 $7,409 $5,717 $2,657 531 1,079 2,280 1,504 2,663 8,122 1,719 2,338 532 1,511 2,341 1,878 2,826 8,160 2,116 2,440 533 2,176 2,806 2,444 4,882 11,921 3,298 3,075 534 4,252 8,100 (1,900) 15,839 39,384 7,001 8,537 535 1,705 2,421 2,193 3,599 6,491 2,531 3,037 536 3,300 4,246 4,410 6,382 11,958 3,698 5,429 537 9,823 7,522 9,347 12,180 34,479 8,225 13,929 538 917 6,538 2,226 5,664 6,700 2,094 1,958 539 5,197 1,009 9,282 6,770 14,701 5,214 7,490 Total $2,487 $3,358 $1,949 $4,286 $12,542 $3,974 $4,067 Page 29 GAO/GGD-99-30 IRS’ Return Selection Process Appendix IV Tables of Audit Results for Filing Years 1992, 1993, and 1994 Table IV.8: Average Audit Results by Source of Audit and Activity Code—Returns Received in 1993 a Average audit results by source of audit Average audit Activity Regular Compliance results—all code DIF classification Preparers projects Referrals All other audits 530 $1,430 $2,280 $1,319 $2,157 $5,754 $1,697 $2,420 531 1,143 1,679 1,569 2,755 7,941 1,851 2,181 532 1,515 1,505 1,892 3,008 6,439 2,710 2,079 533 2,236 2,268 2,397 3,287 12,355 4,364 3,023 534 5,773 3,090 4,718 10,011 20,144 6,984 7,398 535 1,757 2,892 2,256 5,528 5,483 1,976 2,776 536 4,235 3,626 4,523 6,817 8,394 3,325 5,058 537 15,269 4,439 8,229 10,628 19,726 7,113 13,410 538 1,116 2,129 2,749 3,363 2,542 4,815 1,709 539 8,297 3,383 2,980 2,596 8,498 5,559 5,817 Total 2,945 2,507 1,912 3,771 8,804 3,373 3,630 Table IV.9: Average Audit Results by Source of Audit and Activity Code—Returns Received in 1994 a Average audit results by source of audit Average audit Regular Compliance results—all Activity code DIF classification Preparers projects Referrals All other audits 530 $1,422 $1,810 $1,466 $1,810 $4,343 $2,085 $1,915 531 1,155 1,885 1,615 3,182 6,412 1,189 1,663 532 1,616 1,678 1,938 2,458 5,067 1,346 1,820 533 2,058 2,146 2,840 3,223 8,398 2,612 2,407 534 6,144 5,851 6,509 8,134 23,286 9,236 7,277 535 2,157 3,254 2,108 3,015 5,815 1,271 2,382 536 5,077 3,424 4,840 4,172 9,156 2,546 4,910 537 13,077 6,323 11,375 7,344 21,767 7,458 12,243 538 1,599 2,141 2,477 1,402 7,984 835 1,924 539 14,738 6,406 4,021 3,003 13,851 2,168 9,428 Total 3,082 2,969 2,094 3,357 8,329 2,218 3,324 Page 30 GAO/GGD-99-30 IRS’ Return Selection Process Appendix V Comments From the Internal Revenue Service Page 31 GAO/GGD-99-30 IRS’ Return Selection Process Appendix V Comments From the Internal Revenue Service Page 32 GAO/GGD-99-30 IRS’ Return Selection Process Appendix V Comments From the Internal Revenue Service Page 33 GAO/GGD-99-30 IRS’ Return Selection Process Appendix VI Major Contributors to This Report Thomas D. Short, Assistant Director, Tax Policy and Administration Issues General Government Tim Outlaw, Senior Evaluator Division, Washington, D.C. Louis G. Roberts, Evaluator-in-Charge San Francisco Field Kathleen E. Seymour, Senior Evaluator Office Samuel H. Scrutchins, Senior Data Analyst Tre M. Forlano, Evaluator Page 34 GAO/GGD-99-30 IRS’ Return Selection Process Page 35 GAO/GGD-99-30 IRS’ Return Selection Process Page 36 GAO/GGD-99-30 IRS’ Return Selection Process Ordering Information The first copy of each GAO report and testimony is free. Additional copies are $2 each. Orders should be sent to the following address, accompanied by a check or money order made out to the Superintendent of Documents, when necessary. VISA and MasterCard credit cards are accepted, also. Orders for 100 or more copies to be mailed to a single address are discounted 25 percent. Order by mail: U.S. General Accounting Office P.O. Box 37050 Washington, DC 20013 or visit: Room 1100 th th 700 4 St. NW (corner of 4 and G Sts. NW) U.S. General Accounting Office Washington, DC Orders may also be placed by calling (202) 512-6000 or by using fax number (202) 512-6061, or TDD (202) 512-2537. Each day, GAO issues a list of newly available reports and testimony. To receive facsimile copies of the daily list or any list from the past 30 days, please call (202) 512-6000 using a touch-tone phone. A recorded menu will provide information on how to obtain these lists. For information on how to access GAO reports on the INTERNET, send e-mail message with “info” in the body to: firstname.lastname@example.org or visit GAO’s World Wide Web Home Page at: http://www.gao.gov United States Bulk Rate General Accounting Office Postage & Fees Paid Washington, D.C. 20548-0001 GAO Permit No. G100 Official Business Penalty for Private Use $300 Address Correction Requested (268829)
Tax Administration: IRS' Return Selection Process
Published by the Government Accountability Office on 1999-02-22.
Below is a raw (and likely hideous) rendition of the original report. (PDF)