FULL TEXT: GAO REPORT EVALUATING OF IRS' REFUND OFFSET STUDY.

United States General Accounting Office

IRS' study analyzed the effect of tax year 1981 offsets on tax law compliance in tax years 1982 and 1983 and the effect of tax year 1982 offsets on compliance in tax year 1983. IRS concluded that offsetting the refunds caused taxpayer compliance problems in subsequent years. Specifically, IRS concluded that those offset, as a group, were more likely in subsequent years to (1) not file their income tax returns, (2) become delinquent in their taxes, (3) decrease their income tax withholdings, and (4) have smaller refunds.

SCOPE AND METHODOLOGY

To respond to your request, we discussed the report's methodology and conclusions with IRS officials in the Office of the Assistant Commissioner for Planning, Finance, and Research who did the study to ascertain whether their study adequately accounted for other plausible explanations for the reported results. Our evaluation focused primarily on the adequacy of IRS' methodology to support the conclusions reached, but we also obtained IRS' data and replicated certain analyses to do a limited review of IRS' calculations.

Because of the large sample sizes drawn, the numerical results that appear in IRS report are statistically significant and therefore not due to chance. We used the same samples to do our analyses, so the figures we generated for this report are also statistically significant. We discussed our evaluation results with IRS and HHS officials.

We did our review in accordance with generally accepted government auditing standards. We did our work in Washington, D.C., between April 1987 and January 1988. The results of our review are presented in detail in appendix 1 and are summarized below.

RESULTS IN BRIEF

IRS' methodology, which compared those offset with control groups not offset, was limited because it did not consider whether the observed differences between the two groups existed before the offsets occurred. With respect to the conclusion on increased nonfiling, we extended IRS' analysis to earlier years and found that it could be supported. However, IRS' conclusions concerning increased tax payment delinquencies, lower withholdings, and smaller refunds for offset taxpayers are not fully supported due to the methodological limitation. IRS' data base did not include sufficient data from earlier years to extend the analysis regarding these three conclusions. IRS corrected this methodological shortcoming for nonfiling and balance due behavior for its forthcoming study of tax year 1985 refund offsets.

IRS' methodology also had two unavoidable constraints. First, IRS selected taxpayers for its control groups to closely match the taxable income and filing status of the HHS refund offset group. However, the offset and control groups may not be strictly comparable in nontax-related respects. Although IRS believes the differences in taxpayer compliance it found between the groups are indicative of the effects of the refund offset program, it is unknown whether considering nontax characteristics would have produced different results. The appropriate methodology to resolve the comparability issue would have been to divide the HHS referrals into two groups -- one group with refunds offset and the other with refunds not offset -- and then analyze the effects on compliance. However, IRS officials said they could not implement this approach while, at the same time, ensuring equitable enforcement of the law. Second, due to the limited number of post-offset years that had elapsed at the time of IRS' study, it is unknown whether the taxpayer compliance problems identified by IRS are temporary or permanent. IRS plans to extend its research to address this issue.

CONCLUSION ON INCREASED NONFILING CAN BE SUPPORTED

For the 1986 report, IRS evaluated taxpayer behavior after the offset and concluded that offset taxpayers were more likely not to file a tax return in subsequent years than a general control group. Our analysis of the data, which extends IRS' analysis, also supports this conclusion. We evaluated taxpayer behavior both before and after the offset and found that offset taxpayers were more likely to not file a tax return in subsequent years than the general control group. However, our analysis showed that the difference in nonfiling rates between the offset group and the general control group was less than what IRS reported (4.5 percent versus 7 percent). Even with this supplemental analysis, it is still unknown whether the two groups have similar or dissimilar nontax characteristics. The results could be affected if the characteristics are dissimilar.

OTHER CONCLUSIONS NEED MORE SUPPORT

Our analysis showed that the other three IRS conclusions were not fully supported. To substantiate empirically its conclusions concerning increased tax payment delinquencies, lower withholdings, and smaller refunds for offset taxpayers, IRS should have analyzed data before the offsets to evaluate whether offsets affected taxpayer behavior. However, IRS did not and we were unable to do the required analyses because IRS' data base did not include sufficient data from earlier years.

Although the conclusions concerning lower withholdings and smaller refunds need more empirical support, a decrease in withholdings leading to smaller refunds seems plausible. Those offset could lower their withholdings and, consequently, their refunds, to prevent or mitigate further offsets. They also could become a compliance problem if they had a balance due and did not pay it. But changes in withholdings are not a tax compliance problem per se; IRS permits taxpayers to reduce the amounts withheld to minimize overpayments that necessitate refunds.

EXTRAPOLATION OF IRS' STUDY

Section 2653 of the Deficit Reduction Act of 1984, Public Law 98-369, authorized IRS to collect delinquent debts owed the government, such as student loans, by offsetting them against tax refunds. However, the conclusions reached in this study cannot be extended to other groups whose refunds are offset for other outstanding government debts because IRS' study pertained only to the offsets of taxpayers who were delinquent in their child support payments.

IRS is now completing a study of the refund offset program for tax year 1985 that will cover those referred for nonpayment of student loans, agricultural loans, housing loans, disaster loans, and child and spousal support. For that study, IRS corrected the methodological weaknesses related to nonfiling and balance due behavior as discussed above by obtaining and analyzing taxpayer behavior data before and after the refund offset. IRS plans a follow- up study that will extend the prior year analyses to include withholding and refund size.

AGENCY COMMENTS AND OUR EVALUATION

We provided a draft of this report to IRS for comment. IRS responded that our evaluation of its November 1986 study raised several valid points that were worth considering and that it was pursuing a number of our suggestions as it studies further the effect of refund offsets on taxpayer compliance. IRS noted, however, that it was not certain that such changes to its methodology would significantly alter its findings. IRS' technical comments, along with our evaluation, are included as appendix II.

As arranged with the Committee, we are sending copies of this report to the Subcommittee on Oversight, House Committee on Ways and Means; the Commissioner of Internal Revenue; and other interested parties upon request. If you have any questions, please contact Charles Vehorn of my staff on 272-7904.

Jennie S. Stathis

BACKGROUND

The Omnibus Budget Reconciliation Act of 1981, Public Law 97-35, authorized the Internal Revenue Service (IRS) to use individual income tax return refunds to offset delinquent child support payments owed by the taxpayer to the custodial parent. State child support enforcement agencies refer such cases to the Department of Health and Human Services' (HHS) Office of Child Support Enforcement (OCSE), which in turn refers them to IRS for offset. ¹ In addition, the Deficit Reduction Act of 1984, Public Law 98-369, authorized IRS to offset taxpayers for other nontax debts. The Office of Management and Budget and IRS selected five agencies to participate during the first year of the program. ²

The first year of refund offset occurred for tax year (TY) 1981. The refund offset process for individuals who are delinquent in their child support payments works in the following manner. The states submit to OCSE a certified list of individuals who are delinquent in their payments. The amount of the delinquency must be $150 or more and at least 3 months past due. HHS forwards this list to IRS for possible refund offset. Beginning with TY 1982 the states or OCSE gave advance notice at the end of October before the upcoming tax season of the potential offset to individuals who were delinquent. ³

Historically, IRS has not offset every taxpayer referred by HHS. Table I.1 shows that for TY 1981 and TY 1982, less than half of those referred by HHS had a refund offset by IRS. According to IRS, a major reason for not offsetting refunds of every individual referred by HHS is that those referred did not file a tax return for the offset years. If a tax return is not filed, IRS cannot determine whether a potential tax refund exists to offset. Other reasons for not offsetting HHS referrals include IRS not being able to find taxpayers in its files and taxpayers having a small refund (less than $25) or no credit balance.

      TABLE I.1: INDIVIDUALS REFERRED BY HHS WHOSE TAX REFUNDS

The 1986 report states that its findings cannot be extended to other types of refund offsets, but a former IRS Commissioner said the results would be indicative of the effects of other nontax offsets. ⁵ IRS is now drafting a report to address the effects of the various refund offsets made in tax year 1985 on taxpayer compliance.

OBJECTIVES, SCOPE, AND METHODOLOGY

The Joint Committee on Taxation asked us to evaluate IRS' report, Study of the Effect of Refund Offsets for Delinquent Child Support Payments on Compliance, issued in November 1986. Specifically, the Joint Committee asked that our evaluation focus on the adequacy of IRS' methodology and the validity of both the data and IRS' findings concerning taxpayers' filing practices following a refund offset. To do our evaluation, we first reviewed the report's methodology and conclusions with IRS officials in the Office of the Assistant Commissioner for Planning, Finance, and Research who did the study. We then did a limited review of IRS' data to check its accuracy. Our evaluation focused primarily on whether IRS' methodology supports the conclusions reached. We discussed the results of our evaluation with IRS and HHS officials.

In our methodological review we examined whether IRS' data analysis provided sufficient evidence to support the causal inferences drawn in the report. For example, IRS analyzed delinquency data and concluded that the refund offsets caused an increase in delinquencies. We evaluated whether the study had adequately accounted for other plausible explanations for the same results. We also analyzed the sensitivity of IRS' findings relative to other factors, such as adjusted gross income, taxable income, and total positive income. In this methodological review we assumed that IRS' data were accurate and IRS' analyses were validly carried out. Because of the large sample sizes drawn, the numerical results that appear in IRS' report are statistically significant and therefore not due to chance. We used the same samples to do our analyses, so the figures we generated for our report are also statistically significant.

We replicated certain analyses made by IRS of various groups of offset taxpayers and groups of taxpayers not offset for TY 1981 and TY 1982 to determine whether IRS' calculations and the numbers generated were mathematically correct.

SUMMARY OF IRS' REPORT

IRS issued two studies that evaluated the effect of refund offsets on taxpayer compliance. The first report, issued in October 1983, analyzed the effect on taxpayer compliance for TY 1982 of offsets made for TY 1981. The second report, issued in November 1986, extended the analysis of TY 1981 offsets to TY 1983 and also analyzed the effect on taxpayer compliance for TY 1983 of offsets made for TY 1982.

For its analyses, IRS first separated the HHS referrals into two study groups: those whose refunds were offset and those who were not offset. The nonoffset group represents taxpayers referred but not offset due to their not filing a tax return, having a refund less than $25.00, or having filed a balance due return. Next, IRS created two control groups that were matched with the refund offset group according to taxable income and filing status. The general control group was a random sample of individual taxpayers who had received refunds but were not offset for tax delinquencies. The tax-related control group had their refunds offset for tax delinquencies. Neither control group was offset for nontax indebtedness.

IRS' analyses were based on monitoring the four groups described above. Those offset for TY 1981 were tracked for TY 1982 and TY 1983. Those offset for TY 1982 were tracked for TY 1983. Thus, IRS' study contained eight groups of taxpayers: the HHS referrals offset for TY 1981 and TY 1982; the HHS referrals not offset for TY 1981 and TY 1982; the general control group for TY 1981 and TY 1982; and the tax- related control group for TY 1981 and TY 1982. IRS reached its conclusions by contrasting summary statistics pertaining to filing, delinquency, withholdings, and refunds for the HHS referrals and the control groups for the years following the offsets.

On the basis of its analyses, IRS concluded that offsetting refunds of those referred by HHS for nontax indebtedness affected taxpayer compliance in four ways. However, due to the time frames of IRS' studies, it is unknown whether the taxpayer compliance problems identified by IRS are temporary or permanent. IRS plans to extend its research to address this issue.

First, IRS concluded that the refund offsets caused some taxpayers not to file a tax return in the succeeding year(s). IRS reached this conclusion from two analyses. It first determined the number of taxpayers who had their TY 1981 refunds offset but did not file tax returns for TY 1983, and then it compared the result to the control groups. IRS found that 32 percent of those offset for TY 1981 did not file for TY 1983 (up from 26 percent in TY 1982), compared to 22 and 23 percent from the general control and the tax-related control groups, respectively. IRS then determined how many of the taxpayers who had their TY 1981 refunds offset had filed returns for the 2 previous tax years -- TY 1979 and TY 1980 -- but not for TY 1982. It did the same for the general and tax-related control groups and compared the results. Here again, the offset group had a higher nonfiling rate. Of the taxpayers who filed returns for TY 1979, TY 1980, and TY 1981, IRS found that 14 percent of the offset group failed to file a return for TY 1982, compared to 7 percent for the general control group and 4 percent for the tax-related control group.

Second, IRS concluded that the refund offsets in TY 1981 and TY 1982 caused some taxpayers to not pay delinquent taxes in TY 1983. ⁶ IRS data show that 4.9 percent of the taxpayers offset in TY 1981 and 4.0 percent of the taxpayers offset in TY 1982 were delinquent in paying their taxes for TY 1983. In comparison, the general control group had about half the tax payment delinquency rate for TY 1983 (2.6 percent for the TY 1981 group and 2.1 percent for the TY 1982 group). However, the tax-related control group had the highest percentage of tax payment delinquency cases (9.2 percent for the TY 1981 group and 8.8 percent for the TY 1982 group). IRS explained this result as the tendency of the tax-related control group to fluctuate between refund and balance due (with subsequent payment delinquency) situations.

Third, IRS concluded that those offset in TY 1981 and TY 1982 decreased their tax withholdings for TY 1983. IRS data show that 70 percent of those offset in TY 1981 decreased their dollar withholdings relative to salary for TY 1983, whereas 62 percent of taxpayers in the general control group and 55 percent of the tax- related control group decreased withholdings.

Fourth, IRS concluded that those offset in TY 1981 and 1982 had smaller refunds available in TY 1983. IRS data show that the TY 1981 refund offset group had the smallest amount of tax refund available ($756) for TY 1983 compared to the general control group ($795) and the tax-related control group ($892). The TY 1982 refund offset group also had the smallest amount of tax refund available ($694) for TY 1983 compared to the general control group ($795) and the tax-related control group ($846).

REVIEW OF IRS' METHODOLOGY

Our review of IRS' methodology showed that in selecting its study groups, IRS did not ensure that the HHS refund offset group and the control groups exhibited similar taxpayer compliance characteristics before the offsets. Due to this lack of comparability the groups are referred to as being nonequivalent. ⁷ Consequently, any taxpayer compliance differences after the offset may not be attributed solely to the offset. These differences could have existed before the offset, and IRS' findings could represent the continuation of the earlier differences. To strengthen its analysis, IRS would need to examine the taxpayer compliance patterns before the offsets. This would essentially establish a benchmark by which to examine the differences after the offset. ⁸ IRS corrected this methodological shortcoming for nonfiling and balance due behavior in its forthcoming study of tax year 1985 refund offsets.

Figures I.1 and I.2 present a hypothetical case to illustrate this methodological shortcoming. Each figure compares the behavior of the refund offset group to the general control group both before and after the offset occurred. The behavior considered is the extent to which these groups of taxpayers reduced their withholdings relative to salary following the offset. The after offset results in figures I.1 and I.2, which are the actual numbers, clearly show that a greater percentage of the refund offset group experienced a decline in their relative withholdings than the general control group.

FIGURE I.1: PERCENTAGE OF TAXPAYERS WITH A DECREASE IN THEIR RELATIVE WITHHOLDINGS ASSUMING GROUP EQUIVALENCY BEFORE THE OFFSET

[Figure omitted]

The conclusion that could be drawn from the after offset results depends on the condition existing before the offset. If it were true that before the offset both the offset group and the control group had the same percentage of taxpayers who reduced relative withholdings, then we would conclude from such evidence that the refund offset caused the offset group to alter their relative withholdings. ⁹ This is shown in figure I.1. If the difference between the two groups after the offset also existed before the offsets were made, we could not conclude that the offsets had an effect. This is shown in figure I.2.

FIGURE I.2: PERCENTAGE OF TAXPAYERS WITH A DECREASE IN THEIR RELATIVE WITHHOLDINGS ASSUMING GROUP NONEQUIVALENCY BEFORE THE OFFSET

[Figure omitted]

Our review also showed that although IRS selected taxpayers for the control groups to closely match the taxable income and filing status of the HHS refund offset group, the control groups, in general, cannot be considered strictly comparable to the HHS referrals in nontax-related respects.

In following up on the nonfilers identified by the study, IRS found that many of the HHS referrals could not be found on the tax rolls, and the HHS referrals on the tax rolls tended to be more transient and therefore more difficult to locate or contact than the control groups. ¹⁰ While acknowledging the distinct nontax-related characteristics of the HHS referral population, IRS officials believe that similar nontax characteristics are present in a general sense in the control groups. Even with this limitation, IRS officials believe that the differences in taxpayer compliance exhibited by the offset and control groups between the offset year and the year(s) after the offset indicate that the offsets caused the noncompliance problems previously discussed. It is unknown whether closely matching nontax characteristics of the offset and control groups would have produced different results.

As mentioned in IRS' report, the appropriate methodology would have been to select two representative samples from the HHS referral having refunds for offset. One sample of HHS referrals with refunds would be offset, the other not. With this procedure, any differences between the two samples would then be attributable to the refund offset program. We would have no reason to believe that the two samples would have any other systematic differences, since they both came from the same population. Although this methodology would have been desirable, IRS officials said they could not implement this approach while, at the same time, ensuring equitable enforcement of the law.

IRS' CONCLUSION THAT OFFSET PROGRAM INCREASED NONFILING CAN BE SUPPORTED

IRS concluded that the refund offset group had a higher nonfiling rate in the succeeding year(s) than the control groups. IRS' analysis showed that 14 percent of the taxpayers offset for TY 1981 did not file in TY 1982 but had filed for TY 1979, TY 1980, and TY 1981. In comparison, 7 percent of the general control group and 4 percent of the tax-related control group did not file for TY 1982 but had filed in the previous 3 years. However, in arriving at these percentages, IRS tracked only those TY 1982 nonfilers who had filed returns for TY 1979, TY 1980, and TY 1981. By not considering the filers and nonfilers for all groups in the earlier years, IRS did not evaluate whether a nonfiling problem existed before the offset occurred. Thus, it is possible that the differences between the offset and the control groups existed before the offsets took place or that factors outside the analysis account for the difference.

To overcome this methodological shortcoming, we considered whether a nonfiling problem existed before the offset occurred. We examined the previous year's and subsequent years' nonfiling rates for all groups of taxpayers included in IRS' TY 1981 data base. ¹¹ Table I.2 shows the percentage of the offset group and general control group that did not file a tax return for the year shown but filed for the previous year. For TY 1980, the nonfiling rate for the refund offset group is higher than the general control group. This higher nonfiling rate not only continues after the offset, but the difference in nonfiling rates between the refund and control groups increases from 7.4 percent (19.1 minus 11.7) for TY 1980 to 11.9 percent (22.7 minus 10.8) for TY 1982.

       TABLE I.2: PERCENTAGE OF NONFILERS IN THE OFFSET GROUP

We did an additional analysis to see how nonfiling varied by income class. This is particularly important because most of those referred by HHS for offset have low taxable income. Table I.3 shows that an association exists between nonfiling and income. For example, the percentage of nonfiling taxpayers for the refund offset group decreases from about 39 percent for the lowest taxable level, to about 26 percent for the next higher level of taxable income, to about 14 percent for the highest level of income. Similar patterns of nonfiling exist for the other groups. The table also shows that at every income level examined, those offset were more likely not to file than the control groups.

             TABLE I.3: PERCENTAGE OF TY 1983 NONFILERS

IRS' conclusion that the refund offset program caused those offset to be more likely not to pay the taxes owed in subsequent years was not fully substantiated by the analysis. IRS' analysis showed that 4.9 percent of the taxpayers offset for TY 1981 did not pay the taxes owed for TY 1983, compared to 2.6 percent for the general control group. However, two plausible explanations exist for the higher delinquency rate among those offset: (1) the offsets caused the higher delinquency rate or (2) the differences in the delinquency rates existed before the offset.

IRS' analysis did not estimate the prior tax payment delinquency rates for the HHS referrals or the control groups. Consequently, IRS does not know whether the differences in tax payment delinquencies after the offsets existed before the offsets. If the differences did exist, IRS' conclusion would not be valid. IRS' data base did not provide sufficient information for us to do the additional analysis needed to substantiate the conclusion.

IRS' CONCLUSION THAT OFFSET PROGRAM DECREASED WITHHOLDINGS NEEDS MORE SUPPORT

The IRS study also did not fully prove the conclusion that the refund offset program caused those offset to decrease their dollar withholdings by more than the control groups. IRS supported its conclusion by showing that a greater percentage of those offset reduced their ratio of dollars withheld to salary from TY 1981 (the year of the offset) to TY 1983 than those not offset. However, IRS did not determine if the difference in the ratios of withholdings to salary existed before the offsets. If before the offsets the ratio of withholdings to salary had been on the decline, then the continued decline in this ratio should not be attributed to the offsets. IRS' data base did not provide sufficient information for us to do the additional analysis needed to substantiate the conclusion.

IRS' CONCLUSION THAT OFFSET PROGRAM DECREASED LEVEL OF REFUNDS NEEDS MORE SUPPORT

IRS' conclusion that the offset program caused a decreased level of refunds in subsequent years was similarly unsubstantiated. The downward trend for refunds could have been the continuation of an already existing downward trend from before the offset and not necessarily the result of the offset. An analysis of the earlier years' refund levels would have established whether the decline was attributable to the offsets. IRS' data base did not provide sufficient information for us to do the additional analysis needed to substantiate the conclusion.

IRS selected the control group on the basis of the tax return for the tax year when the offset occurred. The control group for the TY 1981 study was selected on the basis of its TY 1981 returns. Thus, the refund offset group and the general control group have similar taxable incomes for 1981; however, we do not know whether the two groups had similar incomes for TY 1983, the year for which the refund comparison is made. If, for example, the taxable incomes of the HHS referrals and the control groups diverged between 1981 and 1983, then the change in refund size could be partially attributable to changes in taxable income. Moreover, since the tax rate dropped significantly from TY 1981 to TY 1982, refunds would have changed regardless of income levels.

CONCLUSIONS

IRS' methodology, which compared those offset with control groups not offset, did not consider whether the observed differences existed before the offsets. On the basis of our supplementary analysis of the filing behavior both before and after the refund offsets, we concur with IRS' conclusion that the offset group was more likely not to file in the succeeding year(s) because of the offsets. However, IRS' conclusions concerning increased tax payment delinquencies, lower withholdings, and smaller refunds for offset taxpayers were not fully supported due to a methodological shortcoming. To substantiate these conclusions empirically, IRS should have analyzed data before the offsets to evaluate whether offsets affected taxpayer behavior. IRS did not, and we were unable to do the required analyses as IRS' data base did not include sufficient data from earlier years.

Although the conclusions concerning lower withholdilngs and smaller refunds need more empirical support, a decrease in withholdings leading to a smaller refund seems plausible. Those offset could lower their withholdings and, consequently, their refunds, to prevent further offsets. They also could become a compliance problem if they had a balance due and did not pay it. But changes in withholdings are not a tax compliance problem per se; IRS permits taxpayers to reduce amounts withheld to minimize overpayments that necessitate refunds.

IRS' methodology also had two unavoidable constraints. First, IRS selected taxpayers for its control groups to closely match the taxable income and filing status of the HHS refund offset group. However, the offset and control groups may not be strictly comparable in nontax-related respects. Although IRS believes the differences in taxpayer compliance it found between the groups is indicative of the effects of the refund offset program, it is unknown whether considering nontax characteristics would have produced different results. The appropriate methodology to resolve the comparability issue would have been to divide the HHS referrals into two groups -- one group with refunds offset and the other with refunds not offset -- and then analyze the effects on compliance. However, IRS officials said they could not implement this approach while, at the same time, ensuring equitable enforcement of the law. Second, due to the limited number of post-offset years that had elapsed at the time of IRS' studies, it is unknown whether the taxpayer compliance problems identified by IRS are temporary or permanent. IRS plans to extend its research to address this issue.

Although section 2653 of the Deficit Reduction Act of 1984 authorized IRS to offset taxpayers who are indebted to the government for a variety of reasons, IRS' study pertained only to the offsets of taxpayers who are delinquent in their child support payments. The study results cannot be extended to other groups whose refunds are offset for other outstanding government debts, such as defaulted student loans. Without more evidence it would not be realistic to assume that the factors that led to the nonfiling by those delinquent for nonpayment of child support would necessarily exist for other indebted groups.

IRS is now completing a study of the refund offset program for TY 1985 that will cover those debtors referred for nonpayment of student loans, agricultural loans, housing loans, disaster loans, and child and spousal support. For this study, IRS corrected the methodological weaknesses related to nonfiling and balance due behavior as discussed above by obtaining and analyzing taxpayer behavior data before the offset. IRS plans a follow-up study that will extend the prior year analyses to include withholding and refund size.

² The five agencies and assistance programs are: the Department of Agriculture's Farmers Home Administration Loan Program; the Department of Education's Guaranteed Student Loan, National Direct Student Loan, and the Federally Insured Student Loan Programs; the Department of Housing and Urban Development's Title I Housing Loan Program; the Small Business Administration's Disaster Home Loan Program; and the Veterans Administration's Direct Loan and Guaranty Loan Programs. A recent GAO report discussed the procedures used by IRS to administer the refund offset programs. See Tax Administration: Collecting Federal Debts by Offsetting Tax Refunds (GAO/GGD-87-39BR, Feb. 9, 1987).

³ Evaluating how this procedure affected the refund offset program was beyond the scope of our study.

⁴ Report on the Effect of Refund Offsets for Delinquent Child Support Payments, issued in October 1983; and Study of Effect of Refund Offsets for Delinquent Child Support Payments on Compliance, issued November 1986.

⁵ Statement of Roscoe L. Egger, Jr., Commissioner of Internal Revenue, Before the Subcommittee on Oversight, House Committee on Ways and Means (Sept. 19, 1985).

⁶ For its study, IRS separated delinquent taxpayers into two groups: tax payment delinquency cases and taxpayer delinquency accounts. Because the distinction between the two groups was minor and the analytical results alike, we reported the results from the tax payment delinquency group.

⁷ Cook and Campbell provide further elaboration on the usage of nonequivalent groups. Thomas D. Cook and Donald T. Campbell, Quasi-Experimentation: Design and Analysis Issues for Field Settings (Chicago: Rand McNally College Publishing Company 1979), pp. 95 to 146.

⁸ See Cook and Cambell, pp. 95 to 146.

⁹ This conclusion assumes that no other factors occurred during the offset year that would have affected one group more than the other.

¹⁰ Department of Treasury officials also analyzed IRS' report and expressed doubts as to the adequacy of the IRS control groups. These officials believed that IRS' conclusions could not be adequately supported because the control group was dissimilar to the offset group.

¹¹ The TY 1982 offset data base did not have sufficient information to make any estimates for nonfiling before the year of the offset.

APPENDIX II

COMMENTS FROM THE INTERNAL REVENUE SERVICE

Note: GAO comments supplementing those in the report text appear at the end of this appendix.

June 9, 1988

We have reviewed your recent draft report entitled "Tax Policy: Evaluation of IRS' Refund Offset Study".

Because of the nature of our November 1986 study, we believe that your subsequent evaluation raises several valid points that are worth considering in future studies to further determine the effect of refund offsets on taxpayer compliance. In fact, we are pursuing a number of these recommendations in our ongoing studies on refund offsets of nontax Federal debts. We are not certain, however, that such changes to our methodology will significantly alter our findings.

In addition, we would also like to address your comments on lower withholdings. Although we agree with your assessment that changes in withholdings are not a tax compliance problem per se, we wish to point out that if refunds for years subsequent to the offset are reduced this will also reduce the revenues derived from this program.

Enclosed are our responses to the technical issues that you raised in your report concerning the study methodology, study findings and references to the refund offset program for non-tax debts.

We hope you find these comments useful.

With kind regards,

Sincerely,

METHODOLOGY

CONTROL GROUP SELECTION

The report implies that IRS used inappropriate methodology in selecting the control groups. We agree that the methodology was not ideal. However, IRS could not implement the ideal approach while, at the same time, ensuring equitable enforcement of the law. Hence, for reasons of policy, none of the referrals could have been eliminated from the offset program. We suggest that the report language further clarify this issue. As provided by IRC 6402(c), the IRS "shall" reduce the payment of a refund due a taxpayer by the amount of any past due support. Although Congress intended to evaluate the effect of the refund offset program on tax compliance, it did not provide a mechanism whereby the IRS could fail to make an offset for a properly certified debt referred to it.

OFFSET VERSUS NON-OFFSET CASES

The report contains contradictory statements as to whether IRS divided the group of referrals into offset cases and non-offset cases because it uses the terminology "not offset" in two ways. First of all, on page 14, the terminology "not offset" coincides with the IRS defined non-offset cases which are those referred individuals who were found on the tax rolls and who behaved in one of the following ways: (1) they did not file a tax return; (2) they had a refund of less than $25; or (3) they filed a balance due return. On pages 4 and 21, however, the report states that the control group should have consisted of HHS referrals which were not offset. In these instances, the report is not referring to the non-offset cases as defined by the IRS. Instead, it is describing the ideal control group explained above that the IRS does not have the authority to implement. Consequently, we recommend that the wording regarding cases that were "not offset" be further clarified.

COMPARISON WITH HHS REFERRALS

The report further indicates that IRS selected the control group taxpayers to match the characteristics of the HHS referrals. This statement is not accurate. IRS selected the control group cases to match the characteristics of a portion of the HHS referrals, namely the HHS cases in which an offset of a refund occurred.

Since the control group cases were selected to match only a portion of the referrals, comparisons between the entire group of referrals and the control group cases are inappropriate. While it is true that many of the referrals could not be found on the tax rolls, the significant point is that all of the offset cases were found on the tax rolls just as the control group cases were. Similarly, the reference to a greater amount of transience among the referrals than among the control group cases is not really relevant to the report.

TAXPAYER BEHAVIOR BEFORE AND AFTER OFFSET

The report concludes that the IRS' methodology should have included analyses of taxpayer behavior before and after the refund offset. In addition, the report states that this problem has been corrected in the IRS' study of the refund offset program for Tax Year 1985. In fact, the problem has only been partially corrected in this later study. For the years prior to the offset year, only the nonfiling and balance due behavior was analyzed for the Tax Year 1985 study. The Tax Year 1986 study will further correct the methodological problems by extending the prior year analyses to include withholding and refund size.

CONCLUSIONS

CHANGE IN REFUND SIZE

The report examines possible causes for changes in refund sizes and suggests that the drop in the tax rate could have had an impact on refund sizes. This factor was controlled in the IRS study. The IRS compared refund sizes between the control group and the offset group. Both groups of taxpayers were affected by the change in the tax rates. Therefore, any differences in refund sizes between the two group, cannot be attributed to the change in the tax rate.

NON-TAX DEBTS

Certain statements in the report which reference the refund offset program for non-tax debts owed to federal agencies may be misleading. The report refers to authorized IRS offsets for certain specified assistance programs administered by five agencies and a pending IRS study of such offsets. The report language implies that the IRS limited offsets to debts arising out of particular, specified assistance programs. However, in fact, consistent with the statutory provisions authorizing the refund offset program, the IRS never limited offsets to specified agency assistance programs. In 1984, Congress enacted IRC 6402(d) and 31 USC 3720A, providing for a three- year refund offset program for non-tax debts. Treasury was authorized to issue implementing regulations and prescribe the terms of agreements with participating agencies. Treasury was also authorized to test the offset program with selected agencies before proceeding with full implementation of the provisions.

For Tax Year 1985, the first year of the program, memorandums of agreement were executed with the following agencies:

o Department of Agriculture

o Department of Education

o Department of Housing and Urban Development

o Small Business Administration and

o Veterans Administration.

We note that for Tax Year 1986, the IRS expanded participation to include the following five additional agencies:

o Bureau of the Mint

o Bureau of the Public Debt

o Department of Health and Human Services

o Department of Justice and

o Department of Defense.

(It was agreed that the Department of Defense would submit separate tapes and accompanying certifications for the Air Force, Army, Navy, Marine Corps, Army Exchange Service, and NAVRESCO.) Pursuant to the implementing regulations and memorandums of agreement with participating agencies, the IRS offsets refunds of taxpayers for all legally enforceable debts that are properly certified by participating agencies.

OTHER

The titles of the October 1983 report and the November 1986 report are similar but not the same. The October 1983 report is titled The Effects of Refund Offsets for Delinquent Child Support Payments while the November 1986 report is titled Study of the Effect of Refund Offsets for Delinquent Child Support Payments on Compliance.

The following are GAO's comments on IRS' letter of June 9, 1988.

1. No change deemed necessary. We recognized in the report that IRS was unavoidably constrained in its methodology for selecting the control groups.

2. Changes, such as adding the reasons why all HHS referrals were not offset, were made to make a clearer distinction between offset and nonoffset referrals.

3. Changes were made to indicate that IRS selected taxpayers for its control group by matching the taxable income and filing status of the HHS offset group.

While it is true that the HHS offset and control groups filed in the offset year, these groups do not share other characteristics. For example, the HHS referrals are delinquent in their child support and, as noted in IRS' report, the referrals found on the tax rolls are more transient. Different characteristics such as these support our contention that the groups are nonequivalent. Due to this nonequivalency, IRS needs to analyze the years before the offset. Therefore, we believe comparisons of the characteristics between the HHS referrals and control groups are appropriate and relevant.

4. Changes were made as required to properly reflect the status of IRS' efforts to eliminate methodological shortcomings.

5. Assuming all other variables were held constant, we agree that a change in the tax rate would not cause differences in refund sizes between the two groups. However, as the report discusses, IRS controlled for tax rates in the offset year by controlling for taxable income and filing status but did not control for tax rates in the years following the offset. For example, the refund offset group and the control group had similar tax rates for TY 1982; however, tax rates for the groups were not similar for TY 1983. The reason for the differences is that a higher proportion of the refund offset group did not file a return for TY 1983, and the nonfiling by this group differed by taxable income class from the general control group. Thus, the refund offset group and general control groups were no longer identical in terms of taxable income. Consequently, the difference in refund size between the two groups may be partially attributable to the fact that the study no longer controlled for tax rates.

6. We added clarifying language to indicate that the Deficit Reduction Act of 1984 did not specify or restrict what federal agency debt IRS was to offset. The act provided for a 2-year offset program. We understand that the Office of Management and Budget and IRS were involved in the selection of agencies that in turn would refer debtors to IRS for offset in the first year. According to IRS' November 1986 report, IRS was involved in agency selection.

7. Changes made to report titles.