Social Security Benefit Reporting in the Survey of Income and Program Participation and in Social Security Administrative Records

Summary

The quality of Social Security benefit reporting in household surveys is important for policy research on the Social Security program and, more generally, for research on the economic well-being of the aged and disabled populations. This is particularly true for the aged among whom receipt of Social Security benefits is nearly universal and reliance on such benefits is considerable. This paper examines the consistency between Social Security benefit amounts for May 1990 as reported in the Survey of Income and Program Participation (SIPP) and given in the Social Security Administration's (SSA's) administrative records for the respondent.

Results show that only 25 percent of the aged and 42 percent of the nonaged reported consistent amounts, as seen in the summary tabulation below. The tabulation presents SIPP- and MBR-reported Social Security benefit amount comparisons, by subgroup. About three-quarters of both groups reported an amount within 10 percent of that in the records. The analysis suggests that beneficiaries under age 65 who were retired workers, aged spouses, and aged widow(er)s are the more accurate reporters. The result is consistent with the idea that newer beneficiaries are more likely to remember the amount of their benefits. In contrast, only about a quarter of disabled workers and of beneficiaries aged 65 or older (regardless of type) reported consistent amounts. Somewhat more than one-third of those who were concurrent SSI beneficiaries reported Social Security benefits that matched those in SSA records, a rate that fell between the most and the least consistent reporters.

Underreporting of Social Security benefit amounts by the amount of the Medicare premium does not appear to be a major problem among aged or disabled beneficiaries in the SIPP, although disproportionate shares of both groups make such reports. However, possible measurement error, particularly substantial underreporting by those at the low end of reported benefit amounts (and, to a lesser degree, overreporting at the high end), may be a nontrivial problem, especially among the aged. Potential causes of the apparent misreporting are discussed, as are limitations in the SSA administrative record measure used in the analysis. The paper also describes directions for further research.

Introduction

The quality of Social Security benefit reporting is important to policy research on the Social Security program and, more generally, to research on the economic well-being of the aged and disabled populations. This is particularly true for the aged among whom receipt of Social Security benefits is nearly universal and reliance on such benefits is considerable. This paper examines the consistency between Social Security benefit amounts as reported in the Survey of Income and Program Participation (SIPP) and those shown in the Social Security Administration's (SSA's) administrative records. Results provide insights into the quality of data in the SIPP, point to some areas of concern, and lead to suggestions for further research.

A particular interest, especially for the aged, is whether or not the amounts reported in the SIPP include the amount of the Supplementary Medical Insurance (SMI), or Medicare Part B, premium. (For persons electing Part B coverage, except those with a Medicaid buy-in, the premium amount is withheld from the monthly benefit payment.) An earlier study, based on data from the Current Population Survey (CPS), showed that 40 percent of the aged in 1973 reported benefits net of the Medicare premium amount (Poehls 1979).¹

The question of whether reported SIPP amounts include SSI premiums takes on increased importance in light of current discussions surrounding recommended revisions in the official poverty measure that were made by a National Academy of Sciences panel (Citro and Michael 1995). One recommendation is to estimate poverty using the SIPP rather than the CPS, which is currently the official data source. Another recommendation would adjust available resources for medical out-of-pocket expenditures (MOOP), including Medicare premiums. As noted in Vaughan (2000a), insofar as the Medicare premium is erroneously excluded from respondents' benefit income reports but deducted from income as part of MOOP expenses under proposed new poverty measures, the Medicare premium would be effectively double-counted. Misclassifications of poverty status could result, and because Medicare coverage is so widespread among the aged, there could be many such misclassifications.²

Data and Limitations

The data used are from a Master Beneficiary Record (MBR) extract file that is exactly matched to a 1990 SIPP extract file.³ The resulting file was chosen for this analysis to yield some results relatively quickly. Although it is valuable for showing differences between SIPP-reported benefits and those given in SSA administrative records, the extract contains very little ancillary information and is not useful for investigating reasons for some differences in benefits that are found.

Comparisons in this paper focus on Social Security benefits for May 1990. June SIPP reports, using three rotations of a linked wave 2 file, are used from the larger SIPP extract file. SIPP reports for June (rather than May) are used because Social Security benefits for May would be received in June. Following Vaughan (2000b), only SIPP cases with a positive person weight for June 1990 are included. Tabulations are limited to respondents whose MBR record showed them to be in current pay status with at least $1 in benefits paid for May and whose SIPP record showed that at least $1 in benefits was reported.⁴ Patterns of reporting differences among aged beneficiaries and among beneficiaries under age 65 are separately examined.⁵ Of the 4,168 SIPP respondents aged 65 or older reporting a positive benefit and with a linked MBR record, 4,084 (or 98 percent) also showed a benefit in the MBR and are used in this analysis.⁶ Among younger respondents aged 18–64, 1,116 reported a benefit in the SIPP file and had a linked MBR. Of them, 1,047 cases (or 94 percent) also showed a May 1990 benefit in the MBR and are used in the second part of the analysis.

The SIPP asks respondents for the total benefit amount before deductions, including any deductions for Medicare. Therefore, the MBR benefit used is the monthly benefit credited (MBC), which reflects the amount payable before withholdings, if any, for the Medicare Part B premium.⁷ Like several other payment amounts on the MBR, the MBC shows the amount the beneficiary is credited with getting for a given month. It is not necessarily the amount received by the respondent for that month, even after correcting for the monthly lag. (As noted, the MBR amount "for" a given month is received the following month.) The amount received in June for May may reflect various temporary reductions or adjustments for overpayments or underpayments. Examples of such adjustments are given in Appendix B. Thus, sources of disagreement between SIPP and MBR amounts may lie with either data source. This issue is discussed further in the concluding section.

SIPP amounts may not exactly match MBR amounts for several survey-related reasons, and benefit imputations are most likely the main cause of differences. While imputation flags were not available on the extract file used for these tabulations, published estimates suggest that, very roughly, about 15 percent may have imputed Social Security benefit amounts (Jabine 1990, Tables 5.9 and 5.10). For respondents who rounded their benefit amount, for example, to the nearest $10, discrepancies would also result. Data processing errors—for example, in recording survey responses, in developing the data file, or incorrect linkages of the SIPP and MBR data—would result in some discrepancies as well.

The increased use of direct deposit for Social Security benefit payments may play a role in misreporting. In 1992, 58 percent of retired workers and their dependents used direct deposit, up from 16 percent in the mid-1970s (Bondar 1994). Like other beneficiaries, those using direct deposit are informed annually of the monthly amount they will receive, but they do not see the payments and may be only generally aware of the amount. The MBR field indicating the use of direct deposit is not on the extract file used for the tables in this paper, nor is a field from the SIPP that indicates whether the respondent's check was directly deposited in the bank.⁸

Results for Aged Beneficiaries

Only a small percentage of respondents aged 65 or older report a Social Security benefit amount in the SIPP that is relatively close to the MBC amount. The distribution of the difference between the Social Security benefit amount as reported in the SIPP and the MBC amount is shown in Table 1. For only 16 percent of aged respondents did the amounts match exactly, and 9 percent were within $1 (plus or minus) of the exact amount.⁹ The 1973 CPS column in Table 1 shows that the low, general level of agreement is quite close to that achieved in a special June 1973 CPS supplement.¹⁰

As noted, nonmatches may occur with respondents who round monthly benefit amounts in their reporting. In an exploration of potential rounding effects, we found that the end digit for the reported benefit amount was zero in 22 percent of the cases. When all cases with SIPP-reported amounts ending in zero are removed, the proportion in the remaining subsample with agreement within $1 rises from 25 percent to 28 percent.

Could it be that respondents forgot their latest cost-of-living increase? Effective December 1989, a 4.7 percent benefit increase had been put into effect. For a crude look at the possibility of such forgetfulness, we removed that increase from the MBC amounts of the 2,301 respondents whose SIPP-reported benefits were lower than their SSA amounts. Comparisons of the result with the SIPP amount showed that only 51 of the 2,301 respondents under review reported an amount equal to their previous year's benefit. This finding suggests that forgotten cost-of-living adjustments (COLAs) may not play a major role in discrepant reporting, but further examination is warranted.

Aged respondents are more likely to underreport, rather than overreport, benefit amounts. In the 1990 SIPP, the majority (56 percent) reported an amount that was $2 or more too low compared with 19 percent who reported an amount $2 or more too high. Nominally comparable figures from the 1973 CPS are 64 percent and 10 percent, respectively.

The size of apparent underreporting in the recent data is also striking. Although 38 percent reported an amount less than $50 below the MBR amount, 11 percent reported a benefit amount at least $100 less than the MBR amount, and 18 percent reported an amount at least $50 less. In contrast, only 7 percent reported an amount at least $50 more than the amount shown for them in the MBR.

We also looked at apparent misreporting in percentage terms (not shown). Overall, almost 60 percent reported amounts in the SIPP that were within 5 percent of the MBR amount (including those within $1). About 18 percent reported amounts that were more than 10 percent too low, and for 7 percent, amounts were more than 50 percent too low. At the other end of the distribution, 7 percent reported amounts more than 10 percent higher than the MBR amount, and only 3 percent reported amounts more than 50 percent higher.

Table 2 summarizes the results with a focus on the probable role of the Medicare premium amount in reporting patterns.¹¹ In 1990, 12 percent reported a benefit amount equal to the MBR benefit amount minus the Medicare premium as shown on the MBR. That result is considerably lower than the 1973 result where 40 percent underreported by an amount equal to the Medicare premium. Some part of this decline may be due to the new role of Medicaid in paying the Medicare premiums for poor beneficiaries.¹² However, the MBR field that might have helped in sorting this out was not on the file used for the tables presented here.

The distribution of the difference between SIPP-reported benefit amounts and amounts shown in the MBR for all respondents aged 65 or older and for subgroups is shown in Table 3. Survey data, rather than administrative data, were used to describe subgroup characteristics to assist researchers without access to the latter. Sex and marital status, age, race and ethnicity, schooling, type of Social Security benefit, SIPP-reported Social Security benefit amounts, and SIPP-reported receipt of at least $1 in Supplemental Security Income (SSI) benefits define the subgroups shown.¹³ The $1 or more in reported SSI benefits is used as a proxy for SSI recipiency. For additional perspective on the size of the differences between SIPP and MBR amounts, mean monthly benefit comparisons using the two sources are also shown in Table 3.

As seen in the column headed "Zero" in Table 3, the percentage reporting amounts consistent with MBR amounts hovers around 25 percent in most of the subgroups shown. The exceptions are the low of 18 percent among black respondents, the high of 34 percent among concurrent SSI beneficiaries, and the relatively wide range—from 16 percent to 31 percent—across subgroups reporting the highest Social Security benefit amounts.^14,15

The percentage underreporting by the amount of the Medicare premium generally falls in the range of 10 percent to 13 percent, although it rises to 17 percent among persons who never married and Hispanics and falls as low as 2 percent among concurrent SSI beneficiaries. To the extent that the Medicaid program is paying Medicare premiums for SSI beneficiaries, we would expect fewer SSI beneficiaries to underreport by the amount of the premium. This low underreporting level among SSI beneficiaries is not evident among all those reporting low Social Security benefits, however, despite some overlap between the two groups.¹⁶

If underreporting by the amount of the Medicare premium no longer appears to be a major problem, reporting patterns across subgroups defined by their SIPP-reported benefit amount may be one. Those reporting the lowest benefits show the most apparent underreporting, and the size of the differences can be striking. More than 20 percent of the lowest benefit group reported amounts that were $100 or more below the amount given for them on the MBR, and 44 percent to 47 percent of those in the two groups with the lowest benefit amounts underreported by at least the Medicare premium amount ($28.60). In general, these apparent underreporting levels are inversely related to SIPP-reported benefits (except among those in the highest benefit group), and overreporting is generally positively related to benefit level. Among those with reported benefits of $900 or more, more respondents overreported their benefit by $50 or more than underreported by that amount.

As noted above, aged SIPP respondents who reported both SSI and Social Security benefits are most likely (34 percent) among the subgroups in Table 3 to report a Social Security amount that matches the amount shown for them in the MBR. While the amounts they report are generally low (not shown), SSI beneficiaries in the sample are more likely to overreport than underreport the Social Security benefit amount. That result is in strong contrast to the patterns discussed earlier among all those with low benefits.

We looked further at the 66 percent (150 cases) of concurrent SSI cases with under- or overreports relative to the MBR. Do they appear to be confusing the programs and reporting their Social Security benefit when asked for their SSI benefit? A comparison of the SIPP-reported SSI amount with the MBR Social Security amount showed no such tendency.¹⁷

We might expect larger differences between the SIPP and MBR Social Security amounts among those aged 65–69 than among other age groups because of the retirement earnings test. In 1990, only those under age 70 were subject to the test,¹⁸ and benefit adjustments for them might result in apparently inaccurate SIPP reports. As seen in Table 3, however, no particular reporting differences appear by age.¹⁹

The three right-most columns labeled "Mean benefit" in Table 3 show average benefit levels in the two sources and the ratio of the SIPP-reported benefit to that in the MBR. Overall, the average monthly benefit amount in the SIPP ($536) is 95 percent of the MBR amount ($563) for the aged, and relationships between SIPP-reported and MBR mean amounts among aged subgroups echo the distributional results previously discussed. Most show SIPP amounts to be 94 percent to 96 percent of the MBR amount, but the ratio increases consistently across subgroups defined by SIPP-reported benefit amounts. At the extremes, the mean monthly benefit amount for the lowest SIPP-reported benefit group ($147) is only 64 percent of the MBR mean amount ($231) for that group, and the ratio is 110 percent ($1,170 compared with $1,065) for the highest SIPP-reported benefit group.

Again, the result for Social Security beneficiaries concurrently receiving SSI benefits does not resemble results for other low-benefit groups. Reported benefits among concurrents are almost 105 percent of the level given in the MBR, rather than the fractions under 90 percent that are seen among those reporting low-benefits.

Results for Beneficiaries Under Age 65

This section looks at reporting patterns among the 1,047 beneficiaries aged 18–64 with a May 1990 benefit shown in both sources. Somewhat more than half (54 percent) are aged 62–64, and the majority of these near-aged persons (64 percent) are receiving retired-worker benefits. Most of those under age 62 are receiving disability benefits (59 percent). Results for all beneficiaries under age 65 are shown in Tables 4 and 5 and are summarized here with some comparisons with patterns among the aged.

Beneficiaries under 65 are much more likely than aged beneficiaries to report amounts consistent with those on the MBR, and that general difference by broad age group was similar in the 1973 data. However, the 42 percent of the younger beneficiaries in 1990 who reported amounts consistent with the MBR represent a decline from the 1973 figure of 59 percent, as seen in Table 4. Underreporting occurred more often than overreporting among those under 65, a pattern similar to that among the aged and roughly similar to that in 1973.

Examined in percentage terms, 67 percent of beneficiaries under age 65 reported amounts in the SIPP within 5 percent (including differences of $1) of the MBR amount (data not shown). That result is somewhat higher than that found for the aged. About 9 percent of the nonaged reported amounts more than 10 percent lower than the MBR amount, and almost 13 percent reported amounts more than 10 percent higher. The proportion of nonaged persons reporting amounts within 10 percent of the MBR amount was quite similar to that for the aged (78 percent and 75 percent, respectively).

Reporting patterns, including mean benefits, for beneficiaries in 1990 under age 65 and by subgroup are shown in Table 5. The average amount of $483 reported in the SIPP is nearly identical to the $481 average benefit from the MBR, a result consistent with the higher concurrence rates seen in Table 4.

Almost 13 percent of disabled-worker beneficiaries reported an amount net of the Medicare Part B premium (see Table 5).²⁰ The figure is quite similar to the almost 12 percent of the aged with that reporting pattern. Concurrent SSI beneficiaries under 65, all of whom qualify for SSI on the basis of blindness or disability, do not exhibit that pattern. Fewer than 2 percent of them reported amounts net of the Medicare premium amount. That could indicate that the Medicaid program is paying the premium, that they did not elect Part B coverage, or that they are following SIPP instructions and reporting their benefits prior to Medicare withholdings. (Most beneficiaries under age 65 in Table 5 would not have been eligible for Medicare. Therefore, the "Equals Medicare premium" percentages in Table 5 have little meaning.)

Among subgroups for beneficiaries under age 65, consistent reporting was positively related to age as shown in Table 5.²¹ Almost one-half (48.7 percent) of beneficiaries aged 62–64 reported amounts consistent with the MBR; the figure dropped to 35 percent among the middle-aged and to only 28 percent among those aged 18–40. Because most 62- to 64-year-old beneficiaries would have recently retired, their recall of the exact benefit amount may be better than that of longer-term beneficiaries, and we might expect better reporting.²² Indeed, 50 percent of retired workers aged 62–64 reported benefit amounts matching the MBR amount compared with 30 percent of their disabled counterparts in that age group (data not shown). The comparison of mean benefits showing SIPP as a percentage of MBR in Table 5 indicates that inconsistent reporting occurs primarily among the youngest beneficiary group. Their mean reported amount is 106 percent of the MBR amount for the group.

Reporting patterns vary considerably across some beneficiary types. More than half of the retired workers and aged spouse and widow(er) beneficiaries reported amounts consistent with the MBR, compared with only 28 percent of disabled workers. Average benefit comparisons present a somewhat different picture, however. Retired and disabled workers reported SIPP amounts that, on average, were very similar to MBR amounts. By contrast, the "other" beneficiary group reported an average benefit that was 117 percent higher than the amount given in the MBR. That group primarily included those receiving benefits because they were adults disabled in childhood or spouses or widow(er)s caring for minor children. Some of the latter groups may be including their child(ren)'s benefits, for which they are representative payees, in their benefit reports.

Patterns by Social Security benefit level are similar to those seen earlier for the aged, although they are less consistent and far less pronounced. Persons reporting very low benefit amounts are more likely to underreport (relative to MBR amounts), and those reporting the highest amounts are more likely to overreport. The SIPP-to-MBR mean benefit ratios increase consistently with the SIPP-reported Social Security benefit amount, rising from 67 percent among those reporting the lowest benefits to 119 percent among those reporting the highest. However, the percentage reporting amounts consistent with the MBR varies considerably—from 23 percent to 54 percent—and shows no particular pattern.

Reporting among concurrent SSI beneficiaries reflects the general pattern for all aged 18–64, and consistent reporting is only 5 percentage points higher than it is among concurrent beneficiaries aged 65 or older.

Size Distributions

This analysis, thus far, has compared the exact amount of benefits (plus or minus $1) reported for the same individual as it appeared in the SIPP and the MBR using the linked data. A final set of comparisons looks at distributions of Social Security benefit amounts grouped into categories of benefit amounts according to their size. In effect, SIPP and MBR results developed for this final set of comparisons were obtained as if the files were not linked. (Mean benefits shown in Tables 3 and 5 were similarly developed.) The picture, based on benefit categories, is both more reassuring than results from the linked data and generally consistent with them.

To gauge the similarity of the distributions from the SIPP and the MBR, measures of the index of dissimilarity are included for each set of comparisons. The index indicates the percentage of beneficiaries in either the SIPP or the MBR distribution who would have to change benefit categories to obtain equivalent distributions from the two sources. Among the aged, 5 percent would have to make such a change (Table 6), and roughly the same proportion would have to do so among subgroups defined by sex and age. Results by marital status subgroups show a slightly wider range—from 4 percent to 8 percent (Table 7).

Less than 3 percent of the nonaged would have to change benefit categories (Table 8), reflecting the greater concurrence among the nonaged seen in the earlier comparisons. Nonaged subgroup results show the lowest proportion needing to change occurs among retired workers and the highest proportion (almost 7 percent), among those receiving "other" benefit types.

Concluding Remarks

We cannot say with certainty which—the SIPP report or the MBR amount—is the more accurate record of the amount received in a given month, and more could be done to understand the causes of the discrepancies. Two different, but complementary, areas for further work are suggested here. One would use a richer set of linked SIPP/MBR data to examine potential causes of the discrepancies. With such data, the role of imputations, particularly in large discrepancies, should be explored. Depending on the results, ways to improve the quality of Social Security benefit imputations in the SIPP might be developed. The role of rounding in benefit reporting could also be more systematically assessed. With more MBR ancillary data, the role of adjustments for the retirement earnings test, payment changes following widow(er)hood, or other adjustments could be explored. MBR entitlement dates would allow a systematic look at the effect of the recency of entitlement on reporting quality for all beneficiaries, not just those aged 60 or 62–64. MBR indicators would also allow investigation of the degree to which the direct deposit of Social Security benefits may result in less accurate reporting. The study could provide a more careful look at the role of forgotten COLAs in underreports, limiting such an analysis to sufficiently long-term beneficiaries and accounting for recomputations that may also have occurred. It would also be useful to repeat the comparisons made in this study with more recent survey data because the data used here were more than a decade old.

A second and complementary area for further investigation would be aimed at providing a better understanding of SSA's payment system and the frequency of payment adjustments for over- or underpayments.²³ For example, a series of MBR extracts showing benefit status, payment amount, and other indicators for a common month(s) could be pulled at 6-month intervals for a sample of respondents. Because retroactive payment adjustments may occur with greater frequency in the early calendar months, January might be one useful common month to include. The resulting data set could provide a better understanding of the size, frequency, and timing of benefit adjustments and the subgroups most likely to experience them. That would inform research using SSA administrative data linked to the SIPP or any other surveys, because SSA data have often been extracted for linkage a few years after the focus of the survey (and, hence, after record adjustments may have been made).

If possible, additional work would incorporate data from SSA's Payment History Update System (PHUS)—a portion of the MBR that has not traditionally been available for SSA research but may become available. The PHUS maintains all transaction information on payment amounts, including all short-term payment adjustments. The amount of benefits that was actually paid in a given month (rather than for a given month) would be discernible in the PHUS data and could be compared with the MBC amount and the SIPP-reported amount for the month.

As discussed earlier, further research could lead to improvements in the quality of SIPP imputations of Social Security benefits. In addition, depending on what was learned from those investigations, SSA and the Census Bureau could explore introducing benefit information from SSA administrative records on to the SIPP public-use files. Such an addition would complement the information about the type of Social Security benefit recently added to the SIPP public-use files and allow users to make their own adjustments to benefit amounts given in the survey data.

Appendix A: Benefit Reporting Indicators

Comparisons in this paper are of Social Security benefit amounts for Survey of Income and Program Participation respondents who reported receiving benefits in June 1990 and had Social Security Administration administrative data showing that benefits were in current pay status for May. (Benefits are received in the following month.) The analysis excluded cases for which SIPP data and administrative data did not agree about benefit receipt and cases for which no SSA administrative data were available. As noted earlier, somewhat more than 200 records from the Master Beneficiary Record for respondents who received spouse or widow(er) benefits and who were not dually entitled are known to be missing. For the analysis, at least $1 in benefits had to be shown in both the SIPP and the MBR; the MBR also needed to show that the benefit was in current pay status.

This appendix compares Social Security benefit status in the SIPP and MBR and looks at some results that might be obtained under a different approach. The discussion is generally restricted to cases with a positive weight for June 1990 (as noted, they are considered to be "in sample" for the month), but the tables include other cases on the file. The aged, persons aged 18–64, and disabled workers aged 18–64 are examined separately in Tables A-1 through A-6. Two tables are presented for each beneficiary group. The first table in each set looks at reporting and linkage rates in the linked SIPP/MBR extract files. The second compares mean benefit amounts and proportions with low benefits across various subsets of respondents defined by the status of their benefit in one or both sources. Unweighted data are used. Cells for cases used in the analysis in the main paper are highlighted.

Overall, results in this appendix generally show the aged (65 or older) to be better at reporting benefit receipt than persons aged 18–64. That finding is in contrast to results of the paper's benefit amount comparisons, which show greater concurrence between SIPP and the MBR among the nonaged on the amount of benefits received. The appendix results also generally show that mean benefit amounts and the percentages with low benefit amounts are quite similar when all respondents reporting benefits in the SIPP are compared with the respondent subset used in the paper.

Aged (Tables A-1 and A-2)

Concurrence between the two sources on benefit receipt was extremely high for the aged. Cases used in the analysis reported benefits that were almost identical to those reported by all SIPP respondents and to those shown for all linked records in the MBR.

• Of the 4,477 aged respondents who reported Social Security benefits in the SIPP, 93 percent (4,168) have linked records, and receipt for 91 percent (4,084) of the reporters was confirmed in the MBR. Of reporters with linked records, receipt was confirmed for 98 percent.
• Of the 4,117 aged respondents showing benefits in the MBR and with SIPP weights for the month, 99 percent (4,084) reported benefits in the SIPP.
• The latter result is quite consistent with Vaughan (2000b), who shows that 98 percent of retired workers, 98 percent of aged wives, and 99 percent of aged widow(er)s in the MBR reported benefits in the SIPP for December 1990 and December 1991.
• Finally, of the 4,201 aged respondents showing benefits in either source, 97 percent show receipt in both. The data show 4,084 with receipt in both; 84 with receipt in the SIPP and not the MBR; and 33 with receipt only in the MBR, yielding 4,201 with receipt in either.

For the aged, Table A-2 shows that mean reported benefits are virtually identical when all SIPP respondents who reported benefits are compared with the linked subset who also show positive benefits in the MBR ($535 and $536, respectively). Proportions reporting low benefits in the SIPP are also the same or nearly so in the two groups. Under the MBR indicators, mean benefits and proportions showing low benefits are also very similar for all cases and for corroborating cases.

Among the relatively small numbers of aged respondents reporting benefits in the SIPP but without corroborating evidence in the MBR, mean reported benefits are lower than in the analysis sample used. Indicators of the lowest benefits come from the 84 respondents who reported Social Security in the SIPP but whose linked MBR data did not agree. It could be that some of those respondents received only SSI benefits but mistakenly reported them as Social Security. (As noted, the analysis file lacked data from SSI administrative files.)

Nonaged (Tables A-3 and A-4)

Concurrence between the two sources on benefit receipt was lower among those aged 18–64 than among the aged, and a somewhat smaller proportion of the group aged 18–64 who reported benefits had a linked MBR. The cases used in the analysis report slightly higher benefits than either all SIPP respondents reporting benefits or all MBR cases with positive benefits.

• Of the 1,244 respondents aged 18–64 who reported Social Security benefits in the SIPP, 90 percent (1,116) had linked records, and benefit receipt for 84 percent (1,047) of reporters was confirmed in the MBR. Of reporters with linked records, receipt was confirmed for 94 percent.
• Of the 1,230 nonaged respondents showing benefits in the MBR and with SIPP weights for the month, 85 percent (1,047) reported benefits in the SIPP.
• Finally, of the 1,299 nonaged respondents showing benefits in either source, 81 percent (1,047) show receipt in both. The data show 1,047 with receipt in both; 69 with receipt in SIPP and not the MBR; and 183 with receipt only in the MBR, yielding 1,299 with receipt in either.

Among nonaged persons, mean reported benefits of all SIPP respondents is a bit lower than that of the subset whose report is corroborated in the MBR ($479 and $485, respectively). An almost identical difference occurs between those of all respondents showing positive benefits in the MBR and those of the subset with benefits in both sources ($479 and $484, respectively). Interestingly, the mean benefit reported by all SIPP respondents is identical to that shown in the MBR for all those aged 18–64 with linked records ($479).

A slightly higher proportion of all respondents reported benefits in the SIPP of less than $199 compared with such reports from those with benefit reports in both sources (7.9 percent and 7.5 percent, respectively), but proportions with benefits under $299 are identical in the two groups (20.3 percent). MBR indicators show higher proportions of respondents with low benefits across all MBR cases, compared with the subset used in the analysis (9.5 percent versus 7.3 percent and 21.2 percent versus 19.8 percent).

Disabled Workers (Tables A-5 and A-6)

Estimates for disabled workers (all aged 18–64) are developed using the type of benefit information in the MBR. The result is that estimates are not available for respondents without linked administrative data, including those reporting disabled worker as their benefit type in the SIPP. (Tables A-1 through A-4 used only the age variable given in the SIPP data to define the groups under scrutiny.)

Among those with MBR disabled-worker benefits, concurrence is lower than among either the aged or nonaged groups.

• Of the 436 weighted cases showing a disabled-worker benefit in the MBR, 361 (83 percent) reported Social Security benefits in the SIPP. That rate is similar to the 85 percent obtained from 1992 and 1993 SIPP panels for December 1994 reported in Social Security Administration (2001), Appendix A.
• However, the result is lower than that given in Vaughan (2000b). He found that 90 percent of disabled-worker cases in the MBR also reported benefits in the SIPP. Vaughan's results were based on weighted data and were from SIPP/MBR linked files for the December 1990 and December 1991 benefits.
• MBR indicators for all disabled workers compared with the subset who also report benefits in the SIPP are quite similar. The mean is slightly higher ($551 versus $547), but the proportion with the lowest benefits is also higher (5.1 percent versus 4.2 percent), and the proportions with benefits under $299 are very similar (12.6 percent and 12.9 percent).

Program and payment features might contribute to the apparently lower SIPP reporting rates noted earlier. Successful SSA Disability Insurance applications may have a relatively long review period, sometimes including additional time for appeals, before an award is made. Once an award is made, however, the disabled worker can be paid retroactively back to the date of initial entitlement, and the MBR is adjusted to show benefits paid in each month of retroactivity. Thus, some SIPP respondents may have been awaiting a disability award decision during the SIPP interview (and correctly reporting no benefit for May), but subsequent changes to the MBR would have made it appear that the benefit had been received. (Those kinds of program features and their representation in the MBR are discussed in Appendix B.) An additional possibility, noted by Vaughan (2000b), is that some new beneficiaries may report their Social Security income as SSI because they had received SSI payments in the 5-month waiting period. Not all new beneficiaries may understand the subsequent change in the source of their benefit income. We were not able to isolate new beneficiaries for this analysis.

Appendix B: Benefit Amounts in the Social Security Administration's Master Beneficiary Record

A benefit amount in the Social Security Administration's Master Beneficiary Record shows the amount paid for a given month, not necessarily the amount paid in that month.²⁴ The latter—the amount the respondent received in a particular month—may differ from the MBR amount because of adjustments for overpayments or underpayments. An overpayment occurs when SSA pays a beneficiary too much. As soon as SSA detects the overpayment, a refund is requested, or, if the beneficiary prefers, amounts may be withheld from future benefits to cover the overpayment, or other arrangements may be made. An underpayment is an amount due a person that has not been paid. SSA may pay in a single check or add the repayment to another benefit payable to the person. Adjustments are subsequently made to the amounts recorded in the MBR to reflect the amount that should have been paid in any given month.

We are not aware of any estimates of the size and frequency of such adjustments in 1990, but estimates for 2000 suggest that about 15 percent of beneficiaries had at least one over- or underpayment. Some examples of the conditions surrounding the occurrence of such adjustments are given here.²⁵

1. A worker receiving benefits at age 66 in 1990 reports no plans for future work. But he does work in 1990, and in 1992 SSA determines that his 1990 earnings exceeded amounts set by the earnings test and that he was overpaid $2,600 in 1990.

   If his monthly benefit amount is $600 (for easier exposition, examples assume no COLAs or recomputations), SSA might withhold all of his current $600 benefit for 4 consecutive months and $200 in the following month to repay the $2,600. In that case, the 1990 MBA history, which previously showed $600 per month, is changed to show that benefits were reduced. The payment history would then show no benefit for January 1990 through April 1990, an MBA of $400 for May 1990, and an MBA of $600 for the remaining months in 1990.²⁶

   The 1992 record—rather than showing that no benefits were received for 4 months and $400 was received for May (as actually happened)—would show $600 every month. If the SIPP respondent reported what actually happened in June 1990—a benefit of $600 was received—the records would agree only if the MBR record pull occurred before SSA determined that an overpayment had been made. If the pull occurred after the determination of overpayment, the record for May (received in June) would show the lower adjusted benefit of $400, and the respondent would appear to have overreported. If the SIPP respondent also reported what actually happened in 1992 (4 "$0 months," 1 "$400 month," and 7 "$600 months") the records would not agree, and the respondent would appear to be underreporting in the early months.

2. An underpayment could arise if the beneficiary tells SSA that he expected high earnings in 1990 and SSA reduces his MBA for the earnings test. If, in fact, he did not work and SSA determined that in 1992, the agency would adjust the 1990 underpayment with a lump sum. The 1990 record would automatically change to show the unreduced amount that should have been paid, and the 1992 record would not reflect the lump-sum payment that was made in that year.

   The respondent's report of what actually happened in 1990 would agree with the records only before the lump-sum adjustment was made. After that, the respondent would appear to be underreporting. In 1992, if the respondent included the lump-sum amount in his benefit report, he would not agree with the records and would appear to be overreporting.

3. SSA conducts benefit recomputations to credit beneficiaries for additional recent covered earnings. While recomputations occur continuously as needed, major systems operations that result in MBR changes usually occur in October and March. (In roughly 25 percent of all cases, a manual review is required and the cases are processed as workloads permit.) A recomputation to include a particular year's earnings is effective with the January benefit in the year following the year in which the earnings were paid. However, even when things go well, there are lags of several months for earnings to be posted, so payments reflecting the January increase are made retroactively.

   For example, an automatic recomputation for a beneficiary's 1989 earnings would occur in October 1990, if the earnings were posted before the recomputation operation. If the recomputation resulted in higher benefits, a new January 1990 MBA would be created and retroactive payments would be made for January 1990 and later. If the 1989 earnings were not posted until after the October 1990 operation, the automatic recomputation would occur in March 1991, a new January 1990 MBA would be created, the COLA at the end of 1990 would be applied, and a new January 1991 MBA would be created. Retroactive payments based on those amounts would be made for January 1990 and later.

   Depending on the month of the interview and the month the 1989 earnings were posted and benefits recomputed, SIPP-reported benefit amounts for 1990 might show the amount received before the recomputation adjustment. In those cases, and assuming the records were pulled after the recomputation adjustment, respondents would appear to be underreporting. Amounts reported in the month a lump-sum retroactive payment was received would appear to be overreports, relative to what the MBR would show for the month.

4. A 62-year-old insured worker develops a disability that prevents his employment, and he applies to SSA for disabled-worker benefits. While waiting for SSA's disability award (which can take many months), he takes reduced retired-worker benefits to meet his income needs. Those benefits begin for January 1990, in the amount of $580 per month. In October 1990, SSA awards disability benefits in the amount of $725, retroactive back to April 1990.

   In that case, the MBA history, which had showed $580 for each month from January through October, would change to show the $725 monthly amount for April through October. If the SIPP respondent reported the $580 that he actually received for May 1990, he would appear to be underreporting relative to the MBR (assuming that the SSA records are extracted after the changes are made.)

   The period surrounding new disability awards can also result in discrepancies that are somewhat outside the scope of this paper. For example, workers who are not eligible for retirement benefits while awaiting a disability award might rely on a spouse's income or their own savings to meet their income needs. When retroactive disability awards are made in such cases, a comparison of benefit status in the linked records will show an apparent failure by the respondent to report benefits at all for some months. During the 5-month waiting period, an alternative, temporary income source for some people is the SSI program. Subsequent confusion may arise for some beneficiaries about whether benefits in any given month were obtained from the SSI or the Social Security program.

5. A woman receiving benefits as a retired worker might learn that her divorced husband died the previous year. If she received higher benefits as a dually entitled surviving divorced spouse and established eligibility for those benefits, she would be retroactively paid the difference between the retired-worker benefit she had received and the survivor benefit she was due back to her former husband's month of death.

   If she was interviewed before the adjustment was made and the MBR records were pulled after that time, the monthly benefit amount she reported would be less than the total benefit that the records showed she was paid. She would appear to be underreporting. If she was interviewed in the month she received a retroactive payment plus her retired-worker benefit, she would appear to be overreporting.