Elderly Poverty and Supplemental Security Income, 2002–2005

Introduction

The Supplemental Security Income (SSI) program acts as a safety net by providing a minimum level of income to the aged, blind, and disabled. As of December 2008, approximately 7.5 million persons received SSI, of which 2 million (27 percent) were aged 65 or older (SSA 2009). This group of recipients is about 5 percent of America's senior citizens. Thus, SSI for the elderly is not a major factor in the social assistance landscape. Nevertheless, it does establish an income floor, and it offers an institutional framework for caring for older people who for some reason reach later life with few resources. Given recent economic developments, it is possible that SSI enrollment may grow. Thus, continuing review of SSI outcomes is valuable.

The success of programs like SSI in ensuring minimum incomes for Americans can be measured in various ways. Typically, leaders and researchers have evaluated persons' economic standing using the official Census poverty standard and data from the Current Population Survey's (CPS's) Annual Social and Economic Supplement (ASEC). The official poverty standard is commonly described as "absolute" because it is based on a family budget established in the 1960s and is fixed in real terms (Fisher 1992). In recent decades, the prevalence of poverty among elderly Americans as measured by the official standard has declined substantially. From 1966 through 2006, the poverty rate for persons aged 65 or older fell from 28.5 percent to 9.4 percent. In 1966, the elderly poverty rate exceeded that of adults aged 18–65 by 18 percentage points. By 1993, parity with the poverty rate of other adults was achieved; since that year, the elderly poverty rate has generally been over a percentage-point lower than that registered for adults of "working age" (DeNavas-Walt, Proctor, and Smith 2007, 50). However, it is difficult to trace the connection between SSI and poverty because receipt of SSI is substantially underreported in the CPS. For example, the estimated number of SSI recipients in 2002 derived from the CPS is about 30 percent lower than the count obtained from administrative data (Nicholas and Wiseman 2009, Table 8).

In a recent article, we addressed the underreporting issue by merging CPS/ASEC survey data for 2002 with administrative data on earnings and benefits from the SSI and Old-Age, Survivors, and Disability Insurance (OASDI) programs (Nicholas and Wiseman 2009). We encountered two major problems in this effort. First, for various reasons only about three-quarters of persons surveyed for the CPS could be matched to Social Security administrative records. Second, in a significant number of cases, income sources and amounts reported in the CPS do not match administrative records, although often the differences are slight. We developed two alternatives to address these problems. For the problem of the unmatched records, we experimented both with simply leaving unmatched observations in the data set and relying on income as reported in the CPS and with using only the matched observations, but reweighting them using a propensity score technique. For the problem of the difference between administrative data on incomes and amounts reported in the survey, we developed estimates based on alternative "restrictive" and "inclusive" assumptions about which source to use. After addressing these methodological issues, we used our adjusted data to recalculate the prevalence of poverty using the official poverty standard and to investigate the prevalence of poverty using an alternative, "relative" standard.

Our data adjustments had appreciable effects on the estimates for calendar year (CY) 2002. We managed to reduce the weighted CPS undercount of elderly SSI recipients from 42 percent to 5 percent. Adjustment of income with administrative data reduced the national absolute poverty rate by 0.3–2.8 percentage points, depending on the procedure for incorporating unmatched observations and application of the restrictive or inclusive income adjustment procedures. The effect on estimated poverty rates for elderly SSI recipients was sizable. Adjustment of income with administrative data reduced the estimated aggregate poverty rate for elderly SSI recipients by 7.4–9.4 percentage points, again depending on the method adopted for incorporating unmatched observations and whether the restrictive or inclusive income adjustments were applied.

In addition to poverty estimates that are based on the official standard, we experimented with a relative poverty standard that identifies people as poor if their gross income adjusted for family size is less than half the national median. (We employ the same income measure for both absolute and relative poverty calculations.) This common relative poverty threshold yields a much higher aggregate poverty rate than is registered using the official standard—22 percent versus 12 percent before adjustment of income using administrative data. This difference persists in virtually the same magnitude after adjustment with administrative data because such adjustment generally shifts the entire distribution of income, not just the lower tail. However, for SSI recipients, adjustment does lower poverty rates, but those rates remain at very high levels—from 75.1 percent without adjustment to 70–72 percent, again depending on the choice between using restrictive or inclusive income adjustments.

When this study began, the 2003 CPS/ASEC was the latest public-use file for which matched administrative data were available. Since that time, comparable studies have been completed within the Social Security Administration (SSA) for the 2004, 2005, and 2006 CPS/ASEC data, allowing replication of our methodology for CYs 2003–2005. This article reports the results of our 2002–2005 analysis and outlines opportunities for additional research.

SSI Background

The SSI program provides a basic monthly national income guarantee, called the federal benefit rate (FBR), to persons aged 65 or older, blind individuals, and qualified children and adults with disabilities. The FBR is adjusted annually for inflation and stays constant in real terms. In 2002, the baseline year for this study, the FBR was $545 per month for a single individual and $817 for a couple (the 2009 amounts were $674 and $1,011, respectively). SSI is a program that provides a minimum level of income for needy aged, blind, or disabled individuals and acts as a safety net for those who have little or no Social Security or other income and limited resources (SSA 2009). To be eligible for SSI, applicants must pass financial tests involving certain assets and net ("countable") income and a medical test if disabled and nonelderly. Once eligibility is established, the SSI payment is the FBR minus the recipient's countable income and/or any "in-kind support and maintenance" received from others. In all states but two, the federal SSI payment is augmented for at least some SSI recipients by a state supplement (SSA 2008).

Because SSI eligibility is not determined by total household or even family income, a substantial number of recipients living with persons other than their spouse are not poor, although by official standards, anyone living solely on the FBR is considered to be poor. In 2002, the official poverty standard was $9,359 for a nonelderly single person ($8,628 if aged 65 or older) and $12,047 for a nonelderly couple ($10,874 if the "householder" was aged 65 or older). The annualized FBR—$6,450 per year for a single individual and $9,804 for a couple—was therefore even less than the poverty standard applicable to elderly persons. Despite this shortfall, it is possible for SSI payments to lift some persons out of poverty when considered in combination with the income of other family members. For others, SSI at least reduces the gap between income and the poverty standard, especially in states with substantial supplements.

The Data

We use CPS/ASEC data in conjunction with various Social Security administrative files to examine trends from CYs 2002 through 2005.¹ Our administrative data provide information about a person's wages and salaries, self-employment, OASDI, and SSI income. We rely on the CPS for information about all other categories of income.

The Current Population Survey

The CPS is a monthly household survey conducted by the Census Bureau. This survey is the main source of employment information about the civilian noninstitutionalized American population. The CPS provides household, family, and person-level data about employment, unemployment, earnings, hours of work, and other indicators. Additional data are collected in the ASEC for CPS households on various family characteristics plus income received in the previous year. For poverty calculations we follow Census Bureau practice and exclude a small number of children living in households with no relatives because no income data are collected for such persons.

To protect confidentiality, income data in the CPS are subject to top- and bottom-coding. When reported amounts exceed certain thresholds, the actual amounts reported are replaced (top-coded) with average reported amounts for the same item for all surveyed persons with above-threshold amounts and identical (on certain dimensions) demographic characteristics. Bottom-coding occurs for losses from farm and nonfarm self-employment income. When persons are known to have received certain types of income but amounts are not reported, the Census Bureau imputes the missing amount using "hot-deck" methods. In this procedure, missing values are imputed using the amounts reported for sample observations with identical (on certain dimensions) demographic characteristics. It is possible for top- or bottom-coded amounts to be used in such imputations, depending on the data processing sequence.

Social Security Administrative Files

The administrative files we employ from SSA include records of individual earnings in employment covered by the OASDI program as well as SSI payments and OASDI benefits. The data sources for earnings are the Summary Earnings Record (SER) and the Detailed Earnings Record (DER), the Payment History Update System (PHUS) for OASDI, and the Supplemental Security Record (SSR) for SSI.

Summary Earnings Record. Data herein are an extract from SSA's Master Earnings File (MEF). A primary MEF record is created when a person receives a Social Security number (SSN); thus, every person in the CPS/ASEC for whom an SSN match was successfully accomplished will have an SER. The SER is the first administrative file examined when assessing the extent of the CPS/administrative match.

Detailed Earnings Record. These data are an extract from the MEF, which includes data on total earnings from all sources—wages, salaries, and income from self-employment that are subject to Federal Insurance Contributions Act (FICA) and/or Self-Employment Contributions Act (SECA) taxation. DER coverage extends to all earnings reported by employers on workers' W-2 forms, and amounts are not capped.² These data include deferred wages such as contributions to 401(k) retirement plans.³ Because individuals do not make SECA contributions if they lose money in self-employment, only positive self-employment earnings are reported in the DER. Our data are aggregated across all employers for each individual and include earnings from wages, salaries, and self-employment, in addition to deferred income.⁴

Payment History Update System. These data record OASDI (Social Security) benefits when paid. PHUS data include both total benefit and the amount of benefit subtracted for Medicare Part B premiums. A key feature of the PHUS is that monthly amounts recorded here represent actual payments, not entitlement. Hence if a person begins entitlement for a Social Security benefit in November 2004, but does not actually receive a check for the amount until February 2005, the payment will be recorded for 2005. This corresponds to income received as reported in the CPS/ASEC.⁵

Supplemental Security Record. Data herein provide the information that is needed to calculate and distribute SSI payments. SSA typically creates an SSR record when an individual files an SSI application. Each person's record includes eligibility and payment information as well as income information about ineligible spouses and parents that is pertinent to establishing and maintaining the individual's eligibility. SSR payments are recorded as disbursed. The SSR includes state SSI supplements if federally administered (that is, if SSA makes the payment on the state's behalf). Payments made by state-administered SSI supplement programs are not included in the SSR. For the most part, state supplements are small, and some of the largest (from California, Massachusetts, and New York, for example) are federally administered (SSA 2008, 7). However, benefits in Alaska, Connecticut, Wisconsin, Minnesota, and a few other states are substantial and are administered by the state. By far the largest state-administered SSI supplement is Alaska's. In 2002, that state added $362 to the FBR for singles and $528 to the FBR for couples living independently (SSA 2008, 13).

The Match

The common element among original CPS/ASEC and administrative files is a Social Security number. CPS interviewers request SSNs for all persons aged 15 or older in each household in the address-based CPS household sample. Interviewees are not required to provide these data, but most do, or at least permit the Census Bureau to search Social Security's administrative files for their SSN using name, birth date, and address. SSNs for persons younger than age 15 are all obtained by searching administrative data. Once collected, the CPS data are extensively reviewed and reorganized, missing values are imputed, and potentially identifiable outlier income values are top- or bottom-coded. Upon completion of these adjustments, the Census Bureau produces a public-use data set. CPS public-use data sets do not include respondents' SSNs, but do contain unique household sequence and, within households, person identifiers. These identifiers relate to file structure only and convey no information useful for determining the actual identity of CPS respondents.

Upon release of the public-use CPS data, the Census Bureau provides a special encrypted file to SSA. This "cross-walk" file specifies the SSN for each person in the CPS for whom an SSN has been reported, identified by the household sequence number and person identifier. Only one person at SSA has access to the cross-walk file, who then uses the SSNs to construct SER, DER, PHUS, and SSR files for each person with a corresponding household sequence number and person identifier. Only the CPS identifiers are retained and used to link persons' CPS and administrative records.

Unweighted match rates for CPS person observations and Social Security administrative data are given in Table 1. The key match is for the SER. Primarily because of diminishing respondent willingness to provide SSNs, the match rate declined from the 2003 to the 2005 CPS/ASEC interviews (pertaining to CYs 2002 through 2004). However, the match rate increases substantially for the March 2006 interview. Beginning with the 2006 CPS/ASEC, the Census Bureau altered its policy for collecting SSNs. Rather than asking respondents for their SSNs and for an affirmative agreement for use of such information for data matching, the new protocol requires that respondents not wanting such matches to occur to notify the Census Bureau through that agency's Web site or to use a special mailed response. If no such instruction is received from respondents, SSA uses both the SSN and other information (name, address, age, and sex) that are provided to establish correct SSNs for data matching. As the table indicates, substituting an "opt-out" option for the former "opt-in" procedure for SSN reporting had a major effect.

Match rates for earnings (the DER), OASDI benefits (the PHUS), and SSI (the SSR) are lower than for the SER because not everyone for whom a match was achieved in a particular year had earnings or received SSI payments or OASDI benefits. Note that the DER, PHUS, and SSR match groups are subsets of the SER counts.

The Merge

We turn now to procedures for merging the CPS data with administrative records. "Adjusted data" is the term used for any CPS-reported values that have been replaced with administrative data. We discuss income adjustment first and then describe creation of a reweighted sample subset based on persons for whom we have a successful SER match. The outcome is three CPS samples for each year. "Baseline" samples are comprised of the same CPS/ASEC data applied by the Census Bureau to calculate official poverty estimates for any given year. (The terms baseline, official, and unadjusted refer to the same sample.) "Intermediate" samples involve CPS income adjustments that have only been applied to CPS observations with matching SER records. The "final" samples are restricted to individuals living in families with at least one person with a successful SER match and are reweighted to adjust for variation across families in the likelihood the match criterion is met.

Income-Adjustment Strategy

The baseline for our calculations is income as reported in the unadjusted public-use CPS/ASEC data. We distinguish between restrictive and inclusive assumptions at each step of our adjustment process. For a summary of the procedural protocol, see Nicholas and Wiseman (2009, Table A-1). In general, the restrictive assumption set gives credence to administrative data when both administrative and CPS reports are available, and the inclusive assumption set gives credence to CPS income reports when such reports are not imputed and exceed amounts recorded in our administrative sources.

Our income-adjustment procedure incorporates three important choices. First, when comparing CPS data with income reported in the DER, we generally work with total earnings—the sum of wages, salaries, and self-employment income—rather than distinguish between wages and salaries and income from self-employment. Second, we use reported earnings from the DER, but accept CPS earnings reports in the absence of DER amounts or in cases of loss from self-employment. Third, we rely solely on administrative sources for income from OASDI and SSI. The CPS collects data on 17 types of income, from alimony and veterans' benefits to wages and salaries. Our adjustments involve only earnings, OASDI benefits, and SSI payments. For all other sources the CPS amounts, including imputations and top-coded values, are retained.

The reasons for the earnings strategy are discussed in detail in our previous article. For OASDI and SSI, we rely on administrative data for both our restrictive and inclusive income adjustments. Incorporating OASDI and SSI administrative data is complicated by evidence that CPS respondents sometimes confuse SSI payments with OASDI benefits. In the previous article, we argue that this underreporting is due in part to misidentification of SSI payments as Social Security benefits. If such confusion does in fact exist, we should expect to see and actually do see greater reported OASDI benefits in the CPS among known SSI recipients who fail to report SSI than is the case for individuals who correctly report SSI receipt (Nicholas and Wiseman 2009, Table 4). Given the misreporting problem, we focus our income adjustment on the combined OASDI and SSI benefit. Our calculations also include an adjustment for state-administered SSI supplements (SSA 2004).

The Consequences of Adjustment

Table 2 reports the outcome of our CPS income adjustments, differentiating observations by their CPS/SER match status and whether their earnings were changed, their combined OASDI/SSI total was changed, or whether both earnings and OASDI were adjusted. We are interested here in the prevalence of adjustments within the sample, so the data are unweighted. The table has two panels: one incorporating the restrictive income adjustments and the other incorporating the inclusive income adjustments. We have tabulated here only income changes, without respect to whether the CPS-reported numbers were increased or decreased. (Our previous article provides greater detail for 2002.)

The following four findings should be noted:

1. Income adjustments are made for only CPS observations with an SER match. The bottom row of Table 2 indicates that the proportion of affected observations ranges from a low of 69.4 percent in the 2005 CPS/ASEC (2004 reference year) to a high of 88.1 percent in the following year.
2. Income adjustments are common. This finding is to some extent misleading because any difference between what is in the CPS and what we gain from administrative data is recorded. Moreover, in considering the large number of cases with no changes for both earnings and the sum of OASDI and SSI benefits, it is important to recall that many of these cases receive neither, so zero matches with zero.
3. The 2006 Census data linkage policy change not only increased the 2005 CPS/SER match rate, but also the proportion of CPS earnings and SSI/OASDI totals that our procedures adjust. This outcome might be attributed to a higher incidence of imputations among those observations added on the basis of the new Census "opt-out" procedure. Our adjustment procedure generally substitutes administrative data for imputations under both the restrictive and inclusive income protocols.
4. Adjustments in earnings are generally less prevalent under the inclusive adjustment procedure. This outcome is a consequence of accepting survey earnings reports by the inclusive procedure if reported amounts exceed administrative data and are not imputed. The restrictive procedure substitutes DER data in most of these cases, and each substitution counts as an adjustment. The obvious question is whether the size and distribution of these adjustments have significant effects on our perception of poverty for the elderly and for individuals and families in general.

We began with the CPS baseline samples. Applying the income adjustments to persons with an SER match creates for each year a second, intermediate data set, which is somewhat of an amalgam because at least 24 percent of observations in each year lack an SER match. For this group it is necessary to rely solely on income as reported in the CPS.

The Final Sample

Our objective in constructing our third, final sample is to create a data set for which the administrative match is near "universal." However, because poverty is assessed on the basis of family income, universal is somewhat ambiguous. Three alternatives were considered. One was to limit consideration only to singles living alone who were matched to the SER and to families in which every member was matched. A second, less rigorous, alternative was to limit consideration to persons who were themselves matched even if every person in their family was not. The third was to restrict the sample to singles living alone who were matched as well as any person living in a family in which at least one family member was matched. We chose the third alternative, in part because a majority of unmatched persons who ended up being included under this strategy appeared unlikely to have income. The effect of the most rigorous "every family member matched" approach and the second "every person matched" approach would be to reduce the final unweighted samples on average by about 35 percent and 24 percent, respectively.

For population inference, the original CPS weights still work for CPS observations without matching SER records because all original CPS observations are used for our intermediate analyses. However, this is not true for the final sample, which excludes unmatched observations. Before generating our final estimates, we must adjust the person weights of our CPS restricted sample members.

The absence of a CPS/SER match can be treated as a problem in unit nonresponse—as if failure to provide an SSN that could be matched to the SER is equivalent to refusing to cooperate with the survey at all (Lehtonen and Pahkinen 2004, 115). Adjusting data for nonresponse then requires specifying, to some extent, the circumstances that affect the likelihood of cooperation (Groves and Couper 1998). The simplest assumption is that such outcomes are a random phenomenon, and each sampling unit shares a common probability of responding. The response rate for the survey then provides an estimate of this common probability, and population totals for various features of interest could be obtained by multiplying the analysis weights for respondents by a nonresponse adjustment factor. However, even the simplest tabulation indicates that the match rate is not independent of demographic characteristics. Hence without adjustment, the subset of observations for which matches are achieved cannot be used to make inference about the U.S. population as a whole.

We address this problem by reweighting our matched sample in a manner that reflects the varying propensity across interview units to provide SSNs or the information required for SSA to obtain them. Both poverty and income distribution statistics are based on families and single individuals. Given that absolute poverty assessment involves considering the income of all family members, it would be convenient if every family member had a CPS/SER match. In practice, there are families who have members without a CPS/SER match, and this issue presents a choice of what sample to use in generating population estimates. We choose to generate our final estimates from CPS observations who live in families in which someone in the family is matched, but not necessarily the observations themselves because this selection criteria is the least restrictive. For each year's data, we compute the parameters of a logistic regression for the log odds of being matched in this sense for each of the persons in the CPS sample (Folsom 1991; Iannacchione 1999). We estimate separate functions for persons who are either younger than age 18, aged 18–64, or aged 65 or older (Nicholas and Wiseman 2009, Appendix C-2). We use this function to calculate i and an adjusted weight for each individual observation. These calculations produce a third or final sample made up of unrelated individuals with an SER match and persons in families with at least one member with an SER match, each with a propensity-adjusted weight and both restrictive and inclusive income estimates.

It should be emphasized that these estimates are not only experimental, but we have not attempted to estimate variances for the sample estimates. Because of confidentiality issues, the design information necessary to estimate variances for sample statistics from the CPS is not publicly released, and the variance estimation methodology provided by the Census Bureau is not applicable to the final sample we construct because of the additional reweighting step applied (Census Bureau and Bureau of Labor Statistics 2002; Valliant 2004).

The Results: Absolute Poverty

We turn now to the results, treating 2002 (that is, the income data from the 2003 CPS/ASEC) as the baseline of this study. The same data presentation employed in our previous article for 2002 is used here, and results for 2003–2005 are given in Table A-1.

Poverty in 2002

We begin by examining the consequence of CPS income and weight adjustments on poverty rate estimates using the same poverty thresholds applied in Census Bureau publications. As already noted, for 2002 a single, nonelderly adult living alone was considered poor if his or her gross cash income after transfers but before taxes for the year fell below $9,359; for a family of four with two children, the reference amount is $18,244 (Proctor and Dalaker 2003, 4). The standard increases with family size and varies with composition. Elderly persons living alone or with spouses are assumed to require about 10 percent less income than do nonelderly persons in the same circumstance.

Poverty rates by age group for CY 2002 are reported in Table 3. The table is divided into two parts: (1) results for the total U.S. population as covered by official poverty statistics and (2) results for the SSI recipient population. For both groups, we present results (a) using the same baseline CPS/ASEC data applied for official estimates published by the Census Bureau, (b) based on an intermediate CPS/ASEC data that only involve income adjustments, and (c) from a final sample involving a CPS/administrative matched data set limited to observations with matching SER records as well as CPS income- and weight- adjusted records. Within each estimate group, we present results for children aged 17 or younger, adults aged 18–64, and for those aged 65 or older.

Tabulations in panels 1(a) and 2(a), in Table 3, are based on the same CPS data used by the Census Bureau to generate official poverty estimates. (Our estimates differ slightly from figures published by the Census Bureau because it uses data without top-codes, and we use the public-use sample, which is top-coded.) The official estimates appear for reference for both the restrictive and inclusive computations. We are particularly interested in poverty rates among the elderly and SSI recipients. National poverty rates for working-age and elderly populations in 2002 were 10.6 percent and 10.4 percent, respectively. As anticipated, poverty rates for SSI recipients in all age groups are much higher than rates estimated for the age groups in the U.S. population as a whole.

Tabulations in panels 1(b) and 2(b) report the results of applying our restrictive and inclusive income-adjustment protocols. At this stage of our research, the entire CPS sample is retained, and CPS data are used for all persons for whom a CPS/SER match was not achieved, so the total sample size does not change from that recorded for the CPS. Looking first at the data for all persons, the effect of incorporating administrative data is sensitive to the assumption set. The restrictive income adjustment decreases the estimated aggregate poverty rate from 12.1 percent to 11.8 percent; the estimated rates for all three age groups decline, with the greatest change for the elderly. The inclusive income adjustment produces a much larger reduction in poverty rates for all groups, most notably among the nonelderly. Both adjustments produce lower SSI poverty rates. The effect is most dramatic for persons aged 17 or younger. Under the restrictive income estimate procedure, the poverty rate for elderly SSI recipients is 40.6 percent, more than 7 percentage-points less than the unadjusted CPS estimate. Using our inclusive income-adjustment procedure, the estimate is 39.4 percent, 8.6 percentage-points less than the unadjusted CPS estimate. The unweighted SSI recipient count (the number in the "person records" column under the data summary section of the table) goes up by over a fifth, from 3,635 to 4,381 when administrative data are employed. This outcome is another manifestation of underreporting of SSI in the CPS.

Tabulations in panels 1(c) and 2(c) report the results of applying CPS income adjustments, reweighting the observations' CPS person weights using propensity scores, and restricting the sample to persons living in families with at least one member with matching individual CPS and SER records. The combined effect of our CPS income and weight adjustments (panel 1(c)) is a modest additional decrease in estimated aggregate poverty rates under the restricted convention when compared with estimates based only on adjusting the CPS income data for respondents who could be matched to SSA records. When the inclusive adjustment is employed, estimated poverty rates fall further. The effect varies among SSI recipients; child and nonelderly adult SSI poverty estimates are greater, and elderly rates are less than those estimated without sample restriction and reweighting (Table 3, panels 1(b) and 2(b)).

What drives the difference between the final restrictive and inclusive income estimates? Our previous article indicates that the most sizable difference between our two sets of final estimates is that for earnings and self-employment income, the restricted calculations rely on the DER, that is, earnings reported by employers. The inclusive alternative takes CPS reports when the amounts reported in the survey exceed what appears in matching administrative records. Therefore, inclusive income estimates are larger than those that are restrictive. For the elderly, earnings are less important (although they count because poverty is estimated on the basis of total family income, not just the income of the elderly themselves), but correcting for SSI underreporting has a noticeable impact. Aside from imputations for state-administered SSI supplements, the same correction is applied in both the restrictive and inclusive procedures, and the consequence in both cases is an 8–9 percentage-point reduction in estimated poverty, particularly among elderly SSI recipients. This alteration comes about principally because of the effect on prevalence of SSI receipt, not amounts reported.

Changes in Poverty, 2002–2005

CPS adjustment with administrative data produces poverty estimates for 2002–2005 that differ from official ones generated from unadjusted CPS data. Charts 1 and 2 focus on the differences between unadjusted CPS baseline estimates (reported in panels 1(a) and 2(a) of Table 3) and our final restrictive and inclusive estimates based on adjusted CPS/administrative matched data (reported in panels 1(c) and 2(c) of Table 3). (A complete version of Table 3 is presented for each reference year in Table A-1.)

Chart 1 illustrates absolute poverty rates estimated for the entire national and elderly populations. The basic relationships between baseline and final estimates change marginally in later years. For the U.S. population as a whole, poverty estimates based on our restrictive final data are slightly below those generated from unadjusted CPS data, and estimates based on CPS inclusive final data are lower. The noted restrictive and inclusive income adjustments produce the same outcomes for the elderly from one reference year to another by reducing their absolute poverty estimates by approximately 1–2 percentage points.

Chart 2 plots baseline and final estimates for elderly SSI recipients. This chart is based on poverty estimates appearing in panels 2(a) and 2(c) of Table 3. For 2002, the chart shows that incorporating CPS elements with administrative data produces a sizable reduction in estimated poverty rates for elderly SSI recipients. In contrast, for 2003–2005, adjusted estimates for the elderly are greater, regardless of the income adjustment applied.

The relationship between our baseline and final poverty estimates for elderly SSI recipients differs substantially from the corresponding national estimates. For both the total U.S. population and all elderly persons, our restrictive and inclusive final estimates of the poverty rate for the 2002–2005 period are consistently below the baseline official estimates. For elderly SSI recipients, however, this is true for 2002 (as reported in our previous article), but not for the 2003–2005 period.

What is going on in Chart 2 is unclear, but it is possible to say more about what is driving the change between the outcome for 2002 and for subsequent years. Recall that in moving from baseline to final estimates in each year, the sample base changes. The baseline includes only observations for people who report SSI receipt in the CPS. The sample base for the final estimates includes only observations for people in families matched to Social Security administrative records (using our match criteria) for the year. Persons who report SSI receipt and for whom matches to the SSR SSI records are found are included in both the baseline and final samples. The baseline poverty rates for this group are quite high, ranging from 44–50 percent over the 4 sample years. The final rates are only slightly changed with adjustment of family earnings, family income from other sources including SSI, and reweighting. Poverty rates adjusted for actual SSI receipt among persons who did not report SSI receipt in the 2002 CPS, but in fact were SSI recipients, were substantially lower than rates for those who reported SSI receipt, but for whom no administrative match was obtained. For the final poverty estimates, persons not meeting our CPS/SER match criteria were deleted from the sample of elderly SSI recipients, and persons known from administrative records to be recipients but who did not report so in the CPS were added and counted as SSI recipients. The observations in the resulting subsample were reweighted to reflect the sample adjustments. The result is a lower overall poverty rate than what is obtained from the baseline sample (Chart 2).

For subsequent years, things change. Persons who did not report SSI receipt to the CPS but in fact were SSI recipients have poverty rates higher than estimated for persons in this category in 2002. In 2003 and 2004, these higher rates are similar to those for the persons reporting SSI, but for whom no administrative confirmation is available. The effects of adjustment on the family income of those who did report receiving SSI are larger and result in substantial reduction in average estimated family income. The combination of changes causes the final samples to have a higher overall poverty rate that exceeds the baseline estimates. Because of the procedural change for collection of SSNs, discussed earlier, the match rate for the 2005 data is much higher, and the proportion of the elderly persons reporting SSI receipt that is verified with administrative data increases. Nevertheless, the final poverty estimates are similar to those for 2003 and 2004. In sum, to our knowledge the difference between the 2002 and 2003–2005 samples cannot be related to some change in the way the CPS collects SSI data or other administrative factors, so the outcome remains an anomaly.

SSI Population Estimates

CPS income and weight adjustments substantially increase the sample-based estimates of the total population of SSI recipients. Estimates of the total SSI recipient population by age group for the original and modified CPS samples for each year are given in Table 4. The first bank in column (1) specifies the sum of sample weights for persons for whom the unadjusted 2003 CPS/ASEC reports receipt of SSI in 2002. The second column indicates intermediate estimates generated from the same CPS sample used for official poverty estimates, but matched to administrative sources and involving adjustment to only CPS income records. The third column gives our final estimates of the number of recipients calculated on the basis of our restricted CPS/administrative-matched sample with CPS income and weight adjustments. Column (4) shows the average monthly SSI caseload for 2002, indicated by SSA's 1 percent SSR sample. Column (5) gives, from the same 1 percent SSR sample, an estimate of the number of persons in the CPS sample universe who had income from SSI in 2002.

Relative Poverty

We turn now from absolute to relative poverty assessment. Reliance on absolute poverty measures, especially measures as old as the official U.S. standard, is controversial. In our previous article, we considered the consequences of evaluating poverty on a relative basis, using the common Organisation for Economic Co-operation and Development standard of 50 percent of median income before taxes (Förster and Mira d'Ercole 2005). We convert family income into "individual equivalents" using an equivalence scale suggested by a recent National Research Council (NRC) review of recommendations for poverty standard reform.⁶ Because of data limitations, we conduct this analysis using the same "pretax, posttransfer" income measure as that employed in official statistics. Ideally we would include income benefits such as food stamps, earned income credit, and housing subsidies, but we could not do so. This issue is discussed further in our conclusions.

The Income Distribution in 2002

Again, we use 2002 and our previous analysis as an anchor. The results appear in the two parts of Table 5: part 1—based on the restrictive adjustment protocol, and part 2—based on the inclusive alternative. Both parts of the table show results for unadjusted CPS data, the sample that combines adjusted data for matched households with CPS data alone for the unmatched, and a third sample of matched CPS data reweighted to adjust for nonresponse. Looking first at part 1, the table identifies the points of demarcation for various deciles of the income distribution for each of the three samples and then the proportion of all observations that fall within the corresponding interval (to save space, deciles 30 and 40 and deciles 70 and 80 are combined). By definition, for each sample, 10 percent of all people fall within each decile. What is of interest here is the location of the median, the corresponding poverty standard, and the proportion of the elderly and elderly SSI recipients who fall below this standard. The median is quite similar across the three samples, causing the relative poverty standard to vary by less than $100. For the estimates in panel (c) the standard is $12,764. This is the amount for a single individual; the NRC equivalence scale says that for a family of two adults and two children, this should be increased by a factor of (2 + .5 * 2)^0.7 = 2.16, that is, to $27,540. As noted in our earlier article, the relative poverty standard assesses a larger proportion of the population to be in poverty (22 percent versus the 12.1 percent reported in Table 3). In contrast to the results for absolute poverty rates, the poverty rate for the elderly now exceeds that for the population as a whole, and the poverty rate for elderly SSI recipients rises to 70.7 percent for the adjusted and reweighted sample.

The same calculations using the inclusive version of the data set are shown in part 2 of Table 5. The inclusive income estimates increase estimated median income and thereby increase the poverty standard. However, the estimated poverty rates do not change much at all. We do find that a larger fraction of elderly SSI recipients are estimated to fall in the lowest decile of the income distribution. On the other side of the distribution, between 8.2 percent (inclusive) and 8.6 percent (restrictive) of elderly SSI recipients live in families with total incomes that place their members in the upper 20 percent of the income distribution.

Changes in the Income Distribution, 2002–2005

Both parts of Table 5 are replicated for 2003–2005 in Table A-2. Our text discussion is based on an extract of that data and focuses on comparison of baseline estimates with the final estimates developed with the restricted/reweighted data set and the restrictive and inclusive income-adjustment protocols. We begin with changes in median income over time and the resulting changes in the poverty standard. The standard for all 4 years of our data set is reported in Table 6. To facilitate comparison, we have adjusted the data to 2002 prices using the Consumer Price Index.

Income distributions change slowly, so we do not expect much change over a 4-year interval. By and large, the restrictive income-adjustment procedure produces a relative poverty standard slightly lower than what is computed using the baseline, unadjusted data; the inclusive measure moves the estimated income distribution to the right and raises the standard. Perhaps the most interesting feature is the general decline in the relative standard from 2002–2004, followed by an increase in 2005. Recall that the federal SSI payment for a single individual is indexed for price changes. Annualized, the 2002 monthly individual FBR amounted to $6,540 per year, or nearly 45 percent of the 2002 "final inclusive" relative standard ($14,350).

Medians capture only one feature of the income distribution. Dispersion is relevant as well, especially in the context of relative poverty assessment. Table 7 reports the 90/10 and 80/20 decile cutoff ratios for the total population for each of the 4 years under study. The 90/10 ratio is equal to the ratio of the demarcation point for the 90^th decile in the income distribution to the demarcation for the 10^th decile. The 80/20 ratio is defined in a similar manner, but obviously does not reach as far out on the tails of the distribution.

Four things stand out in these results.

1. Adjustment with administrative data generally reduces estimated dispersion of the income distribution.
2. Estimates based solely on the inclusive income-adjustment protocol generally produce the lowest dispersion.
3. Dispersion as measured by the 90/10 ratio grew over this period, regardless of the income-adjustment protocol followed.
4. Most changes in the distribution of income occur among those belonging to the bottom and top quintiles of the national income distribution. The estimated change in the 90/10 ratio is larger than the estimated change in the 80/20 ratio.

Finally, relative poverty rates for each of the 4 years under study for all persons—the elderly and the subset of the elderly who are SSI recipients—are given in Table 8. Basically, no trends are evident in the general income distribution. The baseline shows some decline for the elderly and for elderly SSI recipients. This is consistent with trends in the official poverty rate shown in Chart 2. However, the adjusted data show little change. As with the data for the official poverty rate, results after adjustment using administrative data provide little evidence of improvement in the prevalence of poverty among elderly SSI recipients using either poverty standard.

Conclusions

In this article, we have applied the experimental procedures developed in our earlier study of the incomes of elderly SSI recipients in 2002 as well as 3 subsequent years of data. In general, the results for 2003–2005 are consistent with 2002. Even given the incomplete match between CPS and administrative records, we have produced an adjusted data set that yields estimates of the prevalence of SSI receipt that are much closer to administrative totals than can be achieved using the standard CPS data set. Unlike what might be inferred from unadjusted CPS data, we see no evidence of significant decline in poverty rates among the elderly or among elderly SSI recipients over this interval.

Several features of this analysis deserve more attention. The difference between our restrictive and inclusive estimates is quite broad generally because of reliance on administrative data alone for the restrictive estimates. For a significant number of persons with an SER match, earnings reported in the CPS are substantially greater than what is recorded in the DER. On balance, the reduction in incorporated earnings for this group under the restrictive protocol almost offsets the addition to income made for those without CPS earnings, but with a DER report. Clearly more thought needs to be given to alternatives for using administrative data, and the sensitivity of the outcomes to procedural variation deserves more thorough investigation. Beyond sensitivity to definition, some investigation of confidence intervals for the many point estimates we have tabulated is essential.

There are noticeable differences between results for 2004 and 2005. The 2005 data are the first collected following the procedural change in obtaining CPS respondent consent for data matching. It is possible that the observed changes are the product of differences between those persons who prior to 2005 would not have been matched and those who would have been captured in the sample had procedures gone unchanged. Of course it is impossible to identify just who would and who would not have consented under the Census Bureau's opt-in interviewer policy. But it would be possible to use the match propensity models estimated for prior years to identify those observations in 2005 that would have been least likely in previous years to have been matched to Social Security administrative data and to use propensity scores to reduce the 2005 sample to a rate consistent with earlier years. The analysis could then be replicated with an eye toward consequences for income distribution estimates obtained using the procedural change adopted with the 2006 CPS/ASEC.

Recently, various groups have shown renewed interest in the recommendations of the NRC for reform of the poverty standard. In March, the Census Bureau announced plans for a "supplemental poverty measure" (SPM) "broadly based" on the NRC recommendations, to be first published in the fall of 2011 (Census Bureau 2010). As the name suggests, at least initially, the new measure will not replace the current poverty standard, but rather provide a broader perspective on both the resources and needs of families and individuals. The Census Bureau's Web site now includes an ingenious table generator for experimenting with alternative equivalence scales and poverty standards, including relative measures based on position in the income distribution. However, aside from differences in top- and bottom-coding, the generator, like the Census Bureau's other experimental analyses, relies on reported amounts of income from sources such as SSI, OASDI, and Temporary Assistance for Needy Families (TANF). A major part of the reform agenda and the modifications incorporated in the SPM involves addition to measures of income from sources such as Supplemental Nutrition Assistance Program (SNAP, formerly the Food Stamp Program) benefits and payments from the Earned Income Tax Credit, which are not now included. Underreporting will need to be addressed as well, possibly through more systematic incorporation of administrative data. Our experience suggests that incorporating administrative data is important, but not easy.

Appendix