Low Levels of Retirement Resources in the Near-Elderly Time Period and Future Participation in Means-Tested Programs

by
Social Security Bulletin, Vol. 70, No. 1, 2010

This article describes the de facto standards of low income and resources reflected in the eligibility standards of the largest means-tested programs that serve the elderly and then applies these standards to a near-elderly cohort. Through juxtaposing retirement resources in the near-elderly time period with program participation in the elderly time period, the author indirectly examines some of the changes between the two time periods that could affect program eligibility, including spend-down of resources and marital dissolution. Retirement resource levels are estimated using the Survey of Income and Program Participation, and subsequent participation in one of the means-tested programs—Supplemental Security Income (SSI)—is examined using matched administrative records. Although spend-down of resources is shown to occur for only 8.7 percent of eventual SSI program participants, it is more common in the part of the near-elderly population that faces the greatest incentive to decrease resource levels.


The author is with the Division of Policy Evaluation, Office of Research, Evaluation, and Statistics, Office of Retirement and Disability Policy, Social Security Administration.

Acknowledgments: The author is grateful for the helpful comments of numerous colleagues, especially Glenn Springstead and Michael Wiseman.

The findings and conclusions presented in the Bulletin are those of the authors and do not necessarily represent the views of the Social Security Administration.

Introduction

Selected Abbreviations
DI Disability Insurance
FRA full retirement age
SIPP Survey of Income and Program Participation
SNAP Supplemental Nutrition Assistance Program
SSI Supplemental Security Income

This article examines the segment of a near-elderly cohort that has low retirement resources to answer three research questions. First, who are the people who have very low levels of retirement resources in the near-elderly time period? These individuals are described in terms of demographics, current financial situation, and lifetime labor force attachment. Second, what is the relationship between having low retirement resources in the near-elderly time period and participation in the Supplemental Security Income (SSI) program upon reaching age 65? Third, what changes in the years just before turning age 65 can affect future eligibility in means-tested programs? Two potential changes are examined: spend-down of resources and marital dissolution.

Rather than using current-period income, a broad measure of retirement resources is used, which includes wealth holdings and the potential Social Security benefit for which the person would be entitled upon claiming. The focus here is on the population whose levels of income and resources would make them eligible for one or more of the means-tested programs that serve the elderly, if they were otherwise eligible.1

I establish unified eligibility criteria under which an individual could be financially eligible for any of the three largest means-tested programs. The three programs are considered together because they may create joint behavioral incentives. The unified criteria represent upper-bound eligibility measures for thresholds that vary across the three programs and, in some cases, also across states.

These thresholds are used to examine potential future financial eligibility for means-tested programs among those in a near-elderly cohort. This cohort has generally not reached the age of categorical eligibility for these programs,2 but financial eligibility is evaluated with an eye toward future eligibility and possible participation. Eligibility in the near-elderly time period is compared with later receipt of benefits from one of the three programs, the SSI program.

I reach several broad conclusions about the near-elderly population with low levels of retirement resources. Individuals in this group have had low labor force attachment over the course of their lifetimes, both in terms of the frequency and level of earnings. For about half, the lifetime earnings patterns were affected by disability. Further, the near-elderly population with low retirement resources has different marital histories than the remaining near-elderly population. This affects well-being in both the near-elderly and elderly time periods. Not only can the presence of a spouse affect family income and poverty status in the near-elderly time period, but the earnings history of a spouse can also increase the potential Social Security benefit.

Among those with low retirement resources in the near-elderly time period, the rate of SSI payment receipt upon reaching age 65 varies by a number of financial factors. Participation rates are far higher for those with very low levels of potential Social Security benefit amounts or low resource level amounts in the current period, even when compared with other people who would be financially eligible for SSI. Participation rates are even higher for people who additionally lack a defined benefit pension. These trends indicate that many people who would be eligible for means-tested programs upon reaching age 65 already have very low levels of retirement resources in the near-elderly time period. In fact, among the group that receives SSI payments upon reaching age 65, the vast majority are already financially eligible in the near-elderly time period.

I examine two kinds of behavioral changes that could occur between the near-elderly time period and age 65 and that may be of interest to policymakers: spend-down of resources, which could affect program eligibility; and divorce, which could affect both program eligibility and the potential Social Security benefit amount. Of these two possible behavioral changes, I demonstrate, using a counterfactual illustration, that the potential reach of changes in marital status is far greater than the potential reach of spend-down of resources in terms of the numbers of near elderly who could gain eligibility through behavioral changes. This is particularly true of near-elderly women. In contrast to this hypothetical result, the findings indicate that gaining eligibility is far more common through spend-down of resources. This occurs about four times as frequently as gaining eligibility through divorce.

Confirming earlier studies, I find that declining resource levels are common among the very narrow part of the near-elderly population that could gain SSI eligibility through a modest reduction in resource levels. This is observed among the near elderly who participate in SSI upon reaching age 65, and it is not observed among people with similar resource levels who do not.

The next section of the article presents background issues that are relevant to the research questions, followed by a description of how the Survey of Income and Program Participation (SIPP) is employed to estimate program eligibility and how the matched administrative data of the Social Security Administration (SSA) are employed to examine SSI participation. Finally, the results are presented.

Background

Three aspects of program eligibility are discussed in this section, including the thresholds that are relevant to means-testing, using the thresholds to estimate eligibility for households in a sample, and the incentives provided by means-testing. Each aspect is discussed in turn.

Program Thresholds

The three largest means-tested programs serving the elderly are Medicaid, the Supplemental Nutrition Assistance Program (SNAP),3 and SSI. The Medicaid program subjects the largest number of elderly households to means-testing, followed by SNAP and SSI (Chen and Lerman 2005). The maximum income and resource levels that allowed for eligibility in these three programs in 2001 are given in Table 1.4 Along with the nominal income thresholds, the table also presents the effective income thresholds, which sum the income thresholds and the primary unearned income exclusion.5,6

Table 1. Maximum income and resource levels for eligibility for the means-tested programs that serve the elderly, 2001
Program Individual threshold Couple threshold Primary unearned income exclusion Effective individual threshold a Effective couple threshold Ratio of couple to single
Monthly income thresholds
Medicaid 632 947 20 b 652 967 1.48
SNAP c,d 716 968 134 850 1,102 1.30
SSI 593 933 20 613 953 1.56
Monthly resource thresholds
Medicaid 2,232 3,247 . . . . . . . . . 1.45
SNAP c 3,000 3,000 . . . . . . . . . 1.00
SSI 2,000 3,000 . . . . . . . . . 1.50
SOURCES: Bruen, Wiener, and Thomas (2003); Department of Agriculture, Food and Nutrition Service; and the Social Security Administration.
NOTES: See the Technical Appendix for more information.
. . . = not applicable.
a. The effective thresholds are the sum of the income thresholds and the primary unearned income exclusions.
b. The Medicaid income exclusion figure assumes that SSI eligibility is the path to Medicaid eligibility. Variation in income exclusions corresponding to other eligibility paths is not considered here.
c. SNAP was previously called the Food Stamp Program until October 2008. The author uses the term SNAP even though the data used in this analysis refer to the Food Stamp Program.
d. SNAP income and income exclusion figures assume that the individual or couple are the only household members.

Table 1 represents a simplified view of program eligibility standards. The primary source of complexity is state variation in eligibility standards. For Medicaid, there is state variation in both the income and resource thresholds. For SSI, variation results from the differing thresholds of state supplemental programs. For SNAP, both the income and resource thresholds are uniform across states.7 Where there is state variation for any of the three programs, the thresholds are summarized by averaging across states using the number of elderly SSI recipients as weights. Thus, the figures in the table represent the thresholds that are relevant to the average elderly SSI recipient.

Another source of complexity is the links in eligibility across programs. State Medicaid programs are required to cover SSI recipients, and SNAP has automatic eligibility for households that are entirely comprised of beneficiaries of certain other programs. However, some states have Medicaid thresholds that are less restrictive than SSI thresholds, and some have thresholds that are more. Because the goal is to define thresholds under which an applicant could be eligible in any of the three programs, the more restrictive thresholds are not considered, while the less restrictive thresholds are included in the averages presented in the table.8 Thus, the figures represent upper-bound estimates of the thresholds that are relevant to near-elderly individuals.

One salient feature of Table 1 is that the income and resource thresholds have similar orders of magnitude across programs. In this respect, the eligibility requirements are similar for the three programs. Although the thresholds are similar, some differences are notable. The income thresholds are highest for SNAP. Further, the differences are exacerbated when the effect of the primary unearned income exclusions are considered. The SNAP income exclusion is substantially larger than that for the other programs. This leads to the higher effective thresholds shown in the table.

The resource thresholds are more difficult to compare. The SSI thresholds are $2,000 for individuals and $3,000 for couples. This standard is also followed by Medicaid in the majority of states; however, some states have higher "poverty-related" resource thresholds. Of the six states that had higher resource thresholds in 2001, two populous states,9 Florida and Pennsylvania, used thresholds that were 2.0 or 2.5 times higher (Bruen, Wiener, and Thomas 2003). This has a disproportionate effect on the resource threshold figures given in Table 1, which are weighted averages. By contrast, SNAP uses a resource threshold of $3,000 for households containing an elderly (aged 60 or older in this case) member.

Considering all the thresholds together, SNAP generally provides the highest income and resource levels under which a person could be eligible for benefits in any of the three programs. The exception is the resource threshold for couples for which Medicaid is the highest. Thus, the highest of these levels are used to define low levels of income and resources for the indicators used in this article: $850 and $1,102 of monthly income for individuals and couples, respectively; and $3,000 and $3,247 of resources for individuals and couples, respectively.10

Measurement of Eligibility and Potential Benefits

Several difficulties arise when using a sample survey to estimate program eligibility and potential benefit amounts. One issue arises from the possibility that reported income amounts (and, thus, estimated eligibility and benefit amounts) may not be exogenous with respect to program application. For example, program participation or intent to participate in the future may lead to withdrawal from the labor force or a decline in work hours. Neumark and Powers (2000) find empirical evidence that SSI program rules lead to a decline in labor supply for men aged 60–64. Further, the SSI program requires applicants to file for all other kinds of benefits for which they are potentially eligible, including Social Security and pension benefits. Thus, the event of filing for SSI payments could lead to changes in income and consequent changes in eligibility and benefits for SSI, SNAP, and Medicaid.

Another difficulty when using sample surveys to estimate eligibility arises from the timing of the receipt of earned and unearned income. Variation in income from month to month can lead to changes in program eligibility and benefit amounts. This is partially accounted for by SSI and SNAP rules, which disregard $30 of "irregular" or "infrequent" income per calendar quarter. Still, estimates of program eligibility and benefit amounts depend on the month of observation in the survey and the ability of the researcher to identify irregular income. Farrell and others (2003) examine the relationship between monthly variation in income and participation in SNAP, and Elder and Powers (2007) explore this relationship for the SSI program. In this study, the relationship is complicated by the fact that eligibility is estimated in the near-elderly period, and program participation is observed in a subsequent period.

Elder and Powers (2004) address these difficulties by using only Social Security benefit amounts when measuring income and the corresponding program eligibility among the elderly. They discuss several theoretical advantages of using this measure. First, it minimizes endogeneity problems because Social Security benefit amounts are not affected by the claiming of means-tested benefits. Second, it removes measurement error that is due to irregular or infrequent income. And third, this measure reduces recall biases because Social Security benefits are constant across months except for cost-of-living increases.

In addition, Elder and Powers (2004) present empirical evidence that their measure leads to less measurement error. To accomplish this, they compute SSI payment amounts using all reported income and also using only Social Security income, and then they compare these figures with the reported SSI payment amounts. Using all reported income, the estimated SSI payment differs from the survey-reported amount by $247, on average. When using only Social Security income, by contrast, the difference is only $109.11 Although this difference is notable, the value of this evidence is diminished by the fact that Elder and Powers use a very imprecise benefit simulator; other authors are able to simulate benefit amounts to within $1 on average (Davies and others 2001/2002).

The issues previously discussed also generally apply to measuring countable resources. As with income, resources may not be exogenous with respect to program participation. Neumark and Powers (1998) find empirical evidence that people who are likely to participate in SSI reduce their savings as they approach age 65. Unfortunately, there is no equivalent solution to the one discussed earlier for income. This highlights the importance of spend-down issues between the near-elderly and elderly time periods.

Program Incentives

The literature about the incentives associated with means-testing has focused on effects on savings behavior and labor force participation. Savings behavior is the greater concern partly because labor force participation is less common among the elderly and near elderly. Also, means-tested programs treat resources more punitively than income. For example, if the resources of the elderly were to be annuitized, Radner (1990) estimates that the ratio of resource holdings to annual annuity income would be about 15 to 1. Thus, a resource holding that is near the thresholds, $3,000, for example, translates into annual annuity income of $200, or $17 per month. This is far less than the comparable income threshold.12

Although some research addresses the effects of resource testing on savings behavior over the entire life cycle,13 the most relevant research focuses on behavior near the ages when people gain categorical eligibility for the programs, that is, near the full retirement age (FRA).14 For the Medicaid program, Gruber and Yelowitz (1999) find that eligibility has a negative association with savings. Further, this negative association is exacerbated in states where Medicaid eligibility involved a resource test. For the SSI program, Neumark and Powers (1998) find reduced savings among likely beneficiaries approaching the traditional Social Security FRA. For SNAP, I am unaware of any similar research. In this article, I examine resource spend-down between the near-elderly and elderly time periods for future SSI recipients.

Means-tested programs may also provide incentives related to the forming and dissolution of marriages. For example, in many cases, the SSI program would provide a higher benefit to two unmarried adults than to two otherwise identical adults who are married (Balkus and Wilschke 2003). Also, the methods for determining Social Security benefit amounts have marriage and divorce incentives inherent in them, which will be discussed later.

Methods

This study uses a 1996 SIPP subsample of near-elderly people who have not reached the traditional FRA, and then examines SSI participation behavior in the first 6 months after reaching the traditional FRA. Observation of participation at later ages is not possible for all of this sample. Also, similar matched data for Medicaid and SNAP are not currently available.

The analysis subsample was born from November 1931 through March 1941. Thus, the subsample represents a prewar cohort (at least from the American perspective). As of the reference period in November 1996, this subsample was 55½ to 65 years of age. Correspondingly, the subsample reached age 65 from November 1996 through March 2006. Because categorical eligibility for SSI based on age occurs at age 65, the first payment on this basis could be received the following month. Thus, the window of potential SSI payment receipt, which is referred to here as the first 6 months after reaching age 65, is from December 1996 through May 1997 for the oldest in the subsample and from April 2006 through September 2006 for the youngest in the subsample.

This study defines low retirement resources as income and resources15 in November 1996 that are less than the highest income and resource thresholds presented in Table 1, that is, the near-elderly time period. I follow Elder and Powers (2004) in counting only potential Social Security income and treating other income as irregular. In this study, there is no way to estimate other income for the elderly time period. The measure of resources follows the concept of "countable resources" used in the SSI program and is defined more precisely below.

Although this is an individual-level analysis, program eligibility is evaluated on a couple basis if a spouse exists. Therefore, the potential Social Security benefit and countable resources of a person's spouse are included in the measures that are compared with the thresholds for couples given in Table 1.16

I am also able to improve the income measure used by Elder and Powers (2004) by using Social Security income amounts from administrative data rather than the self-reported amounts the authors used. In addition, I use the Summary Earnings Record of earnings histories to calculate potential Social Security benefit amounts and corresponding SSI payment amounts for the entire sample, rather than just for observed beneficiaries.

The potential Social Security benefit amount is a snapshot as of November 1996. From this time to the time of the start of Social Security benefits, further wage- and/or price-indexing would be applied for 0 to 9 years; however, the potential benefit is measured as of the reference period. As a result of ignoring this additional indexing, there will be some false positive indications of low income, but this will lead to no false negative indications (because the additional indexing can only increase the potential Social Security benefit amount).17 In other words, the low-income measure represents an upper-bound measure of potential income eligibility.

The assumptions underlying the calculation of the potential Social Security benefit amount are compatible with an upper-bound interpretation of the measure. The calculation assumes that Social Security benefits begin at age 62 and applies the corresponding early retirement reduction.18 Powers and Neumark (2003) have shown that there is little incentive for people who expect to be eligible for means-tested programs in retirement, particularly SSI, to delay the start of Social Security benefits beyond the earliest possible date. Further, the calculation of the benefit amount is based on the assumption that observed marriages meet the requirements for spousal benefits.19

The marital status assumption is relaxed later in the analysis when potential Social Security benefit amounts are also calculated for the case of a hypothetical divorce. The calculation illustrates the effect of a couple becoming two individuals with no other changes. This calculation also assumes that marriages meet the requirements for spousal benefits.

The countable-resources measure is also a snapshot as of November 1996. The means-tested programs generally count liquid resources such as checking account balances, savings bonds, 401(k) and KEOGH accounts, stocks, bonds, money market accounts, vehicles not needed for employment, and the cash value of life insurance policies. These items are included in the low countable-resources measure, except the cash value of life insurance (which is not measured in the 1996 SIPP). The value of the primary residence is not counted as a resource. For the case of a hypothetical divorce, the resources of the couple are divided. Further, liabilities are generally not part of the countable-resource measure. Although resources could be used to pay off liabilities, it is not assumed that this occurs.

Results

This section progresses from a description of the population with low retirement resources in the near-elderly time period to an analysis of the relationship between low retirement resources in the near-elderly time period and participation in the SSI program upon reaching age 65. Subsequently, I examine behavioral changes between the near-elderly and elderly time periods that could affect eligibility for means-tested programs.

Population with Low Retirement Resources

The population with low retirement resources is defined as the group with both low potential Social Security benefits and low countable resources. However, it is also illuminating to separately examine the groups with low potential benefits only and low countable resources only. In this section, I examine the demographic and economic differences across groups based on cross-sectional data from the SIPP. In addition, the analysis is extended by using matched administrative data to examine labor force and program participation20 over the lifetime.

The proportions of people who have low potential Social Security benefits or low countable resources in November 1996 are shown in the first panel of Table 2. The groups are presented in mutually exclusive categories including low potential benefits only, low countable resources only, both low potential benefits and low countable resources, and neither low potential benefits nor low countable resources (the comparison group). Approximately half of the sample cohort is in the "neither condition" category. Within the remaining sample, low potential benefits are more prevalent than low countable resources. The proportion of people with low potential benefits, including both the first column (low potential benefits only) and the third column (both conditions), is around 45 percent. By comparison, the proportion of people with low countable resources, including both the second column (low resources only) and the third column (both conditions) is around 30 percent. Accordingly, by these measures, the more restrictive measure is low countable resources.

Table 2. Sample characteristics and means of variables, by category, November 1996
Characteristic Low potential benefit only Low resources only Both conditions Neither condition
Sample characteristics
N (unweighted) 1,209 386 1,361 2,420
Weighted percentage of population 22.4 7.3 22.8 47.4
Population (millions) 4.3 1.4 4.4 9.2
Demographic variable means
Age (years) 60.5
(2.7)
60.2
(2.7)
60.1
(2.7)
60.0
(2.7)
Hispanic (%) 6.0
(23.8)
7.1
(25.4)
14.9
(35.7)
2.8
(16.6)
Black (%) 7.1
(25.7)
13.9
(34.6)
24.9
(43.2)
3.6
(18.6)
Women (%) 61.2
(48.8)
38.5
(48.7)
58.8
(49.2)
46.0
(49.9)
Married (%) 58.6
(49.3)
89.0
(31.3)
34.7
(47.6)
93.1
(25.3)
Widowed (%) 16.1
(36.7)
a
. . .
18.9
(39.1)
1.3
(11.3)
Divorced (%) 17.5
(38.0)
7.2
(25.8)
29.4
(45.6)
4.0
(19.6)
Never married (%) 6.2
(24.2)
a
. . .
10.4
(30.5)
1.1
(10.4)
Born outside United States (%) 11.0
(31.3)
7.4
(26.2)
18.1
(38.5)
5.1
(21.9)
Number of years in United States 58.3
(7.5)
58.3
(7.3)
56.8
(8.5)
58.8
(6.0)
Earnings variable means
Average indexed monthly earnings 957
(702)
1,924
(1,067)
726
(639)
2,064
(1,228)
Number of years with earnings 25.4
(12.1)
33.7
(11.0)
23.2
(12.9)
33.4
(11.4)
Highest annual earnings (wage-indexed) 27,667
(13,766)
39,107
(16,481)
21,607
(12,965)
41,806
(17,719)
Earnings above average (years) 6.0
(7.8)
18.2
(13.4)
3.4
(6.4)
18.9
(14.3)
Earnings above the taxable maximum (years) 1.4
(3.1)
7.3
(8.6)
0.8
(2.5)
9.7
(11.3)
Total family income (current month) 3,498
(3,792)
3,303
(2,126)
1,857
(1,764)
5,427
(5,016)
Poverty indicator (current month, %) 9.5
(29.4)
7.3
(26.1)
31.7
(46.6)
2.0
(14.1)
Program participation variable means (%)
SSI participant upon reaching age 65 1.8
(13.2)
0.8
(8.9)
19.8
(39.9)
0.0
(2.9)
SSI recipient in current month 1.0
(10.1)
0.1
(3.6)
16.6
(37.2)
0.0
(1.7)
Ever a disability beneficiary—either DI or SSI b 13.5
(34.2)
23.4
(42.4)
47.2
(43.4)
9.1
(28.8)
Ever a DI beneficiary c 10.8
(31.1)
22.3
(41.7)
25.3
(43.5)
8.9
(28.5)
Ever an SSI recipient  d 4.4
(20.4)
3.8
(19.2)
38.1
(48.6)
0.6
(8.0)
Social Security beneficiary (current month) 27.0
(44.4)
34.8
(47.7)
35.7
(47.9)
22.2
(41.6)
SOURCE: 1996 SIPP matched to Social Security administrative records.
NOTES: Sample members are aged 55–64.
Standard deviations are in parentheses; . . . = not applicable.
a. Estimate not shown because of inadequate sample size.
b. Data obtained from the Master Beneficiary Record and Supplemental Security Record.
c. Data obtained from the Master Beneficiary Record.
d. Data obtained from the Supplemental Security Record.

The characteristics of the four analysis groups are given in the remaining panels of Table 2. The central result is that people with low potential Social Security benefit amounts are clearly different from people who do not have low potential benefit amounts. For the demographic characteristics given in Table 2, there is a general pattern that the two columns representing low potential benefit amounts (the first and third columns) are similar to each other, and both are different from the other columns. For example, the low potential benefit category and the both conditions category have the highest proportion of women.21 Similarly, these two categories have relatively high proportions of persons in the three unmarried categories, including widowed, divorced, and never married.

There are also differences in immigration status. The groups with low potential benefit amounts and the both conditions category have a higher proportion of persons born outside the United States. Immigration is related to earnings patterns through a number of social and economic factors (see Bean, Stevens, and Van Hook (2003) for a discussion). One factor is that the date of immigration has a mechanical effect on the potential benefit amount because it determines the number of potential years of covered earnings in the United States. Only small differences of this sort are observed in this particular sample, however. A measure of the potential years of covered employment is given in Table 2 as the "number of years in the United States." Although the difference between the neither condition and the both conditions categories is statistically significant, the difference is only 2 years.

There are also differences in earnings histories as measured in Social Security's administrative records. This is shown by the differences in the average indexed monthly earnings (AIME), which is a primary input into Social Security's benefit formula.22 The fact that the two categories with low potential benefits have lower AIMEs is true by definition; however, the composition of the differences is interesting. One question follows: Are the differences the result of lower earnings or a greater dispersion of earnings? Further, a greater dispersion of earnings could be due to less frequent employment or less consistent earnings levels.

The means of the earnings variables given in Table 2 show that low potential Social Security benefits are due to a combination of low earnings and low frequency of earnings. The earnings level is illustrated by the highest annual earnings over the lifetime, the number of years with earnings above average, and the number of years with earning above the maximum level that is subject to the Social Security payroll tax. The differences in means for these variables are all statistically significant. For people with low potential Social Security benefits, the most common experience is to have an earnings history where earnings peak in the vicinity of $20,000–$30,000 in 1996 dollars. This is shown in the two panels on the left side of Chart 1. In comparison, for people who do not have low potential earnings, the most common experience is to have an earnings peak around $60,000 or higher (see the two panels on the right side).

Chart 1.
Distribution of highest annual earnings over the lifetime for groups with low potential Social Security benefit amounts or low countable resources
Chart 1 has histograms showing the highest annual earnings over the lifetime for the four analysis groups used in the text. For people with low potential Social Security benefits, the most common experience is to have an earnings history where earnings peak in the vicinity of $20,000-30,000 in 1996 dollars. In comparison, for people who do not have low potential earnings, the most common experience is to have an earnings peak around $60,000 or higher.
SOURCE: 1996 SIPP matched to the Summary Earnings Record.
NOTE: Sample members are aged 55–64. Earnings are wage-indexed to 1996 levels.

The average frequency of earnings is shown by the number of years with positive earnings (Table 2). This measure is also lower for people with low potential Social Security benefits. Chart 2 shows that there is no typical experience for this group; that is, the distribution of the number of years with positive earnings resembles a uniform distribution (in the two panels on the left side). This is compared with a highly skewed distribution for people who do not have low potential benefits (the two panels on the right side). The most common experience for this group is to have 40 or more years of positive earnings.

Chart 2.
Distribution of number of years with positive earnings for groups with low potential Social Security benefit amounts or low countable resources
Chart 2 shows the distribution of the number of years with positive earnings for the four analysis groups used in the text. For the low potential benefits and both conditions groups, the distribution resembles a uniform distribution, whereas for the low countable resources and neither condition groups, the distribution is skewed to the right. The most common experience for these latter groups is to have 40 or more years of positive earnings.
SOURCE: 1996 SIPP matched to the Summary Earnings Record.
NOTE: Sample members are aged 55–64.

The results shown so far illustrate that the group with low potential Social Security benefits differs from the other groups by a number of demographic and earnings history measures. Differences within this group are examined here. Among persons with low potential benefits, what distinguishes those who have low retirement resources in general—that is, those who have low potential benefits and also low countable resources—from those who do not? One can point to disability, marital history, and the income of other family members as major factors.

Disability is a major factor in having low retirement resources in the near-elderly time period, as shown in the last panel of Table 2. Within the group with low potential benefits, there is a notable difference in disability history. Those who also have low countable resources have a much higher rate of having received disability benefits in the past (either from the DI or SSI programs). The 47.2 percent who have received disability benefits from the Social Security Administration breaks down into 25.3 percent who have received DI and 38.1 percent who have received SSI (some have received both). Also, in the current month alone, 16.6 percent received SSI. This is roughly comparable to the percentage that will receive SSI upon reaching age 65 and suggests that disability during the working ages influences participation in SSI after reaching age 65 (when disability is not required for categorical eligibility). In fact, 61.4 percent of those who participate in SSI upon reaching age 65 were previous SSI participants.

Marital history is also a factor in having low retirement resources in the near-elderly time period. Within the group with low potential benefits, the group that also has low countable resources is less often currently married and more often divorced or never married.

Other family income is also a factor. Within the group with low potential benefits, the group that also has low countable resources has much lower total family income, which leads to a much higher poverty rate for this group (31.7 percent compared with 9.5 percent for the group with only low potential benefits).

In summary, the group with low potential Social Security benefits in the near-elderly time period differs from the comparison groups by a number of demographic and economic variables. This group's lower potential Social Security benefit is due to both lower levels of earnings as well as lower frequency of earnings. For the group that also has low countable resources in the near-elderly time period, disability, marital history, and other family income are highlighted as contributing factors.

SSI Participation After Age 65

In the near-elderly cohort analyzed in this study, an estimated 972 thousand out of the total cohort of 19.3 million participated in SSI within 6 months of reaching age 65.23 Of these eventual participants, 98.1 percent had a low potential benefit at the time of the survey during the preretirement period.24 Thus, low potential benefits effectively define the universe of possible SSI recipients upon reaching the age for categorical eligibility, but around 2 percentage points of the eventual participants had higher potential benefits as measured in the near-elderly time period. This is due to changes in status between the near-elderly period and the period after reaching the traditional FRA. For countable resources, changes in status are slightly more common; 91.3 percent of eventual participants had low countable resources at the time of the survey during the preretirement period.

The participation rates for the four different population groups are given in Table 2. For the group with both low potential benefits and low countable resources, 19.8 percent are observed to eventually participate in SSI, which is considerably higher than the other categories. These figures are not comparable to other participation rate estimates because the measures used in this study are based on potential eligibility for any of the three largest means-tested programs, and SSI is not the most restrictive of the three, as shown in Table 1. Thus, some people who are ineligible for SSI are included in the denominator of this ratio. Indeed, estimated SSI participation rates for the elderly are considerably higher than the figures given here.25

For the groups with either low potential benefits only or low countable resources only, approximately 1–2 percent are observed to participate in SSI upon reaching age 65. These two groups represent different changes in status over the analysis period. The group with low potential benefits only has low potential benefits, but not low countable resources. Thus, there must be spend-down of resources for any members of this group to eventually be eligible for SSI. By contrast, the group with low countable resources only has low countable resources, but not low potential benefits. Thus, the potential benefit must be reduced to attain SSI eligibility, which can only occur through a change in marital status. For the population comparison group (neither low potential benefits nor low countable resources), the percentage that eventually participates in SSI is very low.

More information about specific levels of potential benefits and SSI participation is shown in Chart 3. The chart shows participation rates for SSI at different levels of potential Social Security benefits, while also showing the distribution of those potential benefits as a histogram. Thus, it is possible to view the rate of SSI program participation at each potential benefit level, while simultaneously viewing the prevalence of that potential benefit level in the general population. In Chart 3, the people with zero potential Social Security benefits have a relatively high eventual SSI participation rate, between 35 and 40 percent. The rate drops rapidly as the potential benefit rises. A rising potential Social Security benefit corresponds with a declining expected SSI payment amount; thus, the chart confirms previous research that finds an inverse relationship between SSI participation and the SSI payment amount.26

Chart 3.
Rate of receipt of SSI payments upon reaching age 65, by the size and distribution of the potential Social Security benefit amounts for the individual and spouse
Chart 3 shows participation rates in SSI at different levels of potential Social Security benefits while also showing the distribution of those potential benefits as a histogram. The people with zero potential Social Security benefits have a relatively high eventual SSI participation rate, between 35 and 40 percent. The rate drops rapidly as the potential benefit rises and reaches zero around a potential benefit of $1,000.
SOURCE: 1996 SIPP matched to Social Security administrative records.
NOTE: Lowess smoothing is used to show the mean value of program participation at each value of the potential benefit amount. The procedure calculates a mean at each value using a sampling of other data points where the other data points are weighted by the distance from the point being measured. A bandwidth of 0.1 is used, meaning that 10 percent of other points are used. A low bandwidth forces the graph to closely resemble the data. Also, the lowess procedure is very localized and, thus, minimizes the effects of changes in the tails of the distributions on the localized means. The mean of the presented means is forced to equal the observed overall mean.

At a potential Social Security benefit level of around $1,000, the SSI participation rate declines to nearly zero. By comparison, the most prevalent potential Social Security benefit amounts are higher than this. As a result, the SSI participation rate among the majority of the distribution and, thus, the population it represents is zero.

As a comparison, the same information is shown for countable resources in Chart 4. Unlike with income, the expected SSI payment amount does not decline as countable resources increase except for a complete loss of eligibility and benefits at the point where the resource threshold is reached. Nevertheless, the participation rate declines as countable resources increase, but the rate of decline is much less steep. Because Chart 4 shows wealth using a logarithmic scale, the full extent of the slower rate of decline is not visible. Still, the chart shows results that are similar to the results for the potential Social Security benefit shown in Chart 3: The SSI participation rate declines rapidly at a moderate levels of retirement resources. By the level of the applicable programmatic thresholds, the participation rate is very low, in this case less than 5 percent. For the majority of the population, those who have wealth above—and frequently far above—these thresholds, the SSI participation rate is zero.

Chart 4.
Rate of receipt of SSI payments upon reaching age 65, by the level and distribution of countable resources in the near-elderly time period
Chart 4 shows participation rates of SSI at different levels of countable resources while also showing the distribution of those resources as a histogram. In contrast to Chart 3, countable resources are shown using a log scale. The people with zero countable resources have a relatively high eventual SSI participation rate, between 20 percent and 25 percent. The rate drops rapidly as countable resources rise and reaches zero around a resource level of $10,000.
SOURCE: 1996 SIPP matched to the Supplemental Security Record.
NOTE: Reported values of wealth of zero are shown as ln(1) or zero in the chart. Lowess smoothing is used to show the mean value of program participation at each value of the log of countable wealth. The procedure calculates a mean at each wealth value using a sampling of other data points where the other data points are weighted by the distance from the point being measured. A bandwidth of 0.1 is used, meaning that 10 percent of other points are used. A low bandwidth forces the graph to closely resemble the data. Also, the lowess procedure is very localized and, thus, minimizes the effects of changes in the tails of the distributions on the localized means. The mean of the presented means is forced to equal the observed overall mean. The line on the far left of the mean line represents uncertainty about the value at the endpoint.

Although income and countable resources are the criteria for the means-tested programs, certain resources that are generally not countable for means-tested programs could become income sources during retirement. Two potential income sources are also considered here: (1) defined benefit pensions and (2) home ownership—a potentially meaningful source of income for low-income elderly through home sales or reverse mortgages.

Specifically, I examine the group of near-elderly people who have low countable resources, no defined benefit pension, and no home ownership. This group and the prevalence of the lack of components of wealth are shown by potential Social Security benefit level in Chart 5. The solid line represents the percentage of individuals with low levels of countable resources at each potential benefit value. At a potential benefit of zero, around 80 percent of the people also have low countable resources. The proportion gradually drops to zero around the potential benefit levels that are the most common according to the histogram distribution presented in the chart. Thus, a meaningful proportion of the population is subject to some risk of having low countable resources. Chart 5 also shows the percentages of individuals who have no defined benefit pension27 and no home ownership at various potential benefit levels. All three measures are then combined into the dot and dash line, which represents the proportion of individuals who have low countable resources and neither a defined benefit pension nor home ownership—the other two kinds of resources. At a potential benefit of zero, the proportion that has few resources by this expanded measure is around 60 percent. This proportion drops to 20 percent at a potential benefit of around $800, and then to zero at a potential benefit of around $1,400. Thus, although all the people who have very low retirement resources by this particular definition have potential Social Security benefits of less than $1,400, not all people with potential benefits below this level are without resources.

Chart 5.
Absence of resource holdings of the individual and spouse and distribution of potential Social Security benefit amounts, by type of resource
Bar chart with line overlays with text description below.
Show text description

Text description for Chart 5.
Absence of resource holdings of the individual and spouse and distribution of potential Social Security benefit amounts, by type of resource

Chart 5 has lines showing the percentages that lack certain kinds of resources at different levels of potential Social Security benefit amounts. There are lines for having low countable resources, no defined benefit pension, no home ownership, and all three of these conditions. At nearly all potential Social Security benefit amounts, the highest percentage is not having a defined benefit pension. The percentage having low countable resources is generally lower than this, and the percentage having no home ownership is generally lower than that. The percentage with all three of these conditions is generally lower than all the other percentages.

SOURCE: 1996 SIPP matched to the Summary Earnings Record.

Within the group with very few resources by this expanded definition, a higher percentage eventually participate in SSI than for the group with only a low potential benefit, as shown in Chart 6. The participation rate for the group with low countable resources is shown at differing potential Social Security benefit levels by the solid line. The highest participation rate, corresponding to a potential benefit of zero, is around 20 percent. This jumps to more than 60 percent when the restriction of no defined benefit pension is added. Thus, at a potential benefit of zero, those who also have few resources in savings and pensions have a higher participation rate (around 60 percent) than those with only a zero potential benefit (around 40 percent; see Chart 3). Adding an additional component of retirement resources—the restriction of no home ownership—does not raise the participation rate further.28 This corresponds to previous studies that have not found a statistically significant correlation between home ownership and SSI participation.29

Chart 6.
Rate of receipt of SSI payments upon reaching age 65 for groups with the absence of three types of resource holdings in the near-elderly time period, by the potential Social Security benefit amounts of the individual and spouse and the distribution of those amounts
Bar chart with line overlays with text description below.
Show text description

Text description for Chart 6.
Rate of receipt of SSI payments upon reaching age 65 for groups with the absence of three types of resource holdings in the near-elderly time period, by the potential Social Security benefit amounts of the individual and spouse and the distribution of those amounts

Chart 6 shows eventual SSI participation rates for groups with a lack of certain types of resources including low countable resources, low countable resources as well as no defined benefit pension, and these two conditions plus no home ownership. For the group with low countable resources, the participation rate starts around 20 percent at a potential Social Security benefit of zero and drops to nearly zero around a potential Social Security benefit level of around $1,000. For the group that has both low countable resources as well as no defined benefit pension, the participation rate is higher at all nonzero points. For example, at a potential Social Security benefit of zero, the participation rate is around 60 percent for this group. Adding an additional restriction of no home ownership does not raise the participation rate further. In fact, it is lower at most points of the distribution. For example, the group with all three conditions has a participation rate around 50 percent at a potential Social Security benefit of zero.

SOURCE: 1996 SIPP matched to Social Security administrative records.

Changes in Financial Eligibility Status Over Time: Resource Spend-Down and Divorce

The group that has both low potential Social Security benefits and low countable resources in the preretirement period is financially eligible for one or more of the means-tested programs in the current period; however, eligibility during retirement (when members of this group would also have categorical eligibility) is of more interest. Consequently, changes between the current period (the survey reference month, November 1996) and the time of reaching the traditional FRA are also of interest. The focus here is on two kinds of changes that could make people who are not eligible in the current period eligible when elderly. First, resources could be decreased below the applicable resource thresholds; and second, marital dissolution could lead to a reduction in the potential Social Security benefit amount and could bring income below the applicable income thresholds.

Resource spend-down. Previous research has provided limited evidence that the SSI program provides a savings disincentive as people approach the traditional FRA. In this section, I examine what observed program participation reveals about changes in resource levels during the near-elderly years. Specifically, I examine how much resources must have fallen between the current period and the traditional FRA for observed participants in order to obtain SSI eligibility.

One way to address the effects of savings disincentives is by calculating the frequency of the resource spend-down that is implied by comparing the current period with eventual SSI participation. To this end, the distribution of countable resources in the current period for people who eventually participate in SSI, but are not recipients in the current period, is shown in Table 3. The programmatic thresholds, given in Table 1, fall between the 75th percentile ($503) and the 90th percentile ($5,802) of this distribution. More specifically, 13.1 percent of current nonrecipients who eventually participate in SSI have countable resources that are above the thresholds in the current period (not shown). For current-period recipients and nonrecipients together, the comparable figure is 8.7 percent.30

Table 3. Distribution of countable resources for future SSI recipients compared with future nonrecipients with low potential Social Security benefit amounts, November 1996
Percentile Future SSI
recipients
Future SSI nonrecipients with low potential Social Security benefits
5th 0 0
10th 0 0
25th 0 232
40th 0 1,624
Median 0 6,000
60th 40 13,204
75th 503 48,953
90th 5,802 166,392
95th 13,847 319,252
N 130 2,182
SOURCE: 1996 SIPP matched to Social Security administrative records.
NOTE: Sample members are aged 55–64. Future payment receipt is defined as participating in the SSI program within 6 months of reaching age 65. The sample is restricted to current SSI nonrecipients. The distributions are weighted.

Another way to address the effects of savings disincentives is to examine which part of the distribution would face the strongest spend-down incentives. For people below the thresholds in the current period, the SSI program would provide no incentive for further resource reductions. At resource levels higher than the resource thresholds, there would be an incentive for resource spend-down; however, the force of the incentive would decline as resource levels increase. At some point, the value of the reduction in resources would exceed the present value of the stream of SSI payments. Thus, the strongest savings disincentive, a priori, would be faced by people who would need only a small reduction in savings in order to attain program eligibility. The strongest incentive for future participants would be for those somewhat below the 90th percentile of the distribution of countable resources, as shown in Table 3. These people have resource levels a moderate amount above the program thresholds.

Excluding the group that eventually takes up SSI (and the group that is eligible in the current period), the remaining population can be divided into two groups. The first group could not become eligible even through resource spend-down and has been referred to as having neither condition (neither low potential benefits nor low resources) in this article. The second group could become eligible through resource spend-down and has been referred to as having low potential benefits only. The members of this group serve as a relevant comparison group for the group of eventual participants because their potential Social Security benefits do not disqualify them from future eligibility, but their countable resources in the current period do. Thus, they face a comparable incentive to decrease their resource levels.

This group is used as a comparison group in Table 3. For this group, the strongest incentive is faced by the people around the median of the resource holdings distribution. This level is somewhat above the level of the resource thresholds. At some point of the distribution, perhaps around the 90th percentile, the incentive becomes negligible.

A complementary way to address the effects of savings disincentives is by using this comparison group in a synthetic cohort. Neumark and Powers (1998) look at resource holdings at different ages to infer whether some people are decreasing resources as they approach age 65. Rather than following a cohort over time, this method synthetically creates a cohort from the ages that are observed at one point in time.

Neumark and Powers (1998, Table 2) compare the net worth of people aged 60–62 with those aged 63–64. After isolating a group of likely participants based on a method similar to propensity scoring,31 the authors examine changes in wealth at various points of the wealth distribution. They note that, at the 75th percentile, wealth decreased from around $13,000 for people aged 60–62 to around $5,000 for people aged 63–64.

The synthetic cohort of Neumark and Powers is reproduced here, but with several improvements. Most importantly, I use data on actual people who eventually participate in SSI rather than relying on estimates of likely future participation. Also, I use all people rather than just male heads of the household. Further, I use resource measures that correspond to the program criteria; that is, countable resources for the SSI unit are used rather than net wealth for the family.32

A downward trend is visible for all people who participate in SSI in the future and have positive levels of countable resources (Table 4, top panel). At the point of the strongest disincentive around the 90th percentile, the trend is pronounced and monotonic; countable resources decrease by over 50 percent across the observed age groups. At the 95th percentile, there is also a decrease of over 50 percent, but the reductions do not bring the resource levels down to levels that would attain program eligibility.33

Table 4. Distribution of countable resources for future SSI recipients compared with future nonrecipients with low potential Social Security benefit amounts, by age group, November 1996
Percentile 55–57 58–59 60–62 63–64
Future SSI recipients
5th 0 0 0 0
10th 0 0 0 0
25th 0 0 0 0
40th 0 0 0 0
Median 0 0 0 0
60th 200 100 100 20
75th 464 1,000 928 300
90th 5,911 5,000 3,500 2,800
95th 14,300 22,232 5,911 6,000
N 35 29 42 24
Future nonrecipients with low potential Social Security benefits
5th 0 0 0 0
10th 0 0 0 0
25th 90 232 232 232
40th 1,000 1,860 1,624 4,000
Median 3,249 5,847 6,175 11,292
60th 7,935 11,836 13,994 17,550
75th 39,408 41,167 50,185 60,459
90th 125,997 170,800 153,161 218,393
95th 271,127 382,722 292,352 354,932
N 526 467 688 481
SOURCE: 1996 SIPP matched to Social Security administrative records.
NOTE: Future benefit receipt is defined as receiving SSI payments within 6 months of reaching age 65. The sample is restricted to current SSI nonrecipients. The distributions are weighted.

These results tentatively confirm previous research that shows that resource spend-down may be occurring in anticipation of program eligibility during retirement. One reason for caution is that these results might reflect general trends for this population rather than trends that are related to means-testing. For example, a general decline in resource levels may be the norm among people with low lifetime earnings (as measured by the potential benefit) in the near-elderly ages. Alternatively, a general decline may be the norm among people in this age group who also have low levels of resources. Both aspects can be addressed by examining the distribution of the comparison group used earlier.

The distribution of countable resources for the comparison group, people with low potential benefits who do not participate in SSI in the future, is shown in the bottom panel of Table 4. The members of this group are not disqualified from eligibility by their potential benefits and thus could be eligible depending on resource levels. At certain resource levels, members of this group would face the same savings disincentives as those who participate in SSI in the future.

An increase in resource levels is visible at all points of the distribution that have positive resources for the comparison group. This shows that a decline in resources among people in the near-elderly age group with low lifetime earnings is not the norm. For those who also have low levels of countable resources (for example, below the median for the youngest group, aged 55–57), the changes across age groups are generally zero or positive. Thus, declining resource levels are not the norm for this group either. At the point of the distribution that would face the strongest disincentive to save, around the median, there are observed increases in resources across age groups. The end result of the increase is resource levels that are well above the levels of program eligibility shown in Table 1.

The evidence indicates that future SSI recipients tend to decrease resource levels as they approach the age of categorical eligibility. The conclusion is strengthened by comparison with a group that would be eligible, but does not receive payments. This group shows increasing resource levels during the same ages.

The evidence is pronounced at the points of the distribution for which one would expect the savings disincentive to be greatest. However, the evidence is also present at other points in the distribution. For example, future SSI recipients with resource levels well above the program thresholds are also observed to reduce resources, but not to the levels of program eligibility. In addition to the presence of measurement error, this may indicate that other factors are at work. A more comprehensive analysis would control for other factors that would be expected to influence changes in resource levels in the near-elderly age group, particularly health shocks, changes in family status, and changes in employment status; however, this is left to future research.

Divorce. In contrast to resources, the Social Security benefit level can be only indirectly controlled by the individual. Early retirement will reduce the potential benefit amount, but the estimates used in this study use the assumption of universal early retirement among the low-income population. Thus, no further reductions are possible. The only remaining mechanism to change the potential Social Security benefit amount is changes in marital status. Married people whose potential benefit is too high for program eligibility could become eligible through widow(er)hood or divorce.

The Social Security benefit amounts of spouses and widow(er)s are set by specific ratios of the worker's primary insurance amount (PIA). A spousal benefit is one-half of the worker's PIA, and a widow(er) benefit is the full worker's PIA. One simplified example is a worker with an earnings history and a corresponding PIA who has a spouse who has not worked or is not fully insured. In this case, the couple could receive a benefit based on the worker's full PIA and a spousal benefit based on one-half of the worker's PIA. The sum would count as income and be evaluated against the income thresholds for couples, given in Table 1. In the case of widow(er)hood, there would be one benefit for the surviving spouse based on the full PIA. In the case of divorce, the same spouse would receive a benefit based on one-half of the worker's PIA. For both widow(er)s and divorced people, the benefit would count as income; however, the income thresholds for individuals rather than couples would apply.

In the case of widow(er)hood, the economy of scale implied by Social Security benefit amounts could be as high as 1.5 for couples to 1 for individuals. This is the same economy of scale implied by the SSI program eligibility thresholds, although the ratios for the effective income thresholds are slightly different (after considering income exclusions). Table 1 shows the economies of scale implied by the three programs by showing the ratio of income and resource thresholds for individuals and couples. The ratios for the Medicaid program are very similar to SSI; however, the ratios are lower for SNAP. Thus, there will be few changes in SSI and Medicaid income eligibility status based on changing Social Security benefit amounts that are due to widow(er)hood. For SNAP, widow(er)hood may change income eligibility status because the couple to widow(er) benefit ratio is larger than the couple to individual income threshold ratio.

In the case of divorce, the economy of scale implied by Social Security benefit amounts could be as high as 1.5 for couples to one-half for divorced individuals. This is larger than the couple to individual income threshold ratios for all three of the programs. Thus, divorce could change eligibility status for all three programs.

Actual changes in Social Security benefit amounts and corresponding changes in program eligibility could differ from the examples discussed earlier. As an illustration of the potential reach of the divorce issue across the population, I have recalculated Social Security benefit and SSI payment amounts for the hypothetical case in which all observed couples divorce. The potential Social Security benefit amounts for this hypothetical case are based on observed earnings histories. Correspondingly, I have also recalculated countable resource amounts by dividing the resources of couples in half.

The percentages of people who would have low Social Security benefit amounts and low countable resources in the case where all couples in the sample divorce are shown in Table 5. Although divorce rates were very low for this cohort at the time of the survey (Kreider and Fields 2002), estimates for the subpopulations that could become eligible in this hypothetical case are provided in the table.

Table 5. Percentage of people with potential Social Security benefit amounts below the effective programmatic income and resource thresholds: Status quo and hypothetical divorce cases, by sex, November 1996
Circumstance Low potential benefit only Low resources only Both conditions Neither condition
Men
Status quo 18.1 9.4 19.5 53.0
Hypothetical divorce 28.2 6.3 26.5 39.0
Women
Status quo 26.5 5.5 25.9 42.1
Hypothetical divorce 61.7 0.4 33.8 4.2
SOURCE: 1996 SIPP matched to Social Security administrative records.
NOTE: Sample members are aged 55–64. See Table 1 for the effective income and resource thresholds. The hypothetical divorce case assumes that all observed couples get a divorce.

The impact of potential divorce differs dramatically by sex (Table 5). For women, one-third (33.8 percent) are estimated to be eligible for one or more of the means-tested programs in the case of divorce. This compares with 25.9 percent for the status quo. In addition to the one-third that would be eligible, for somewhat less than two-thirds (61.7 percent), there would be the possibility of spending down resources to become eligible. The extent of spend-down would have to be substantial in many cases because this group of women has high resource holdings by the standards of means-tested programs; the median value is around $70,000, and the 25th percentile is around $20,000. The point is not to assert that SSI program participation is likely for this group, but rather that program eligibility is possible for a much larger group of women than was initially shown in Table 2. In fact, less than 5 percent (4.2 plus 0.4) of women in this cohort are definitively ineligible based on Social Security benefits alone in the case of divorce.

In the case of divorce for men, around 45 percent (39.0 plus 6.3) are definitively ineligible based on Social Security benefits alone. Around a quarter (26.5 percent) would be eligible and a bit over another quarter (28.2 percent) could become eligible through spend-down of resources. Both figures are moderate increases from the status quo figures.

Conclusions

Among the near-elderly cohort analyzed in this study, around a quarter are financially eligible for one or more of the three largest means-tested programs that serve the elderly. This group may be eligible for these programs upon reaching the ages of categorical eligibility. Only around a fifth of this group (the financially eligible) is observed receiving SSI payments upon reaching age 65, however. This result leads to the issue of changes in financial eligibility between the near-elderly time period and reaching age 65.

Two components of financial eligibility are low potential Social Security benefits and low liquid resource holdings. This study has shown that low potential Social Security benefits are due to both low frequencies of earnings and also low levels of earnings over the life cycle. Further, those with low potential Social Security benefits are more likely to be women, nonmarried individuals, and immigrants. Low potential Social Security benefits in conjunction with low liquid resources define the group that is estimated to be financially eligible for one or more of the means-tested programs in this study. Distinguishing characteristics of this group are a higher prevalence of disability over the lifetime, a lower prevalence of marriage, and lower total family income.

Within the group that is financially eligible for one or more means-tested programs in the near-elderly time period, the rate of SSI payment receipt upon reaching age 65 varies by the level of potential Social Security benefits and liquid resources. The participation rate is substantially higher at very low levels of both potential benefits and resources, compared with other financially eligible people. The rate is even higher for people who also lack a defined benefit pension. By contrast, a lack of home ownership does not additionally increase the participation rate.

This article concludes with some illustrations of changes in program eligibility status between the near-elderly and elderly time periods. Of the group that is observed to receive SSI payments after reaching age 65, 13.1 percent of current nonrecipients and 8.7 percent of current nonrecipients and recipients combined report that they have resources over the eligibility thresholds at the time of the survey. From this, one can infer that spend-down occurs with moderate frequency. When examining resource levels across age groups leading up to the elderly time period, resource levels decline for future SSI recipients in the part of the resource distribution where the savings disincentive would be the strongest. The opposite trend is observed among a comparable group of people who do not receive SSI payments in the future. This is compatible with the theory that the SSI program induces spend-down of resources, but assessment of causation is left to future research.

Another possible change in eligibility between the near-elderly and elderly time periods could result from a change in the potential Social Security benefit that is due to divorce. Although the incidence of divorce is low during the near-elderly time period, a majority of women would be income-eligible for one or more means-tested programs in the hypothetical divorce scenario examined here. However, only a minority of these women would also have resource levels below the relevant thresholds.

Technical Appendix

The figures in Table 1 are averages when there is variation across states; averages are weighted by the number of elderly SSI recipients in each state. The figures represent the maximum levels of income and resources for which eligibility could be obtained by elderly individuals or couples in some circumstances. Not all applicants at these levels would be eligible in all circumstances because of variation in provisions across states and variation in the methods for counting income across states in the Medicaid and SSI programs.

For SNAP, the thresholds are uniform across states with the exception of different income thresholds for Alaska and Hawaii, which are not considered in this table. The income thresholds follow the Department of Health and Human Services poverty guidelines and assume that the individual or couple are the only household members. The income and resource thresholds are those for households with an elderly member.

Federal SSI thresholds are also uniform across states; however, the availability of state SSI supplements raised the income threshold in some states for the elderly in 2001. The SSI income threshold figures include the federal benefit rate plus the state supplements that are available to aged individuals or couples living independently (see Bruen, Wiener, and Thomas (2003, Table 5)). Although the federal resource thresholds apply for all states, some states had more restrictive resource thresholds for the state supplement than for the federal benefit in 2001. This variation is not considered here because the federal threshold determines the maximum level of eligibility for any kind of benefit in this case.

Medicaid eligibility is closely tied to SSI eligibility because states are required to provide Medicaid to federal SSI recipients. This requirement does not apply to states that follow section 209(b) of Public Law 92-603. Eleven states followed section 209(b) provisions in 2001, of which six had more restrictive income standards than the federal SSI program and eight had more restrictive resource standards than the federal SSI program (Bruen, Wiener, and Thomas 2003). The effects of these more stringent standards are not addressed in Table 1, which makes the Medicaid figures an upper bound of possible estimates. As opposed to the section 209(b) provisions, other provisions make the Medicaid program more lenient in terms of eligibility. These provisions include the "poverty-related" and "medically needy" programs (see Bruen, Wiener, and Thomas (2003) for exact definitions). The Medicaid figures in Table 1 reflect the weighted averages of thresholds including the poverty-related provisions, but excluding the medically needy provisions because the former expands general program eligibility and the later expands eligibility for a subpopulation that also meets the medically needy criteria. The poverty-related thresholds are given in Bruen, Wiener, and Thomas (2003), and the averages are also weighted by the number of SSI recipients aged 65 and older in each state.

Notes

1 That is, they were categorically eligible based on age or another criterion.

2 The cohort is roughly aged 55–64 at the time of the analysis, whereas the age of categorical eligibility is 65 for Medicaid and SSI. For SNAP, different criteria apply to a household with a member older than age 60.

3 SNAP was called the Food Stamp Program until October 2008. I use the term SNAP even though the data used in this analysis refer to the Food Stamp Program.

4 2001 is the midpoint of when the near-elderly cohort used in this study reached the traditional full retirement age (FRA) for Social Security retirement benefits. See the Methods section for more information.

5 The primary unearned income exclusion is the most commonly applied exclusion. For SNAP and Medicaid, the unearned income exclusion listed in Table 1 does not capture all of the variation by state, family size, and route to eligibility.

6 The effective income thresholds could also be higher because of the earned income exclusions if the person has earned income.

7 SNAP thresholds differ for Alaska and Hawaii; however, this variation is not addressed in this table.

8 Also, there is state variation in the extent to which states must provide Medicaid to SSI recipients. See the Technical Appendix for more information.

9 Florida and Pennsylvania together had 9.9 percent of SSI recipients aged 65 or older in 2001.

10 There are also differences in income and resource counting methods between programs and across states, but these differences are difficult to summarize in a table.

11 Author's calculation based on figures reported in Elder and Powers (2004, Table 6) for calendar year 1997. Those authors report unweighted figures; thus, the comparison represents an average in the sample not accounting for the sample selection probabilities.

12 Davies, Rupp, and Strand (2004) consider the effects on SSI program eligibility and participation of eliminating the resource test and counting an annuitized value of resources as income.

13 See Hubbard, Skinner, and Zeldes (1995) and Ziliak (2003) for example.

14 Elderly people may gain financial eligibility at a different age than when they gain categorical eligibility. For example, Medicaid has procedures for reducing resources in order to obtain long-term care benefits.

15 No indexing is used to make dollar amounts more comparable across time periods.

16 Specifically, if the spouse is also categorically eligible for SSI (blind, aged, or disabled), then the spouse's Social Security benefit is treated as income, otherwise the benefit of the spouse is "deemed" as income to the potential beneficiary following SSI program rules.

17 There could also be additional earnings between the reference period and reaching age 65, which would have a similar effect.

18 For people observed to be Disability Insurance (DI) beneficiaries in the administrative records, the Social Security benefit is calculated at the time of the first payment and updated using cost-of-living increases.

19 A spouse is eligible for spouse's benefits if the marriage is valid at the time of claiming, and a divorced spouse is eligible if the divorce occurred after 10 years of marriage.

20 Analysis of program participation is restricted to programs administered by SSA. It would be useful to examine other programs, however, the retrospective program participation data in the SIPP is of minimal usefulness. For example, according to matched administrative records, 38.1 percent of those with both low potential Social Security benefits and low countable resources were at some point recipients of the SSI payments. The comparable self-reported value in the SIPP is only 2.2 percent.

21 The differences mentioned in this section are all statistically significant at the 5 percent level. The standard errors used in statistical tests need to be adjusted in order to account for the complex sample design of the SIPP. I give unadjusted standard errors in the tables; however, I use an approximate adjustment for statistical tests. The Census Bureau (2001) gives design effects (adjustment factors) that account for the effect of the complex sample design on the variances of various survey items. Because the estimated design effect exceeds four only for one item (metro status) and is much smaller for other survey items, I adopt a design effect of four for the variances. This implies true standard errors that are twice as large as the unadjusted standard errors. Assuming a design effect of four provides conservative tests of population differences.

22 The AIME is calculated based on the benefit formula and administrative earnings records and may differ from the official AIME that would be used by SSA in the benefit calculation.

23 The phrase "participate in SSI within 6 months of reaching age 65" is not meant to imply that the first SSI payment receipt necessarily occurred within this time period. In fact, the majority (61.4 percent) of people who receive SSI payments during this period also received SSI payments before reaching age 65.

24 These figures can be derived from the figures given in Table 2, but differ somewhat because of rounding.

25 See Strand, Rupp, and Davies (2009) for a review of this literature.

26 See Davies, Rupp, and Strand (2004), for example.

27 Missing defined benefit pension observations are assumed to indicate lack of a benefit; however, this does not appear to have lead to an overestimate of the proportion of people who lack a defined benefit pension. See Department of Labor (2005) for comparison.

28 Although adding an additional restriction (no home ownership in addition to no defined benefit pension and low countable resources) creates a smaller subgroup, this smaller subgroup can have a higher or lower rate of SSI payment receipt.

29 See Davies, Rupp, and Strand (2004), for example.

30 This figure can also be derived from the information in Table 2, but the derived version differs somewhat because of rounding.

31 Participation likelihood is based on the characteristics of people who are older than age 65 and report receiving SSI payments in the SIPP. Predictors include the existence and maximum amount of state SSI payments, demographic characteristics, and the history of participation in the Food Stamp Program.

32 Also, the reference period is November 1996 in this study, compared with reference periods from 1984 through 1986 in Neumark and Powers (1998).

33 The sample sizes in the top panel of Table 4 are small, but are comparable to Neumark and Powers (1998) who base their analysis on samples of 38 and 30 observations for two groups.

References

Balkus, Richard, and Susan Wilschke. 2003. Treatment of married couples in the SSI program. Issue Paper No. 2003-01. Washington, DC: Social Security Administration.

Bean, Frank D., Gillian Stevens, and Jennifer Van Hook. 2003. Immigrant welfare receipt: Implications for policy. In America's newcomers and the dynamics of diversity, Frank Bean and Gillian Stevens, eds., 66–93 New York, NY: Russell Sage Foundation.

Bruen, Brian K., Joshua M. Wiener, and Seema Thomas. 2003. Medicaid eligibility policy for aged, blind, and disabled beneficiaries. AARP Public Policy Institute Issue Paper #2003-14. Washington, DC: AARP (November).

Census Bureau. 2001. Source and accuracy statement for the 1996 public-use files from the Survey of Income and Program Participation. Washington, DC.

Chen, Henry, and Robert I. Lerman. 2005. Do asset limits in social programs affect the accumulation of wealth? Opportunity and Ownership Project series, Brief #4. Washington, DC: Urban Institute (August).

Davies, Paul S., Minh Huynh, Chad Newcomb, Paul O'Leary, Kalman Rupp, and Jim Sears. 2001/2002. Modeling SSI financial eligibility and simulating the effect of policy options. Social Security Bulletin 64(2): 16–45.

Davies, Paul S., Kalman Rupp, and Alexander Strand. 2004. The potential of the Supplemental Security Income program to reduce poverty among the elderly. Journal of Aging and Social Policy 16(1): 21–42.

Department of Labor. 2005. National Compensation Survey: Employee benefits in private industry in the United States, 2002–2003. Washington, DC.

Elder, Todd E., and Elizabeth T. Powers. 2004. SSI for the aged and the problem of take-up. Michigan Retirement Research Center Working Paper No. 076. Ann Arbor, MI: University of Michigan.

———. 2007. A longitudinal analysis of entries and exits of the low-income elderly to and from the Supplemental Security Income program. Michigan Retirement Research Center Working Paper No. 156. Ann Arbor, MI: University of Michigan.

Farrell, Mary, Michael Fishman, Matthew Langley, and David Stapleton. 2003. The relationship of earnings and income to Food Stamp participation: A longitudinal analysis. Economic Research Service. Electronic Publications from the Food Assistance & Nutrition Research Program, E-FAN-03-011 (November).

Gruber, Jonathan, and Aaron Yelowitz. 1999. Public health insurance and private savings. Journal of Political Economy 107(6): 1249–1274.

Hubbard, Glenn R., Jonathan Skinner, and Stephen P. Zeldes. 1995. Precautionary saving and social insurance. Journal of Political Economy 103(2): 360–399 (April).

Kreider, Rose M., and Jason M. Fields. 2002. Number, timing, and duration of marriages and divorces: 1996. Current Population Reports, Series P70-80. Washington, DC: Census Bureau,.

Neumark, David, and Elizabeth Powers. 1998. The effect of means-tested income support for the elderly on pre-retirement saving: Evidence from the SSI program in the U.S. Journal of Public Economics 68(2): 181–206 (May).

———. 2000. Welfare for the elderly: The effects of SSI on pre-retirement labor supply. Journal of Public Economics 78(1–2): 51–80 (October).

Powers, Elizabeth T., and David Neumark. 2003. The interaction of public retirement income programs in the United States. American Economic Review 93(2): 261–265.

Radner, Daniel. 1990. Alternative estimates of economic well-being by age using data on wealth and income. ORES Working Paper No. 42. Washington, DC: Social Security Administration.

Strand, Alexander, Kalman Rupp, and Paul S. Davies. 2009. Measurement error in estimates of the participation rate in means-tested programs: The case of the U.S. Supplemental Security Income program for the elderly. Paper presented at the Federal Committee on Statistical Methodology Research Conference, Washington, DC (November).

Ziliak, James P. 2003. Income transfers and assets of the poor. Review of Economics and Statistics 85(1): 63–76.