Why Researchers Now Rely on Surveys for Race Data on OASDI and SSI Programs: A Comparison of Four Major Surveys

Introduction

The characteristics of beneficiaries of the Old-Age, Survivors, and Disability Insurance (OASDI, or Social Security) and Supplemental Security Income (SSI) programs are of longstanding interest to researchers, advocacy groups, and policymakers. The Social Security Administration (SSA) addresses the subject of beneficiary characteristics by publishing program statistics that are segmented by various demographic categories. Much of this work, such as one of the agency's most comprehensive publications—the Annual Statistical Supplement to the Social Security Bulletin (or the Supplement)—is based on SSA's administrative data. However, statistics produced using administrative data no longer include a common characteristic many stakeholders use to analyze OASDI and SSI programs—race.

SSA discontinued the publication of data by race for the SSI program after 2002 and for the OASDI program after 2009 (Martin and Murphy 2014). This is largely because of changes to SSA's process for assigning new Social Security numbers (SSNs), a process called enumeration. Nearly all individuals (96 percent) now receive their original SSNs as part of the hospital-birth registration process under a procedure called Enumeration at Birth (EAB). Qualified immigrants receive most of the remaining new SSNs upon entry into the United States under a procedure called Enumeration at Entry (EAE). SSA does not collect any associated race or ethnicity data as part of these expedient and cost-efficient procedures.¹

Today, relatively few individuals manually file an original Social Security card application at an agency field office. Moreover, providing race and ethnicity information is voluntary, as it is unnecessary for administering the program. Consequently, the administrative data on race and ethnicity that SSA does collect comes from a self-selecting sample that represents an ever-dwindling proportion of the population.² Furthermore, although SSA had collected race and ethnicity data in earlier years, the agency changed the number and definition of race categories over the years, making long-term comparisons of race data statistically meaningless.

In the absence of reliable administrative data for race categories, SSA researchers now primarily use data from four major surveys—(1) the Current Population Survey (CPS); (2) the Survey of Income and Program Participation (SIPP); (3) the American Community Survey (ACS); and (4) the Health and Retirement Study (HRS)—to examine OASDI and SSI program use by race and ethnicity. The race data collected in these surveys are self-reported, and although there are inherent technical issues with that process, those data are the primary source of national race data available (Ríos, Romero, and Ramírez 2014). However, each data set possesses a unique set of advantages and limitations for analyzing the OASDI and SSI programs.

This note provides a historical background on SSA's race and ethnicity data collection. It compares the four alternative data sources most commonly used to estimate OASDI and SSI program statistics by beneficiary race and ethnicity. This overview is designed to help individuals better understand the race and ethnicity data available in existing SSA publications. It may be of use to researchers considering future OASDI and SSI analyses as well.

Background

Before EAB began in 1987 (followed by EAE in 2002), all individuals were required to file Form SS-5 to apply for an SSN, to report a change to the information previously submitted (such as a name change), or to request a replacement Social Security card.³ Because federal agencies do not need race or ethnicity data to administer their programs, Form SS-5 is unusual among federal forms (particularly those serving as applications) in that it enables applicants to report their race and ethnicity voluntarily. Individuals have been able to report that information since race and ethnicity designations first appeared on the form in 1936.

Before 1980, Form SS-5 did not indicate whether providing information on race and ethnicity was voluntary, and many applicants probably assumed it was required.⁴ Applicants were allowed to select one of three races: white, negro, or other (specify). SSA classified as “unknown” the race of those who did not make a selection. In 1980, the number of race/ethnicity categories on the Form SS-5 expanded to five: white (non-Hispanic), black (non-Hispanic), Hispanic, Asian or Pacific Islander, and American Indian or Alaska Native.

In 1997, the Office of Management and Budget issued revised standards for the classification of federal data on race and ethnicity that established a minimum of five race categories and the separate consideration of Hispanic or Latino ethnicity, regardless of race. On the SS-5 today, an applicant can select from among seven race categories: white, black/African American, Asian, Native Hawaiian, Other Pacific Islander, Alaska Native, and American Indian. A separate question allows the applicant to select his or her ethnicity—Hispanic or Latino—by answering yes or no. The use of expanded race categories in later years could not easily be reconciled with data from earlier years that used fewer categories. As a result, SSA's primary statistical publication, the Annual Statistical Supplement of the Social Security Bulletin, retained only the original three categories—white, black, or other—for reporting program data by race until the 2011 Supplement in which the agency discontinued providing race and ethnicity breakdowns.

SSA's collection and use of data has always been restricted to statistical and research purposes. Data on race and ethnicity are subject to legal and policy constraints that exceed the standard confidentiality rules for administrative data. The use of race data is narrowly restricted to statistical linkage or summary statistics (Alexander 1980). SSA originally used race data to aid in identifying individuals, but now it uses the data for statistical and research purposes only (Scherr and Nelson 1980). Several SSA publications used only administrative race data to provide comprehensive information on beneficiary racial or ethnic characteristics (Lingg 1978, 1980, 1983, 1984, 1986). Scott (1999) discusses in detail the limitations of using SSA administrative data for that purpose.

In the past, most individuals completed Form SS-5 to request an original SSN when they first applied for a job. In the mid-1980s, individuals younger than age 22 accounted for almost all new SSN applicants (94 percent), with those younger than age 15 accounting for 74 percent and individuals younger than age 5 accounting for 41 percent (Cronin 1985). By contrast, almost all of today's SSN applicants are younger than age 5. The change began when the Tax Reform Act of 1986 required taxpayers claiming children as dependents to obtain an SSN for each child aged 5 or older, which in turn created a spike in the demand for issuing SSNs to children at earlier ages. Because of that demand and the fact that the SSN had by then been adopted almost universally as a standard identifier, SSA in 1987 began entering into agreements with all 50 states to provide the EAB service option (Scott 1999). Today, parents have reasons unrelated to Social Security to acquire SSNs for their children, such as buying savings bonds in their name, opening a bank account, obtaining medical coverage, or applying for certain government benefits.⁵

EAB allows parents to voluntarily apply for SSNs for their newborns as part of the hospital-birth registration process. SSA assigns about 96 percent of new SSNs to infants through this convenient service option, saving the time a parent would otherwise have spent gathering the necessary documentation, completing an SS-5 application, and visiting (or mailing original documents to) a field office for processing. Instead, participating hospitals forward the SSN request and other birth registration data to their states' vital statistics agencies, which then send the information to SSA. EAB does not provide SSA with race and ethnicity data because they are classified under “Information for Medical and Health Use Only” on the birth certificate and are unnecessary for administering the OASDI and SSI programs (Scott 1999). In fiscal year 2015, SSA assigned almost 4 million SSNs to newborns through the EAB program (SSA 2015a).

In 2002, SSA began the EAE pilot, a joint effort with the Departments of State (DoS) and Homeland Security (DHS). EAE initially allowed a noncitizen aged 18 or older who had been admitted to the United States as a lawful permanent resident to obtain an SSN based on the data collected by DoS during the immigration visa process conducted in the individual's home country. DoS transmits the enumeration data to DHS after the visa has been granted. DHS updates certain data, if necessary, before sending the application to SSA for an SSN, which the individual receives upon admission to the United States. Data collected for this program do not include the applicant's race or ethnicity. In 2009, SSA expanded EAE to all people, regardless of age. In fiscal year 2015, SSA issued over 229,000 Social Security cards through EAE—222,570 original cards and 6,754 replacement cards (SSA 2015b).

The SSN as a Record Link

For research and statistical purposes, SSA develops and relies on information from linked administrative and survey data, which the agency then shares with other researchers, policymakers, and the public (Fisher 2007; Davies and Fisher 2009). Linking SSA's administrative data with survey data yields a broader set of demographic and socioeconomic information and provides data of better quality than those available from a survey alone (Haines and Greenberg 2005; Davies and Fisher 2009). SSA uses these data to analyze and research policy initiatives for the OASDI and SSI programs as well as the earnings of the working and beneficiary populations (Fisher 2007). SSA's biggest data-linkage partner is the Census Bureau.

Substantial methodological benefits exist for linking administrative data with survey data. One major advantage involves data accuracy. When both an administrative file and a survey collect comparable data, statisticians and policy analysts are better able to evaluate the extent of underreporting and overreporting. For example, survey respondents often confuse the SSI and OASDI programs when reporting payments or benefits respectively received. Further, by matching survey responses to administrative data, agencies can compare the benefit amounts reported by survey respondents with the actual dollar amounts distributed by the agency (Fisher 2007). In addition, matching administrative and survey data provides operational efficiencies by using data previously collected and validated by another government agency, thereby saving the time and money it would take to collect the data independently. SSA administrative data linkages are available for three of the four surveys this note examines.

Comparison of Surveys

This section presents profiles of the four surveys that provide race and ethnicity data for use in analyzing OASDI and SSI beneficiary populations. The profiles highlight each survey's design, methodology, and history; the categories for which the survey collects race and ethnicity data; and the survey's strengths and limitations for analyzing SSA program data. Table 1 summarizes the attributes of four surveys that collect race and ethnicity data on the OASDI and SSI programs.

Current Population Survey

The CPS, sponsored jointly by the Census Bureau and the Bureau of Labor Statistics, is a monthly cross-sectional survey of about 60,000 eligible households in the United States. It provides comprehensive data on the labor force, including employment, unemployment, persons not actively involved in the workforce, hours of work, earnings, and other demographic and labor force characteristics.

Design, methodology, and history. The CPS is the primary source of U.S. labor force statistics and includes extensive demographic data. The CPS Annual Social and Economic Supplement (CPS/ASEC) is a cross-sectional survey examining the social and economic well-being of the U.S. civilian noninstitutionalized population. It collects annual supplemental data on family characteristics, work experience, prior-year income from all sources, receipt of noncash benefits, program participation, geographic mobility, and health insurance coverage. Households in the CPS/ASEC sample are interviewed once a year for two consecutive years (Census Bureau, n.d. b).

Race and ethnicity categories. Since January 2003, CPS respondents have been able to select from one or more of the following five race categories: white, black or African American, Asian, American Indian or Alaska Native, and Native Hawaiian or Other Pacific Islander. An additional survey question addresses Spanish, Hispanic, or Latino ethnicity, irrespective of race. SSA publications that have used CPS data on racial and ethnic differences among program beneficiaries include Hendley and Bilimoria (1999); Gesumaria and Weaver (2001); Koenig (2003); Olsen (2005/2006); McNabb and others (2009); Davies and Fisher (2009); Weaver (2010); Olsen and O'Leary (2011); Anguelov, Iams, and Purcell (2012); and Iams and Purcell (2013).

Advantages for SSA research. The CPS—

• distinguishes among different types of OASDI and SSI benefits received among all races and ethnicities;
• collects data that are representative at national, state, and regional levels;
• enables linkage with SSA administrative data to yield a broader set of demographic and socioeconomic information; and
• approximates actual gross benefit amounts, including Medicare deductions.

Limitations for SSA research. The CPS—

• underreports OASDI benefits and significantly underreports SSI payments because OASDI beneficiaries and SSI recipients may confuse the two programs or fail to report benefits and payments (Huynh, Rupp, and Sears 2002; Fisher 2007; Nicholas and Wiseman 2009; Davies and Fisher 2009);
• does not collect data on OASDI beneficiaries and SSI recipients younger than age 15, thereby restricting the analysis of program participation to children aged 15–17;
• does not provide local geographic data; and
• employs techniques such as income topcoding, rank proximity swapping, and age perturbation to protect the privacy of respondents aged 65 or older, which produces errors in the data for that age category (Alexander, Davern, and Stevenson 2010; Census Bureau 2010; Leonesio and others 2012).

Survey of Income and Program Participation

The SIPP is the Census Bureau's most extensive source of information on income and program participation. It collects data and measures change with regard to the nation's economic well-being and the family dynamics of households therein.

Design, methodology, and history. The SIPP is a household-based longitudinal survey of the demographic and socioeconomic well-being of the U.S. civilian noninstitutionalized population and of adults in the military living in on- or off-post households with at least one civilian adult member. The survey principally provides monthly information about cash and noncash income and the program participation of individuals, families, and households. It also collects data on taxes, assets, liabilities, participation in government transfer programs, and health insurance coverage (Census Bureau, n.d. c).

The SIPP collects more detailed data than any other national survey on program eligibility, access and participation, transfer income, and in-kind benefits. By design, the SIPP oversamples low-income households; as such, it covers large numbers of minority households. From its inception in 1983 through 2013, the SIPP interviewed sample (panel) households at 4-month intervals over a 3- to 4-year period. Each interview collected monthly data about the previous 4 months which, when aggregated, produced annual data.

In 2014, following a 5-year redesign effort, the Census Bureau revised the SIPP methodology. The new SIPP includes a 4-year panel, but interviews respondents annually and asks them to recall details of demographic characteristics and program participation over the previous calendar year. With some subjects, the new SIPP applies an events-history format to identify important dates of changes. The plan is to link SSA administrative data on earnings and benefits to the survey data. At present, the data properties for OASDI and SSI benefits generated by the revised process are unknown, as are the survey's major strengths and weaknesses. The Census Bureau asserts that the original and revised SIPP methodologies produce similar results, while saving tax dollars.

Race and ethnicity categories. SIPP respondents select from one or more of the following five race categories: white, black or African American, Asian, American Indian or Alaska Native, and Native Hawaiian or Other Pacific Islander.⁶ SSA researchers have used SIPP data in many studies analyzing racial and ethnic differences among OASDI beneficiaries and SSI recipients. Examples include Olson (2002); Huynh, Rupp, and Sears (2002); Koenig (2003); Sears and Rupp (2003); Gottschalk and Huynh (2005); Neumark and Powers (2005); DeCesaro and Hemmeter (2008); Bridges and Choudhury (2009); Tamborini, Iams, and Whitman (2009); Tamborini, Cupito, and Shoffner (2011); Iams and Tamborini (2012); Iams and Purcell (2013); and Bailey and Hemmeter (2014).

Advantages for SSA research. The SIPP—

• distinguishes between OASDI benefit types (retirement, disability, and survivor) and SSI payment types (old-age and disability) among all races, ethnicities, and ages, as well as for differing definitions of disability (Brault 2012);
• compiles data that are representative at a national level;
• enables longitudinal analysis by tracking changes over time;
• does not misclassify OASDI beneficiaries and SSI recipients as much as the CPS does because the SIPP overreports OASDI benefits and SSI payments (Koenig 2003; Fisher 2005, 2007; Davies and Fisher 2009);
• oversamples minorities at panel outset to mitigate later attrition; and
• enables linkage with SSA administrative data to yield a broader set of demographic and socioeconomic information.

Limitations for SSA research. The SIPP—

• does not provide local-level geographic data, and data are not always representative at regional or state levels;
• includes respondents who are prone to report receiving only OASDI benefits when they in fact receive SSI payments only or concurrent OASDI and SSI benefits (Davies and Fisher 2009); and
• includes respondents who are prone to report OASDI benefit income net of Medicare premium deductions rather than actual benefit payments (Iams and Purcell 2013).

American Community Survey

The ACS is an ongoing nationwide survey that provides vital information annually about the United States and its people. Information from the survey generates data that help determine how federal and state funds are distributed each year.

Design, methodology, and history. The ACS is the premier source of information about the American people and the workforce, distributing about 250,000 questionnaires monthly. The ACS samples about 3 million addresses each year, resulting in nearly 2 million final interviews. It has replaced the decennial census long-form sample questionnaire. Following testing from 2000 through 2004, the Census Bureau began collecting ACS questionnaire responses on a rolling basis every month in 2005, instead of once every 10 years. The 7-question ACS form replaced the decennial's long, 50-question form in 2010. The Census Bureau releases the 1-, 3-, and 5-year survey data to the public on a phased schedule the following year, generally in the order of shortest-to-longest survey cycle, and updates the ACS yearly. By producing 1-, 3-, and 5-year estimates, the ACS reflects data collected over time, compared with “point-in-time” estimates (such as the decennial census) that approximate the characteristics of an area on a specific date (Census Bureau, n.d. a).

Of the surveys discussed here, only the ACS includes both the nation's civilian institutionalized and noninstitutionalized populations. Including the civilian institutionalized population—residents housed in group quarters, such as college dormitories, prisons, barracks, shelters, and nursing homes—allows the ACS to capture more data on African American OASDI beneficiaries and SSI recipients, who are overrepresented among the populations in correctional facilities and nursing homes (Martin and Murphy 2014).

Race and ethnicity categories. The ACS features both basic and detailed groups for race and ethnicity. The basic groups include five single-race categories (white, black or African American, Asian, American Indian and Alaska Native, Native Hawaiian and Other Pacific Islander) and Other, which includes two or more races in various combinations. American Indians and Alaska Natives may provide any tribal designation. Detailed Asian groups include Asian Indian, Chinese, Filipino, Japanese, Korean, and Vietnamese. Detailed Native Hawaiian and Other Pacific Islander groups include Native Hawaiian, Samoan, Tongan, Guamanian or Chamorro, and Fijian. ACS respondents may identify ethnicity by selecting “Hispanic or Latino” origin; detailed subgroups include Mexican, Puerto Rican, Cuban, Dominican, Guatemalan, and Salvadoran. SSA studies that used ACS data by race and ethnicity include Martin (2007); Hemmeter (2009); Smith-Kaprosy, Martin, and Whitman (2012); and Martin and Murphy (2014).

Advantages for SSA research. The ACS—

• distinguishes between OASDI benefits and SSI payments;
• compiles data that are representative at all geographic levels: national, state, regional, and local levels (all counties and places, including census tracts and block groups with populations of less than 20,000, 20,000–65,000, and more than 65,000);
• has the largest sample size of all household surveys;
• has the highest response rate of all household surveys because it is mandatory;
• includes both noninstitutionalized and institutionalized populations;
• provides yearly estimates for all states, cities, metropolitan areas, counties, and population groups of 65,000 or more residents;
• allows for more detailed and statistically significant analyses of population subsets than do the CPS and SIPP (Martin and Murphy 2014; Smith-Kaprosy, Martin, and Whitman 2012; Martin 2007);
• includes race and ethnicity data for many broad and detailed groups and subgroups; and
• oversamples the American Indian and Alaska Native population.

Limitations for SSA research. The ACS—

• does not distinguish between OASDI retirement, survivor, and disability benefits, or between SSI retirement and disability benefits;
• includes respondents who may confuse SSI payments with OASDI benefits or fail to report payments;
• excludes OASDI and SSI beneficiaries younger than age 15, so analyses of program participation among children is restricted to those aged 15–17;
• employs techniques such as income topcoding, rank proximity swapping, and age perturbation to protect the privacy of respondents aged 65 or older, which produce errors in the data for that age category (Alexander, Davern, and Stevenson 2010);
• lacks depth in survey questions used to determine disability, meaning that ACS data fails to identify about one-third of the population aged 25–61 that either receives Disability Insurance (DI) benefits or SSI disability payments (Burkhauser and others 2012); and
• does not enable linkage with SSA administrative data.

Health and Retirement Study

The HRS, conducted by the University of Michigan, is a biennial longitudinal panel study that surveys a representative sample of approximately 26,000 Americans aged 51 or older. The survey explores the changes in labor force participation and the health transitions that individuals undergo toward the end of their work lives and in the years that follow.

Design, methodology, and history. Unlike the other three surveys, the HRS is not government-administered. Rather, it is conducted by the University of Michigan under a cooperative agreement with the Department of Health and Human Services' National Institute of Aging and SSA. Longitudinal studies are crucial for understanding how health and economic disparities for individuals evolve with age.

The HRS focuses primarily on the intersection between health, retirement, and economic status in later life. The survey provides detailed information on each of those topics as well as on employment history and the availability of pensions, work-related disability and related benefits, family composition and resource transfers, and health insurance and service use. The HRS measures health conditions comprehensively, encompassing a range of chronic diseases, disabilities, cognition, depression, and sensory functioning, as well as risk factors such as smoking, drinking, exercise, and body weight. In recent waves, the study was expanded to include biomarkers and enriched psychosocial measures, which greatly enhance the potential for studying the mechanisms underlying disparities in health (University of Michigan 2015).

Race and ethnicity categories. HRS respondents select from eight racial categories: white/Caucasian, black/African American, American Indian, Alaska Native, Asian, Native Hawaiian, Pacific Islander, and Other. Respondents may indicate Hispanic or Latino origin regardless of race and may further specify Mexican American/Chicano, Puerto Rican, Cuban American, and Other origins. Prior to 2006, respondents could not report multiple races or ethnicities. SSA publications that have used HRS race and ethnicity data include Olson (1999); Gustman and Steinmeier (2000); Choudhury (2001/2002, 2002); Dushi and Honig (2007, 2008); Dushi and Iams (2008); Purcell (2012); and Murphy, Gourd, and Begay (2014).

Advantages for SSA research. The HRS—

• distinguishes between OASDI beneficiary types (retirement, disability, and survivor) and SSI recipient types (old-age and disability);
• provides national and regional data;
• enables longitudinal analysis by tracking respondents over time;
• oversamples minorities at panel outset to mitigate later attrition; and
• enables linkage with SSA administrative data.

Limitations for SSA research. The HRS—

• provides no unrestricted state and local data; and
• excludes the population younger than age 51.

Discussion

SSA administrative data linkages with CPS, SIPP, and HRS data yield broader sets of demographic and socioeconomic information and provide data of better quality than those that administrative records or surveys alone provide. SSA administrative data linkages with the ACS are not currently available. Each of the surveys, except for the ACS, distinguish among the different types of OASDI and SSI benefit receipt. The CPS and ACS do not collect data for OASDI and SSI beneficiaries younger than age 15, so analyses of program participation among children is restricted to those aged 15–17; conversely, the HRS population is limited to persons aged 51 or older. Only the ACS includes the entire institutionalized population. Some research suggests that the CPS tends to underreport OASDI receipt and substantially underreports SSI payments (relative to the SIPP), but accurately approximates actual gross OASDI benefit amounts, including Medicare premiums. In addition, some research has found that the SIPP has a lower rate of OASDI beneficiary misclassification (versus the CPS), but the benefit amounts recorded in the SIPP more closely reflect net OASDI benefits minus Medicare premium deductions. Therefore, analysts may underestimate total income and thus overestimate the share of the population in poverty and eligible for participation in poverty-related programs. The CPS and ACS are also subject to errors in data for individuals aged 65 or older. In addition, the ACS has data limitations regarding disability program participants.

OASDI and SSI beneficiary data by race and ethnicity are available at the national level from all four surveys and at the regional level from each of the surveys discussed in this note except the SIPP. Beneficiary data by race and ethnicity at the state level are available from the CPS and ACS only, and at the local level from the ACS only. Statistical analyses by race and ethnicity typically use national data.

Conclusion

Policymakers are increasingly interested in the demographic and socioeconomic characteristics of OASDI beneficiaries and SSI recipients, as the minority shares of the senior and disabled populations grow. Although SSA administrative data on race and ethnicity are no longer viable, four major surveys—the CPS, the SIPP, the ACS, and the HRS—provide valuable data on the characteristics of OASDI beneficiaries and SSI recipients. For most policy and social science research, the various surveys, as currently constituted and having improved over time, are adequate. However, each has its own advantages and limitations for analyzing SSA-administered programs. Stakeholders, particularly advocacy groups, have voiced a need for more detailed and robust race data in the future that addresses the demographics of OASDI beneficiaries and SSI recipients. Potential improvements that could assist researchers in meeting stakeholders' needs are better state- and regional-level data and additional linkages with SSA's administrative records.

For further discussion on the importance of collecting race and ethnicity data and other related information, see American Sociological Association (2003); El Nasser (2013); Humes, Jones, and Ramirez (2011); Macartney, Bishaw, and Fontenot (2013); Krogstad and Cohn (2014); and SSA (2014).