Warning: This post is long.
In discussions of the role Social Security plays in providing retirement income, one will often hear some variant of the following: Social Security
“is the major source of income (providing 50% or more of total income) for 66% of the beneficiaries. It contributes 90% or more of income for one-third of the beneficiaries and is the only source of income for 22% of them.”
These are official SSA statistics. Perceptions of Americans' dependence on the Social Security program help influence views regarding the shape of possible reforms. For that reason, as well as others, it is important to have a clear idea what these statistics mean.
The interpretation of these statistics, and their sensitivity to alternate formulations, are the subjects of an important series of papers in the Social Security Bulletin by Lynn Fisher, an economist at the Social Security Administration.
Fisher shows that commonly used figures regarding seniors' dependency on Social Security rely on a series of measurement decisions, any of which could reasonably be decided in other ways. Using plausible alternate methodologies, the percentage of seniors entirely dependent on Social Security – around one-in-five, by the standard measure – could be as low as 3.5 percent.
Fisher examines four potential sources of bias in how we measure dependence on Social Security benefits:
- Unit of measurement: do we count "elderly units" or individuals?
- Benefit reporting: do we use self-reported benefit levels or rely on government data?
- Asset income: do we include only regular income payments, irregular payments such as IRA or 401(k) withdrawals, or even assets that can be liquidated to produce income?
- Non-cash benefits: Should we include non-cash benefits such as energy, food or housing assistance?
Any number of reasonable answers can be made to these questions. What is important is that people understand these choices when they ask "How dependent are retirees on Social Security?"
Unit of measurement: The SSA measures of dependence are expressed in terms of “aged units.” Aged units treat each marital unit (married couple or nonmarried individual) as one unit. A non-married individual has only his or her own income and demographic attributes.
How can this affect measured levels of dependence on Social Security? In two ways. First, single individuals tend to have lower incomes, and therefore be more dependent on Social Security, than do married couples. However, since both a single individual and a married couple count as one unit, this can overstate the percentage of individuals who are dependent on Social Security. For instance, if a single person was entirely dependent on Social Security while a married couple was not, on a ‘aged unit’ basis 50% would be wholly dependent on Social Security while on an individual basis only 33% would be.
Second, a non-married individual may share a household with other individuals, but the aged unit does not include the resources of these non-married cohabitants. (Thus, the ‘aged unit’ measure is not as broad as a ‘household’ measure.) If non-married cohabitants share incomes and costs, this can cause overstatement of unmarried individuals’ dependence on Social Security.
Using the individual as the unit of reporting and assuming that family income is shared, the percentage of seniors wholly dependent on Social Security drops from around one-fifth to around 13%.
Asset income: As the pension world shifts from traditional defined benefit plans to defined contribution plans, in which individuals would draw down their account balances to fund retirement expenses, one would expect that the share of seniors reporting asset income would increase. The measured percentage has actually decreased from 1991-2000, but this may be due to the limitations of the CPS survey data SSA uses in its calculations of Social Security dependency. Fisher turns to another survey – the Federal Reserve’s Survey of Consumer Finances, which emphasizes measures of asset holdings, to supplement existing data.
Fisher found that SCF data supported the view that receipt of asset income had remained roughly constant from 1991-2000. From this improved measure of asset holdings, she was able to infer receipt of asset income. This previously unreported asset income was relatively small, but concentrated among lower earners. While it does not greatly affect the average level of dependence on Social Security, it would lower the percentage wholly reliant on the program, from around 20% to around 10%.
Survey data: Fisher examines two issues dealing with data. First, how the Census Bureau’s Current Population Survey (CPS), which SSA uses to calculate its dependency statistics, compares to the Survey of Income and Program Participation (SIPP), another Census survey that can be used to calculate the income of the aged. Second, Fisher examines how results differ when survey data regarding receipt of Social Security benefits is replaced with administrative data.
Survey choice: The SSA dependency data are derived from the Census Bureau’s Current Population Survey (CPS). One advantage of the CPS is that a new survey is conducted annually, allowing for more up-to-date data. Another Census survey, the Survey of Income and Program Participation (SIPP), is conducted less frequently but asks more detailed questions. Survey subjects are asked about 70 sources of income, versus 35 in the CPS, and the survey takers check back with subjects four times per year, versus only once in the CPS. The SIPP also asks detailed questions about financial assets, while the CPS does not.
Administrative data: SSA’s figures regarding dependency on Social Security use self-reported levels of Social Security benefits contained in the CPS survey. However, researchers have matched CPS survey results to SSA administrative files to see how accurately individuals can recount their Social Security benefits. For a number of reasons, individuals misreport their Social Security benefits – confusing them with SSI or other benefits; reporting them net of Medicare Part B premiums, etc.
Taken together, using both the SIPP survey data and matching survey findings to administrative data and reduce reported dependence on Social Security benefits. Using 1996 survey data, Fisher found that the percentage 100% dependent on Social Security using the CPS with self-reported benefits was 17.9%; add administrative data to the CPS and dependence fell to 17.3%; use the SIPP survey with self-reported benefits and 100% dependence fell to 8.5%, and using SIPP matched to administrative records dependence fell to 8.4%.
Effects in combination: Fisher examines a number of different methodological questions separately, the sees how they affect measured dependence on Social Security when used in combination. Table 1 provides details: Combining all the issues discussed above and applied to 1996 data, the percentage of individuals over 65 depending on Social Security for 100% of their income declined from 17.9% to 4.8%. The percentage depending on Social Security for 90% of their income declined from 30.4% to 13.7%.
In an appendix, Fisher explores the effects of including non-cash benefits, such as food stamps or housing or energy assistance. These are not cash, but are nevertheless valuable resources. If non-cash benefits are included, the percentage wholly reliant on Social Security declines to 3.5%.
As we consider potential reforms to the Social Security program, it is important to have good information regarding the retirement incomes of current seniors, and for policymakers to understand what existing information really means. Put another way, if someone asks the qualitative question, "What percentage of seniors are totally dependent on Social Security?", the answer can range from as high as 20% to under 4%. The two different answers may well lead to two different policy conclusions.