Why Are So Many Americans on Food Stamps?

The Role of the Economy, Policy, and Demographics

James P. Ziliak

Since its inception fifty years ago, the Supplemental Nutrition Assistance Program (SNAP) has become a central component of the social safety net in the United States. The program is unique in its status as something akin to a universal entitlement; that is, subject to meeting low-income and asset tests, the program does not impose restrictions on eligibility based on age (Social Security), family structure (Temporary Assistance for Needy Families, or TANF), or work history (Disability Insurance and Unemployment Insurance). Today, one in seven Americans receives assistance from SNAP at a cost approaching $80 billion in FY2013, making it the second largest means-tested transfer in terms of cost after Medicaid. For much of its first three decades, SNAP operated in the shadows of the safety net, in terms of both a policy or research interest, especially in comparison to politically contentious programs like TANF and more expensive programs like Medicaid. What has caught the attention of policy makers and researchers alike in recent years is the rapid growth of the program. Since FY2000, the number participating has increased 171 percent and inflation-adjusted spending by 286 percent. This has led to calls for programmatic reforms, ranging from the 2013 House Bill H.R. 3102 that would cut $39 billion over the next decade to wholesale decentralization in the form of a block grant to states with additional work requirements similar to TANF (Secretaries' Innovation Group 2013).

The aim of this chapter is to document the factors underlying the evolution of program participation since 1980. I emphasize three major points in my analysis: changes in the macroeconomy, both cyclical forces from the labor market and secular trends in income inequality; changes in public policies, both food and nonfood related; and shifting demographics of the American household. I begin by describing the socioeconomic and policy climate in recent decades that have had bearing on SNAP participation, followed by a formal empirical analysis of those determinants and detailed simulations of the relative contributions of the economy, policy, and demographics to changes in SNAP participation over time. The results suggest that SNAP is operating effectively as an automatic fiscal stabilizer — nearly 50 percent of the increase in participation from 2007 to 2011 is due to the weak economy — but policy reforms expanding access and benefit generosity also affected participation, accounting for nearly 30 percent of the increase after the Great Recession. The changing demographics of the American household are helping restrain growth in SNAP.

THE ECONOMY

The rush to cut or reform SNAP may be misplaced if growth in the program is primarily due to the weak labor market and economy over the past decade. There is extensive research evidence that SNAP functions effectively as an automatic fiscal stabilizer, meaning that, as the economy and market incomes fall during recessions, participation in SNAP "automatically" rises to smooth consumption, and as market incomes rise during economic expansions, participation falls (Wallace and Blank 1999; Blundell and Pistaferri 2003; Gundersen and Ziliak 2003; Ziliak, Gundersen, and Figlio 2003; Bitler and Hoynes 2010; Klerman and Danielson 2011; Ganong and Liebman 2013). This suggests that, as the macroeconomy improves in coming years, participation and cost of SNAP will decline. Indeed, assuming no changes in law, the Congressional Budget Office (CBO) (2012) projected that by 2022 spending on SNAP would fall 23 percent because of improving labor market conditions.

That such a countercyclical link exists seems transparent in Figure 1.1, which depicts trends in aggregate SNAP participation and seasonally adjusted unemployment rates. Highlighted in the figure are years that include a macroeconomic recession as defined by the National Bureau of Economic Research. It is clear that peaks in SNAP usage coincide (perhaps with a lag) with peaks in unemployment rates over the past three decades. A simple time-series regression of SNAP participation on the unemployment rate and a linear trend yields:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)

where standard errors are reported in parentheses. The r-squared of 0.57 says that unemployment rates explain nearly 60 percent of the variation in aggregate SNAP participation (the trend adds nothing to the model). The coefficient on the unemployment rates implies that for each percentage point increase in the unemployment rate, SNAP increases by 0.86 points, or almost 10 percent on the average SNAP participation rate of 9.2 percent over the 1980 to 2011 period. Although this time-series model is simply illustrative, a robust link between the business cycle and SNAP participation remains even in a more fully specified model as estimated in the following discussion.

Concomitant with large swings in unemployment rates, there has also been a sizable secular shift in the distribution of earnings and income that may bear on SNAP participation in recent decades (Piketty and Saez 2003; Autor, Katz, and Kearney 2008; Burkhauser et al. 2012). Figure 1.2 depicts trends in real median household income as well as the ratio of the 90th percentile of real income to the 10th percentile of real income. The median is a robust measure of the center of a skewed distribution, such as income that is used to signify how the typical household is faring, and the 90/10 ratio is a standard measure of inequality. Since 1980, real median income has increased 22 percent to $52,000, whereas 90/10 inequality increased 30 percent. Unlike the expansions of the 1980s and 1990s, however, which each had peaks in income in excess of the prior cycle, median income at the peak of the 2000s cycle was no greater than the 1990s, and in fact by 2011 any gains in median income since 1999 were lost. The flattening out of median income from 2000 to 2007, followed by a sharp decline in the Great Recession, coincides with a sharp increase in income inequality. The latter was driven by continued increases in income at the top coupled with significant declines in income at the bottom of the distribution.

POLICY REFORMS

There was little change in basic federal eligibility rules in the nearly two decades leading up to the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA, also known as welfare reform). Welfare reform, however, had both a direct and an indirect policy effect on SNAP. Directly, it eliminated eligibility for most legal permanent aliens unless they had at least ten years of work experience or were veterans; it eliminated benefits for convicted drug felons; it limited benefits to three months out of any thirty-six-month period for able-bodied adults without dependents (ABAWDS) between the ages of eighteen and fifty working fewer than twenty hours per week or not meeting other work requirements; it reduced the maximum benefit and froze many deductions used in calculating net income; it allowed states to sanction individuals and households for noncompliance with TANF requirements or child support payments; and it mandated that states adopt the electronic benefit transfer (EBT), replacing paper coupons with debit cards (Gabor and Botsko 1998; Gleason et al. 2001). Indirectly, participation was affected by virtue of the fact that half of the food stamp caseload was categorically eligible for stamps via their receipt of TANF benefits, and as welfare reform pulled people off TANF they also dropped receipt of food stamps, at least temporarily (Ziliak, Gundersen, and Figlio 2003). The other policy change that indirectly affected SNAP, even prior to welfare reform (note that in Figure 1.1 participation started to fall three years before PRWORA), was the 1993 expansion of the Earned Income Tax Credit (EITC) that pulled scores of single mothers into the labor force and off welfare and food stamps (Meyer and Rosenbaum 2001).

As participation in SNAP plummeted over 40 percent from 1993 to 2000, so too did the take-up rate of benefits; that is, the fraction of eligible persons participating fell over 25 percent to just over 50 percent (Leftin, Eslami, and Strayer 2011). A decline in participation does not have to be associated with a decline in take-up because the decline in participation could occur only among those no longer eligible. Take-up rates among eligible seniors have always been low (30 to 35 percent depending on the year), and the decline in take-up after 1993 was driven mainly by children and nonelderly adults (Cunnyngham 2002). The fall in take-up, which is likely a spillover effect of former TANF recipients also leaving food stamps even though they remained eligible, was met with alarm (and controversy) in policy circles.

In response, several new initiatives aimed at program outreach and eligibility expansion were introduced, and in some cases codified in the Farm Bills of 2002 and 2008, giving much more discretion to the states to improve take-up and program administration. States had always been in charge of day-to-day administration of food stamps, along with employment and training programs, and paid for these duties generally with a 50 percent federal cost share. With the 1996 welfare reform and the ensuing farm bills, states now had greater leeway to tailor the programs to be more consistent with local policy objectives. These included liberalizing vehicle asset tests, such as exempting one or more vehicles from the test, to foster welfare-to-work; expanding (broad-based) categorical eligibility, which allowed states to use more generous TANF asset and gross-income tests to determine eligibility; restoring eligibility for legal aliens previously excluded by PRWORA; expanding the option for simplified reporting, which allowed states to relax the frequency and form (that is, phone or online) of benefit recertification; and outreach via advertising campaigns. Not all of the reforms were intended to make access easier. Specifically, in the 1990s states responded to financial incentives offered by the USDA to reduce benefit error rates by requiring more frequent certification, especially in households with workers, such that by 2000, 36 percent of working households faced three-month or shorter recertification intervals, up from 4.7 percent in 1992 (Kabbani and Wilde 2003). Some of the simplified reporting requirements adopted in recent years were intended to offset the hardship associated with more frequent certification. Another policy that restricted access is fingerprint imaging and facial recognition software.

Table 1.1 summarizes changes in the number of states (and the District of Columbia) adopting various SNAP policy options between 2000 and 2011 based on information in the Economic Research Service's SNAP Policy Database. By 2011, forty-one states had implemented broad-based categorical eligibility as a mechanism to join SNAP, compared to only two states in 2000; of those forty-one states, thirty-eight elected to eliminate the SNAP liquid asset test as part of broad-based eligibility. The number of states offering call centers, phone applications, and combined application processes with other transfer programs has exploded, along with the number offering simplified reporting and exempting vehicle assets. As of 2011, forty-nine states used simplified reporting, and forty-five states exempted all vehicle assets from resources in determining eligibility for SNAP. At the same time, states pulled back from imposing short recertification windows of one to three months and restored SNAP eligibility to more noncitizens, especially children.

CHANGING DEMOGRAPHICS

Beyond the macroeconomy and policy changes, there have been important demographic shifts in the U.S. population over the past several decades that could differentially affect the size and composition of the SNAP caseload. On the one hand, the aging of the population suggests that the composition of the caseload should be aging as well. However, because take-up rates among seniors are so much lower than those among adults and children, population aging should put downward pressure on the growth of the caseload; thus it is not clear a priori whether the age distribution of the caseload is likely to change due to population aging. Indeed, because of the decline in marriage the fraction of births to unwed mothers accelerated from 15 percent of live births in 1980 to 40 percent by the mid-2000s (Cancian and Reed 2009; Carlson and England 2011), and, as single-mother families are more likely to be poor, then it is entirely possible that the caseload could be getting larger and younger at the same time.

A similar trend toward a younger, larger caseload could emerge from the growth of the Hispanic population, who, all else being equal, tend to be younger, have lower incomes, and have larger family sizes (Landale, Oropesa, and Bradatan 2006). Another secular trend placing upward pressure on the size of the SNAP caseload is the significant growth of disability, both in the Supplemental Security Income (SSI) and Social Security Disability Income (SSDI) programs (Muller, O'Hara, and Kearney 2006; Autor 2011). Households in which all residents receive SSI automatically qualify for SNAP, but SSDI households and those SSI units where some receive disability and others do not must still meet income and asset restrictions.

Figure 1.3 presents trends in the age composition of households receiving SNAP for children under age eighteen, for adults age eighteen to fifty-nine, and for seniors age sixty and older. The age sixty threshold for seniors is consistent with SNAP policy for eligibility determination. The figure shows that in any given year the participation rate among children is double the rate of adults and triple that of seniors. However, the decline in participation among children and adults during the welfare reform era (1993 to 2000) was substantially higher than among the elderly (46 and 48 percent, respectively, compared to 30 percent), but the subsequent growth from 2000 to 2011 was also higher for those two groups compared to the elderly (95, 138, and 75 percent, respectively). Combined, these changes over the past decade have resulted in a shift in the age composition of households receiving SNAP away from children and elderly and toward adults, as shown in Figure 1.4. In all years prior to the Great Recession about 55 percent of SNAP households consisted of children and the elderly, but by 2009, a slim majority were nonelderly adults.

Coincident with the shift in the age composition of households with SNAP is the shift in the composition toward smaller households and those with multiple generations. Figure 1.5 shows that beginning in the welfare reform era there has been a secular decline in the fraction of SNAP households containing three or more persons and a rise in the fraction of one- and two-person SNAP households. At the same time, households receiving SNAP, like the general population, are increasingly likely to contain multiple generations, as seen in Figure 1.6. A multigeneration household is one that contains two or more adult generations, with or without a grandchild, or a grandparent and grandchild household ("skipped generations"). The upward pressure in multigeneration households stems more from the addition of adults to the household than children, which is consistent with the trend toward more two-person SNAP households.

A possible concern with the shift toward smaller, prime-age adult households is that it may coincide with an increasing fraction of the SNAP caseload headed by individuals out of the labor force. A more welfare-reliant population could affect public support for SNAP in light of the centrality of work requirements and time limits that fundamentally altered the TANF program during welfare reform (Ziliak 2009). In fact, Figure 1.7 shows that the share of SNAP households headed by a person out of the labor force has been very stable for the past two decades, averaging about 53 percent. The growth has been most rapid among full-time, full-year workers, as well as part-time, full-year workers. In other words, the heads of an increasing share of SNAP households have a very strong attachment to the labor force. Furthermore, Figure 1.8 shows that the fraction of SNAP households headed by a high school dropout has plummeted by more than half to under 30 percent since 1980, and by 2011 more than a third of SNAP households were headed by someone with some college or more. Figure 1.9, which depicts the distribution of SNAP households by household income in relation to the federal poverty guideline given the household size, shows that since the mid-1980s the composition of SNAP households has trended toward those with annual incomes above the poverty line. This suggests that SNAP has evolved into a work supplement for educated, near-poor households.