The Downs and Ups of the SNAP Caseload: What Matters?*

Stacy Dickert-Conlin, Katie Fitzpatrick, Laura Tiehen

February 2015

Please do not cite without permission of the authors.

Abstract: Since the early 1990s, states have received unprecedented flexibility to determine Supplemental Nutrition Assistance Program (SNAP) eligibility and program administration. We estimate that state SNAP policies accounted for 40 percent of the predicted caseload decline between 1993 and 2000 – primarily through the eligibility restrictions on noncitizens. More recent eligibility expansions and reductions in transaction costs explain 20 percent of the 2000-2011 caseload increase. The state unemployment rate plays a strong role in caseload changes over the study period, accounting for more than 60 percent of the predicted caseload decline between 1993 and 2000 and increase between 2007 and 2011.

                                                            * The authors thank Brian Stacy for excellent research assistance. They also thank Craig Gundersen, Lucie Schmidt, Michele Ver Ploeg, and participants at the Association for Public Policy Analysis and Management Conference, the National Tax Association Conference, the Annual Welfare Research and Evaluation Conference, and a USDA Economic Research Service seminar for helpful comments. This research was supported by a cooperative agreement through the U.S. Department of Agriculture’s Economic Research Service. The views and opinions expressed in this article are those of the authors and may not be attributed to the Economic Research Service or the USDA.  

1  

I. Introduction

The Supplemental Nutrition Assistance Program (SNAP) is one of the

largest social safety net programs in the United States. In 2014, an average of 46.5

million individuals in 22.7 million households received SNAP benefits each

month, and federal spending on the program totaled $74.1 billion

(http://www.fns.usda.gov/pd/supplemental-nutrition-assistance-program-snap).

Originally known as the Food Stamp Program, SNAP received permanent

legislative authority in 1964. Congress typically reauthorizes the program every

five years, most recently in the Agricultural Act of 2014. Because spending on

SNAP was at historically high levels during the Congressional negotiations over

the Agricultural Act of 2014, a number of proposals sought to rein in SNAP,

primarily by eliminating the ability of states to expand SNAP eligibility

requirements.

State flexibility in program rules and administration is relatively new.

Throughout much of its history, SNAP had nationally uniform program eligibility

standards and benefit levels, and states had little latitude in program

administration. Beginning with the 1996 welfare reform legislation and

significantly expanding with the 2002 Farm Bill, legislative and regulatory

changes gave states increased flexibility to simplify program administration and

increase program access, especially for low-income working families. Recent

Congressional testimony by Audrey Rowe, Administrator of the USDA agency

that oversees SNAP, described efforts to “assist States with workload

management while easing the burden of the application process for recipients,

many of whom are new to the program” (Rowe, 2012). Rowe’s testimony

2  

highlighted strategies such as aligning eligibility requirements across programs

and streamlining interview processes.1

The recent caseload growth raised concerns about the extent to which the

unprecedented policy freedom provided to states increased SNAP caseloads and

expenditures. 2 Figure 1 shows that, historically, SNAP was a countercyclical

program; the caseload increases during recessionary times and declines during

economic expansions. One notable exception occurred during the recovery from

the 2001 recession, when the unemployment rate dropped, but the SNAP caseload

continued to increase. While the rise in the number of low income individuals

may explain a portion of the caseload increase, the role that specific state policies

played in the program’s growth is not fully understood. Although the Agricultural

Act of 2014 did not eliminate state flexibility in program administration, the

intensity of the deliberations regarding SNAP suggest that these state policy

options will be a subject of continued debate. The goal of this paper is to

understand the effect of specific state policy options and economic conditions on

SNAP caseloads.

The decline of the food stamp caseload after the Personal Responsibility

and Work Opportunity Reconciliation Act (PRWORA), the landmark 1996

welfare reform legislation that replaced Aid to Families with Dependent Children

(AFDC) with Temporary Assistance to Needy Families (TANF), prompted a

number of studies that focused on the relative contributions of welfare reform and

the improving economy to explain the decrease. Across studies, there is a large

variation in the portion of the food stamp caseload decline explained by welfare

reform, from roughly zero to as much as 40 percent (Figlio et al., 2000; Ziliak et

                                                            1 Available at: http://appropriations.house.gov/uploadedfiles/hhrg-112-ap01-wstate-arowe-20120228.pdf 2 See the following for examples of media commentary on the proceedings: Tanner (2013) and Heritage Foundation (2013). 

3  

al., 2003; Blank and Wallace 1999; Currie and Grogger, 2001). Researchers

attribute between 6 and 20 percent of the Food Stamp caseload decline to

macroeconomic conditions (Wallace and Blank 1999; Currie and Grogger,

2001).3 Related research using individual data finds that specific policies such as

requiring participants to more frequently verify their continued eligibility and

exempting the value of vehicles from the eligibility critieria, affect food stamp

participation and caseloads (Currie and Grogger, 2001; Mickelson and Lerman,

2004, McKernan and Ratcliffe, 2003; Kabbani and Wilde, 2003; Ratcliffe et al.

2008).

Recent work focuses on the factors behind the increase in the food stamp

caseload since 2001. Researchers find SNAP participation to be very responsive

to changes in economic conditions, with a one percentage point increase in the

unemployment rate associated with increases in SNAP participation ranging from

4 to 11 percent (Bitler and Hoynes, 2010; Bitler and Hoynes, 2013; Ganong and

Liebman, 2013; Klerman and Danielson, 2011; Ziliak, 2013). Using individual-

level panel data, Ratcliffe et al. (2008) and Ribar et al. (2008, 2010) find evidence

that policies that influence the transaction costs of SNAP participation affect

program take-up. Overall, estimates from the research most closely related to

ours suggest that economic conditions account for 27 to 45 percent of the recent

caseload increase, while new state SNAP policies account for estimated 16 to 35

percent of the increase (Klerman and Danielson, 2011; Ziliak, 2013). The

primary focus of Klerman and Danielson (2011) is the role of welfare reform,

SNAP policies, and economic conditions on the composition of state, monthly

SNAP caseloads from 1989 to 2009. Ziliak (2013) seeks to explain SNAP

                                                            3 Other research on food stamp caseloads such as Heflin (2004), Richburg-Hayes and Kwakye (2005) and Tschoepe and Hindera (1998) generally finds similar results.

4  

caseload changes using annual, household survey data from 1980 to 2011. Our

research extends the literature on SNAP caseload dynamics by examining a

broader set of SNAP policies than previously available and using over 20 years of

monthly SNAP administrative data that can precisely measure program

participation and capture the timing of state policy implementation.

Our empirical approach is straightforward. We rely on variation across

states and over time in economic conditions and policy from January 1990

through December 2011 to identify the role of these factors in changes in state-

level monthly SNAP caseloads. We broadly group thirteen policies into several,

non-mutually exclusive, types: those that affect SNAP eligibility, those that affect

the transaction costs associated with SNAP participation, those that affect the

stigma of SNAP participation, those that affect outreach towards potentially

eligible non-participants, and those that reflect other means-tested transfer

programs. We compare the estimated effect of economic conditions and these

state-level policies across static and dynamic first-differenced models with

differing numbers of lags. We also compare our results between measuring the

caseload based on the number of households receiving SNAP in each state and the

number of individuals receiving SNAP in each state.

To preview our results, we find across specifications that the economy

plays a central role in determining state caseloads but state-specific SNAP

policies also contribute to the ups and downs of the caseload over the last two

decades, both through policies that affect eligibility and policies that affect the

costs of participation for those eligible. Specifically, we find that restricting the

eligibility of legal noncitizens has a large and significant chilling effect on the

caseload. Broad-based categorical eligibility, a policy that waives the SNAP asset

test and increases income limits for many households, has a positive effect on

caseload growth, but the effect is relatively small. Exempting vehicles from the

SNAP asset test has a small effect in the short-run but the effect grows over a

5  

longer period. Providing transitional benefits to households leaving TANF has a

significant, positive effect on the SNAP caseload.

We find that reducing the burdens associated with SNAP recertification

and reporting requirements for income changes is associated with higher

caseloads. Allowing households to submit applications online appears to have a

small positive effect on the caseload, but estimates are not statistically significant

in the dynamic specifications. Stigma, in the form of requiring SNAP applicants

to be fingerprinted, has large negative effects on the caseload. In contrast,

distributing benefits via electronic benefit transfer (EBT) cards is not robustly

correlated with caseloads. We find little effect of federally funded radio and TV

outreach ads on caseloads. Finally, policies related to other means-tested

programs contribute to SNAP caseload changes, as both the expansion of state

and federal EITC programs and welfare reform both had a long-term, negative

effect on the caseload.

The next section of the paper describes SNAP and its recent history.

Section III considers the caseload data and how to parameterize the institutional

details to capture their effect on caseloads. Section IV describes the estimation

results and Section V presents results from policy simulations. Section VI

concludes.

II. SNAP: Institutional Details and Recent Program History

SNAP is designed to increase the food purchasing power of households

with low levels of monthly income and assets. In 2014, SNAP participants

received an average of $125 in monthly benefits per person

(http://www.fns.usda.gov/pd/supplemental-nutrition-assistance-program-snap). In

contrast with many other programs serving low-income households, SNAP

eligibility does not generally depend on family structure, age, or disability status,

6  

so benefits reach a broad range of economically disadvantaged households.4 The

basic eligibility criteria for SNAP includes household gross income of 130

percent or less of the federal poverty level and household net income (gross

income less certain deductions, such as a standard deduction and deductions for

earned income, dependent care, utilities, and more) of 100 percent or less of the

federal poverty level.5 Additionally, households must have $2,000 or less in

assets. SNAP determines monthly benefits by subtracting 30 percent of household

net income from a maximum benefit level that depends on household size.

Nationwide, over 80 percent of eligible individuals participated in SNAP in 2012

(Eslami, 2014) but SNAP participation rates among eligible persons (take-up

rates) vary widely across time and states, from a low of roughly 57 percent in

California and Wyoming to over 90 percent in eight U.S. states (Cunnyngham,

2014).

SNAP underwent a number of changes in the past two decades that altered

the basic eligibility standards and program administration in ways that could

affect both the difficulty in enrolling in and staying on the program. Table 1

illustrates the increasing variation in state-level SNAP policies. The 1996 welfare

reform legislation made direct changes to SNAP eligibility, imposing strong

restrictions on SNAP benefits to able-bodied adults without dependents

(ABAWDs) and eliminating the eligibility of legal noncitizens to receive SNAP.

When welfare reform eliminated federal SNAP eligibility for legal noncitizens, a

handful of states created and maintained state-funded food assistance programs

for those that were otherwise ineligible for federal SNAP benefits. Subsequent

legislation reinstated eligibility for legal noncitizen children, legal immigrants in

                                                            4 There are certain restrictions on the receipt of food stamps by legal immigrants and able-bodied adults without dependents (ABAWDs). Households receiving Supplemental Security Income (SSI) or Temporary Assistance to Needy Families (TANF) are categorically eligible. 5 Households with elderly or disabled individuals have slightly more generous program rules. 

7  

the country for at least five years, and some specific legal immigrant groups, such

as refugees. Still, changes to the citizenship rules lingered, particularly as some

states eliminated or scaled back their state-funded food assistance programs. By

2011, although all otherwise eligible legal immigrant children could receive

SNAP in all 51 states, nonelderly legal immigrant adults faced at least some

restrictions on their SNAP eligibility in all but three states.

Welfare reform also had an indirect effect on SNAP participation because

of the mechanical link between AFDC and SNAP eligibility and participation:

specifically, AFDC recipients were “categorically eligible” to receive food

stamps. The sharp declines in TANF, AFDC’s replacement, recipients after

PRWORA decreased the number of households that were categorically eligible

for SNAP. Even among those who remained eligible for TANF, most TANF

funds provided in-kind benefits, such as child care, which did not confer

categorical eligibility for SNAP. The sharp declines in SNAP caseloads

following welfare reform prompted regulatory and legislative changes to expand

eligibility and reduce the burden of applying for SNAP. States were given more

flexibility to simplify administration and increase program access, especially for

low-income working families. In late 2000, states were authorized to extend

SNAP categorical eligibility to households that receive noncash, as well as cash,

TANF benefits. Over time, some states adopted a policy referred to as “broad-

based categorical eligibility,” which allows the state to remove the federal SNAP

asset restriction and increase the income limit for most low-income households in

the state. The number of states using broad-based categorical eligibility increased,

particularly after 2006, and 41 states adopted the policy by 2011.

The post-PRWORA era saw additional reforms affecting eligibility,

particularly the treatment of vehicles as household assets. States received the

ability to align their definitions of income and assets with their TANF or

Medicaid programs, as long as these definitions are less restrictive than the

8  

federal SNAP definitions. Some states chose to exclude the value of one or all

vehicles in their TANF cash assistance program, and used the increased flexibility

to do the same in determining SNAP eligibility. States were also given the option

to provide an automatic SNAP benefit to families leaving TANF, for up to five

months after their TANF exit. In 2011, 21 states offered transitional SNAP

benefits to TANF leavers, with all of these states choosing to offer the full five

months of benefits.

Throughout the post-PRWORA period of emphasizing labor force

participation among lower income populations, state policies also changed with

respect to how recipients must establish their continued eligibility for the

program. State recertification periods—the number of months that can elapse

before a SNAP household has to recertify eligibility – shortened dramatically

through the 1990s as states sought ways to avoid federally-administered penalties

for benefit calculation errors (Rosenbaum, 2000). Short recertification periods

increase the difficulty of continued program participation for working families, in

particular, who may need to take off from work to complete the recertification

process. The 2000s saw states lengthen the recertification periods in an effort to

increase access to SNAP, particularly among low-income working households.

In between certifications, households who receive food stamps must report

changes in circumstances that may affect their eligibility or monthly benefit.

Before 2000, recipients had to report monthly and within ten days of a change in

circumstances that could affect eligibility. Since 2000, all states have the option to

allow SNAP recipients with earned income to report income changes on a

quarterly or semi-annual basis, rather than each month or each time a change in

circumstances occurs (U.S. GAO, 2002). Semi-annual, or “simplified” reporting

decreases reporting burdens on SNAP recipients because it requires households to

report changes to their financial circumstances within reporting periods only if the

changes would make them ineligible for the program.

9  

Other SNAP policy changes since the 1990s affected how states utilize

technology in program administration. Beginning with Maryland in 1989, states

rolled out electronic benefit transfer (EBT) cards that can be used like a bank

debit card to purchase eligible food to reduce benefit fraud, program

administration costs, and the stigma of participation. The welfare reform

legislation imposed a deadline on states’ implementation of EBT for food stamp

benefits (USDA, 2006) and by 2004 every state paid all SNAP benefits via EBT.

Also beginning in the mid-1990s, a few states implemented fingerprinting

during the application process to reduce fraud associated with enrolling for

benefits under multiple names. In January 1996, New York was the only state

with a statewide fingerprinting requirement, but by January 2001, three additional

states –Arizona, California, and Texas – required fingerprinting statewide. While

only these four states ever adopted this requirement they comprise a large portion

of the national caseload: accounting for one quarter of the total national SNAP

caseload in 2001. Over time, states eliminated fingerprinting requirements and by

January 2012, Arizona was the only state to have a fingerprinting requirement.

In the early 2000’s, states began utilizing the internet in program

administration. A growing number of states allowed applicants to complete and

submit an application for SNAP benefits over the internet. By the summer of

2012, 35 states allowed all applicants to submit online applications and in three

additional states (California, Indiana, and New York), residents in some counties

could submit their application online. States differ in whether or not they allow

for an electronic signature or if applicants must mail in their signature to complete

an application. Some states also provide SNAP recipients with online access to

information about their case, recertification and changes in income reporting.

Finally, the emphasis on program access during the 2000s manifested

itself in a trend towards increased outreach to potentially eligible non-participants.

The federal government reimbursed states with a federally approved SNAP

10  

outreach plan for half of their administrative costs on outreach spending. Each

fiscal year from 2001 to 2009, the federal government also awarded outreach

grants to non-profit organizations to increase participation among eligible non-

participants, particularly those with historically low participation rates such as

Hispanics and the elderly. In addition, a large-scale federally funded radio and

television advertising campaign in selected media markets, launched in 2004

sought to raise awareness about SNAP among potentially eligible non-

participants.

III. Data and Identification Strategy

We use the significant temporal and spatial variation in SNAP policies

described in the previous section as the source of our identification in estimating

the effect of these policies on the SNAP caseload. We measure the SNAP

caseload at the state, monthly level from January 1990 through December 2011.

We use two measures: the number of households and the number of individuals

receiving benefits.6 We normalize each state-month individual caseload by the

state’s population and each state-month household caseload by the number of

households in each state.7 Measuring a state’s caseload at monthly intervals

(versus annual intervals) can better capture the response to the precise timing of

the change in policy and coincides with the timing of benefit payment.

Administrative caseload data capture the measure of interest to policymakers and

avoid the substantial underreporting of SNAP receipt in household surveys

                                                            6 We are very grateful to Rebecca Blank and Geoffrey Wallace for providing us with the household level data between 1990 and 1998, and to Nadine Nichols of the Food and Nutrition Service, USDA for providing us with the additional data. 7 The Census Bureau provides estimates of the state population in July of each year. The number of households in each state uses data from the March CPS, provided by IPUMS-CPS. For both estimates, we assume a constant rate of change and smooth each population throughout the year.

11  

(Meyer et al., 2009). The household caseload provides estimates for the relevant

SNAP benefit unit while the individual caseload captures the number of persons

that rely on benefits to meet their food needs.

The first set of covariates reflects state-level SNAP policy changes

described in the previous section. Table 1 describes each of the covariates in

selected years over our time period. To capture policies that affect mechanical

eligibility for benefits, we include a vector of dichotomous variables. The first

variable takes a value of one if noncitizens under age 65 face any restrictions to

their SNAP eligibility. Limits on the eligibility for noncitizens should reduce the

caseload through entry and exit, particularly in states with large immigrant

populations. We also include dichotomous variables for state vehicle policies,

parameterized as two dichotomous variables representing state policies to exempt

at least one but not all vehicles in the household and state policies to exempt all

vehicles in a household (state policies that do not exempt any vehicles serve as the

omitted group); state policies providing up to five months of transitional SNAP

benefits to those leaving TANF; and state adoption of broad-based categorical

eligibility. We expect each of these policies to contribute to growth in the SNAP

caseload by mechanically increasing the size of the eligible population.

Another set of variables captures SNAP policies that may affect stigma

and transaction costs among those eligible for SNAP. While not mutually

exclusive from those that affect stigma, we begin by considering policies that

affect transaction costs. We include measures to capture the implementation of

short recertification periods, defined as the proportion of working households

subject to a certification period of three months or less; state reporting policies,

defined as a dichotomous variables indicating the implementation of simplified

(or biannual) reporting (with quarterly, monthly or change reporting as the

omitted group); and, state online application availability, defined as a

dichotomous variable for states that allowed submission of an application for

12  

SNAP benefits online. More frequent reporting and recertification policies should

reduce the caseload by encouraging exit from the program at the reporting or

recertification interval. Online applications should increase the ease of completing

application by eliminating the need to travel to the local SNAP office to apply for

benefits and encouraging entry into the program.

For policies affecting stigma, we include state EBT policies, defined as the

portion of state SNAP benefits issued via EBT, and the implementation of

fingerprinting requirements, defined as a dichotomous variable indicating the

statewide use of fingerprinting during the application process.8 Use of EBT may

increase caseloads by reducing the stigma of redeeming benefits, although the

new technology may pose difficulties for some individuals and grocery stores to

adopt. We presume that fingerprinting will reduce access by increasing the stigma

associated with entry into the program.

We examine outreach efforts, which seek to provide information on the

program to potentially eligible but non-participating populations. To capture this,

we examine federally funded SNAP radio and TV advertisements. 9 The media

campaigns, which we characterize with a dichotomous variable indicating the

airing of a federally funded TV or radio ad campaign in any media market in the

state, may increase SNAP participation by reducing the costs of learning about the

program and its eligibility criteria, in addition to reducing the transaction and

stigma costs associated with the program. Moreover, because state willingness to

accommodate any increase in applications was one criterion for deciding

                                                            8 New York State ended fingerprinting statewide in 2007 but New York City continued fingerprinting SNAP applicants through May 2012. We only consider fingerprinting policies in effect statewide. 9 Previous research captures outreach through state outreach spending under a federally approved outreach plan (Kabbani and Wilde 2003; Ratcliffe et al. 2008; Ratcliffe et al. 2011; Ziliak 2013). This measure of outreach is problematic because it only captures state-level spending each fiscal year that is reimbursed by the federal government, rather than the total of state-level outreach spending and it would require smoothing the annual data to fit our monthly observations. Because of these measurement challenges, we chose not to include this variable in our estimates. 

13  

placement of these ads, this variable also captures the willingness of states to

expand the caseload through outreach and other practices that are difficult to

measure.

Lastly, we control both for non-SNAP policies targeted at low-income

populations and for the economic environment. Because of the close link between

cash welfare programs and SNAP, we include a dichotomous variable indicating

the earliest implementation of either an AFDC waiver or TANF. Acknowledging

the importance of the Earned Income Tax Credit (EITC) to the budgets of low-

income households, we capture the large variation in the generosity and timing of

EITC receipt during the calendar year by multiplying the real EITC maximum

credit value for a family with two children (based on the combined federal and

state EITC, in thousands) by the portion of annual federal EITC payments made

in each month. Table 1 shows the dramatic growth in state level EITC payments

and Appendix Table 1 shows variation in annual payments over the calendar year.

We control for state economic conditions with the state unemployment rate, as

shown in Table 1.

In sum, to identify the relationship between SNAP caseloads, state SNAP

policies, and other low-income policies, we use variation arising from differences

across states and over the 252 months of our data. As shown in Table 1, dramatic

state by time variation in policies and conditions exists. Figure 2 represents the

caseloads of the six states with the largest household caseloads: California,

Florida, Michigan, New York, Pennsylvania and Texas. Like the national picture

in Figure 1, in earlier years the state unemployment rate and the SNAP caseload

move together. Starting around 2001, however, there is more cross-state variation

in the relationship between the unemployment and caseload trends. California,

Florida, and Texas experienced strong declines in unemployment between 2001

and 2007, but caseload increases were more pronounced in Texas than in the other

two states. Michigan and Pennsylvania experienced relatively strong and steady

14  

increases in their SNAP caseloads between 2001 and 2007, despite a falling

unemployment rate in Pennsylvania and a relatively flat unemployment rate in

Michigan. The onset of the Great Recession led to sharp increases in SNAP

caseloads in all six states, but differences again emerge in the timing and

magnitude of the later decreases in unemployment rate and slowing of caseload

growth. It appears possible that state-specific SNAP policies may play a role in

determining the realization of the state SNAP caseload.

To understand the relationship between policy and economic variables and

food stamp caseloads, we adopt two specifications. The first is a static model and

the second is a dynamic model that allows past realizations of the caseload to

have a direct effect on the current caseload because of sluggish adjustment to

changes in the economic environment or policy conditions.10

The dependent variable is the natural logarithm of the per capita SNAP

caseload (SNAP_ PerCap ) in state s in month t (t=1, …, 252). Caseloads are

measured as the number of recipient households in one set of regressions and

number of individuals receiving SNAP in another. SNAP_Policies is the vector of

state-level SNAP policies; LowInc_Policy is a vector that includes our EITC

variable and our control for the earliest implementation of a major AFDC waiver

or TANF, and Unemp is the monthly state unemployment rate. All policy and

unemployment variables include a selected number of lags, l, beginning with the

month prior to the caseload measure. In separate regressions, we estimate 12 and

24 months of lags (L=12 and L=24). Including a large number of lags in our

independent variables allows for the fact that it may take time before the effects of

policy can be measured in the state caseload. We control for characteristics

common to a state over time with a state fixed effect and allow a within state

trend over time by interacting the state fixed effect with a time trend and the time

                                                            10 Our dynamic specification draws heavily on the methodology used in Ziliak et al. (2000).

15  

trend squared. To control for seasonal variation across the calendar year, we

include a dummy for the calendar month m (m = 2, …, 12). We also include a

dummy variable, st

,  in state-months affected by the Gulf Coast Hurricanes in

the fall of 2005, particularly Hurricane Katrina, to control for the temporary

spikes in the affected states associated with the Disaster SNAP program.11

Before estimating, we first difference each equation to address concerns of

nonstationarity. After first differencing, we have the following equations for the

static [equation (1)] and dynamic models [equation (2)]:

ststmt

ssltsUnemp

ltsPolicyLowInc

ltsPoliciesSNAP

stPerCapSNAP

L

ll

L

ll

L

ll

1

11

)(

)(_

)(__log

(1)

ststmt

ssltsUnemp

ltsPolicyLowInc

ltsPoliciesSNAP

ltsPerCapSNAP

stPerCapSNAP

L

ll

L

ll

L

ll

L

ll

1

11

1

)(

)(_

)(_

)(_log_log

(2)

Under the first differenced specification, the state fixed effects, s

and

ts

represent deviations from state trends. We use Driscoll-Kraay standard errors

                                                            11 According to Hanson and Oliveira (2007), there was a 15 percent spike in SNAP caseloads in areas affected by Hurricanes Katrina, Rita, and Wilma. Based on Hanson and Oliveira’s analysis, we set this dummy variable equal to one in September, October, and November 2005 for the following states: Alabama, Florida, Mississippi, Louisiana, and Texas. These state-month combinations represent both SNAP Disaster Assistance claims by residents remaining in the Gulf Coast after the hurricane, as well as claims by Hurricane Katrina evacuees.

16  

to account for spatial and temporal dependence in the covariance matrix (Driscoll

and Kraay, 1998).

IV. Results

Table 2 and Table 3 provide the results of empirical estimation of the

household and individual per capita caseload, respectively. Column 1 of each

table provides static estimates using 12 lags of the explanatory variables while

Column 2 provides static estimates using 24 lags of the explanatory variables.

Columns 3 and 4 provide results for the dynamic specification, which includes

either 12 or 24 lags of the dependent variable. All reported estimates are the long-

run effects and we report p-values from the F-test of joint significance below each

estimate.12

Generally, specifying the dependent variable as either the household

caseload or the individual caseload tends to produce similar long-run estimates.

This suggests that the SNAP policies we consider do not, by and large, have

differing effects on larger households versus smaller households. Models that

include 24 lags rather than 12 lags tend to have larger and more statistically

significant long-run effects, implying that it can take more than a year for the

SNAP caseload to respond to changes in policy and economic conditions.

We turn now to a specific discussion of the estimated effects of SNAP

policies, beginning with SNAP policies affecting eligibility. We first discuss the

effect of vehicle exemption policies, where the omitted category is that no

                                                            

12 The long-run effects for each explanatory variable are given by L

j1 for static models

and

L

L

j

1

1

1

for dynamic models. Note that θ is the coefficient estimate of the lagged

caseload.

17  

vehicles are exempted from the SNAP asset test. We generally find no significant

effect of exempting vehicles after 12 months, except for a small and surprisingly

negative effect of exempting one but not all household vehicles on the household

caseload. After 24 months, we find consistently positive and statistically

significant effects of the vehicle exemption policies on both the household and

individual caseloads. The estimated effects are larger in the dynamic than in the

static model. For example, exempting at least one but not all vehicles from the

SNAP asset test increases the household SNAP caseload by 2.2 percent after 24

months in the static model and by 5.9 percent in the dynamic model. Likewise,

exempting all vehicles increases the household SNAP caseload by 2.0 percent in

the static model and 6.7 percent in the dynamic model. The point estimates of the

coefficients on the two vehicle policy variables are quite similar within each

model, suggesting that a policy to exempt all vehicles does not lead to greater

increases in the SNAP caseload than a policy to exempt one but not all vehicles.

Prior researchers find mixed evidence on the vehicle exemption policies, with

Ratcliffe et al. (2008) and Klerman and Danielson (2011) finding positive effects

on SNAP participation and Hanratty (2006) and Ziliak (2013) finding no effect.

We estimate that other policies intended to extend SNAP eligibility to

more types of households increase caseloads. Providing transitional benefits to

TANF leavers increases the caseload per capita between 2.7 percent and 5.2

percent after 12 months and between 4.5 percent and 6.4 percent after 24 months.

Broad-based categorical eligibility for SNAP increases the household caseload up

to 5.9 percent and the individual caseload up to 10.5 percent after 24 months. The

similarity of estimates in the model with 12 lags versus 24 lags suggests that most

of the caseload increase in response to the adoption of broad-based categorical

eligibility occurs within one year. Recent studies (Klerman and Danielson, 2011;

Ganong and Leibman, 2013; Ziliak, 2013) also find similar effects of broad-based

categorical eligibility.

18  

In contrast to vehicle exemptions, transitional benefits and broad-based

categorical eligibility, the eligibility restrictions on noncitizens reduce the size of

the eligible population. We find large, negative effects of policies that prevent

nonelderly noncitizens from receiving benefits. The effect is much larger in the

long run, with caseload reduction of 14.8 to 24.8 percent after 24 months of the

rule preventing noncitizens from receiving benefits, compared to 8.6 to 15.6

percent reductions in caseloads after 12 months.

Table 4 shows that the large declines in caseloads in response to the

eligibility restrictions have the expected differential effect in states with large

populations of noncitizens.13 In particular, consider the results in the final column

of Table 4, from the dynamic model of the SNAP household caseload that

includes 24 monthly lags of the explanatory variables. Imposing eligibility

restrictions on nonelderly noncitizens lowers the caseloads by 35 percent in states

that have large populations of noncitizens, relative to 22 percent in states that do

not have large populations of noncitizens. In all cases, the estimated magnitude of

the effect of these restrictions on SNAP eligibility of noncitizens on caseload

appears very large, suggesting these policies affected participation not only by

reducing the pool of eligible noncitizens, but also through a chilling effect that

discouraged participation among eligible noncitizens. Evidence of chilling effects

from the post-welfare reform change in the eligibility rules for legal noncitizens

has also been found for other means-tested programs, including SNAP (Bitler and

Hoynes, 2011; Borjas 2003; Borjas 2004, Fix and Passel, 1999).

                                                            13 To define states as differentially affected, we tabulate the size of the adult population that is not native-born citizens in each state using IPUMS-CPS data from 1996 through 2012. Examining the states with the largest non-native citizen populations in each year, we find the largest populations in California, New York, Florida, Texas, New Jersey, and Illinois and a clear break in the distribution after these states. Specifications that defined states as differentially affected by examining the percent of the population that were not native-born citizens or the percent of the population that were not citizens produce similar results. 

19  

Among policies that affect the transaction costs associated with

participating in SNAP, we estimate that short recertification periods for working

households reduce the caseload, as first shown by Kabbani and Wilde (2003).

With a 10 point increase in the percentage of working households with

recertification periods of three months or less, the SNAP caseload declines by 1.3

to 2.4 percent. Kabbani and Wilde (2003) find similar effects during the 1990s.

Replacing monthly reporting requirements with simplified reporting

policies have a long-run effect increasing the caseload by 2.7 to 5.8 percent after

12 months and similar effects after 24 months. While the availability of online

SNAP applications may reduce the transaction costs of entering SNAP by

eliminating the need to travel to a county office to complete the forms, the

estimated effect is positive but not statistically significant in most specifications.

The point estimates are similar to Schwabish (2012) who finds in the first three

years after the availability of online applications, per capita SNAP participation

increases less than one percent per year.

Turning to policy variables that may affect stigma among the potential

caseload, our results confirm the findings of Ratcliffe et al. (2008) and Ziliak et

al. (2003) who find little to no effect of EBT on SNAP participation. Our long

term estimates are statistically significant only in the models with 24 lags,

however the estimates are small and negative in the static model and small and

positive in the dynamic model.

In contrast, we find that requiring SNAP applicants to be fingerprinted,

which we would expect to increase the stigma associated with benefit receipt, has

a large negative effect on SNAP caseloads. Over a 12 month period, the

implementation of a fingerprinting requirement reduces the caseload by 6.4 to

10.5 percent. After 24 months, a fingerprinting requirement reduces the caseload

by 3.9 to 10.4 percent. As previously discussed, these policies occurred in states

with large SNAP caseloads: Arizona, California, New York, and Texas. These

20  

large negative estimates are consistent with other research (Ratcliffe et al., 2008;

Ratcliffe et al., 2011; Ziliak, 2013) and anecdotal evidence by anti-hunger groups

that fingerprinting requirements dissuade eligible households from applying for

SNAP (California Association of Food Banks 2010; Eligon 2012; Goetz 2009).

The estimates on the effect of having a federally funded TV or radio

campaign in at least one media market in the state are weak and contradictory,

despite anecdotally evidence that the ads generated a large number of requests for

program information and application assistance.. In the dynamic model, the

results are positive but never statistically significant. In the static model, the

estimates are negative, but close to zero. Prior studies using a different measure of

outreach--outreach spending in the state under a federally-approved outreach

plan--also found mixed results. Ratcliffe et al. (2008) find no effect of outreach

spending on SNAP participation, Kabbani and Wilde (2003) find a positive effect

only among working households, and Ziliak (2013) finds a small negative effect.

We find, as expected, that welfare reform has a significant negative effect

on SNAP caseloads, which indicates that the mechanical link between AFDC and

food stamps helps explain the decline in the SNAP caseload in the late 1990s. The

long-run coefficient estimate on the indicator variable of implementation of either

an AFDC waiver or TANF suggests that food stamp caseloads were 4.1 to 8.0

percent lower as a result of welfare reform.

Changes in the EITC are also highly correlated with the monthly SNAP

caseload. The federal EITC expansions – particularly the 1993 expansion – were

intended to ensure that joint receipt of SNAP and EITC would raise employed

families’ after-tax and after-benefit income above the federal poverty level. In

three out of the four models, we estimate a positive effect of the EITC on

caseloads after 12 months. However, we find that EITC benefits are associated

with caseload reductions after 24 months. For every $1,000 increase in the

monthly EITC benefit payout in the state, the SNAP caseload declined by 2.6 to

21  

6.6 percent after 24 months. Our results suggest that in the long run increases in

EITC generosity may influence food stamp recipients to either transition off the

caseload or never apply for benefits. It is possible that the EITC’s influence on

labor supply and earned income may reduce eligibility or the benefits associated

with SNAP participation (for example, Eissa and Liebman 1996). In addition, the

annual EITC payment may reduce SNAP eligibility if EITC recipients invest in

assets with their payment (for example, Goodman–Bacon and McGranahan,

2007). Our results contrast with Mickelson and Lerman (2004) who find some

negative, yet non-robust, effect of the EITC on SNAP participation. Ratcliffe et

al. (2008), however, find results consistent with our estimated effects of the EITC

on SNAP participation.

Finally, the results on the long-run effect of unemployment on SNAP

caseloads imply that a one percentage point increase in the unemployment rate is

associated with a 4.5 to 11.4 percent increase in caseloads. The dynamic

specifications predict that a one percentage point increase in the unemployment

rate results in at least a 7 percent increase in the SNAP caseload, which is larger

than the effect found in most recent studies (Bitler and Hoynes, 2012, 2013;

Ziliak, 2013), but similar to the finding of Klerman and Danielson (2011).

V. Simulations

We use our regression estimates to.calculate the contribution of each

explanatory variable to the predicted change in the caseload over a specified time

period. We first calculate the predicted change in the caseload due to all

observable factors over the chosen time period. We then calculate what the

predicted change would have been, if the factor of interest were held constant at

the baseline month. For example, we calculate the predicted change in the

caseload from July 2000 to December 2011, holding SNAP policies at their July

22  

2000 levels. The difference between the overall predicted change and the

predicted change with SNAP policies held at the baseline level gives us the

contribution of SNAP policies to the predicted caseload change. We estimate the

share of our predicted caseload change that is due to changes in SNAP policies,

Welfare Reform, the EITC and economic and seasonal factors. In each of our time

periods, the actual change in household caseloads is larger than our predicted

change. We report the simulated effect of the policy or economic changes as a

share of the predicted change in caseloads, acknowledging that factors

unobserved to us play a further role in actual caseload changes.

We use the estimates from the dynamic specification where the outcome

variable is the number of SNAP households per capita and the estimates include

24 lags of the independent variables (Column 4 of Table 2). In this dynamic

specification, SNAP policies have a direct effect on future caseloads, through the

lag structure (and the estimates of the vector ), and, therefore, an indirect effect

on future caseloads by affecting the lagged caseloads for future time periods (and

the parameter estimates of the vector ). Our simulations account for both of

these effects through an iterative process of predicting the 51 state SNAP

caseloads, holding the set of covariates of interest constant between a beginning

and ending time. Obviously, we estimate these contribution of the policy

changes with error and so, in order to put standard errors on our estimates, we

bootstrap with random draws from the parameters (specific details are available in

an online appendix)..14

Table 5 shows the results of the simulations. We consider three distinct

time periods: March 1993 to July 2000, which includes the period of welfare

reform, an overall improving economy, and the large decline in SNAP caseloads;

                                                             

23  

July 2000 to December 2011, covering two recessions, increased access to SNAP,

and the dramatic caseload increase; and December 2007 to December 2011,

which marked the beginning of the Great Recession through the end of our data.

During the 1993 to 2000 period the actual caseload declined by 50 percent

and we predict a 40.4 percent decline. We simulate that the changes in the

combined SNAP policies explain 38.1 percent of the 40.4 percent decline our

specification predicts. Nearly all of this predicted effect reflects the restrictions

on eligibility for noncitizens in the post PRWORA era. We also capture changes

in state policies that likely affected transaction costs (shorter recertification

periods) and stigma (increased EBT receipt).15 Overall, these combined policies

affecting transaction costs and stigma explain only 1.4 percent of the predicted

decline, with large standard errors around this small point estimate. Welfare

reform alone, either an AFDC waiver or TANF implementation, explains 10.0

percent of the predicted decline. Expansions in the EITC explain 21 percent of

the predicted decline, but the effects of both the EITC and welfare reform are

imprecisely estimated. In contrast, our simulations suggest that the improvements

in the economy account for 60 percent of the decline in predicted SNAP

caseloads. This estimate is precisely estimated at standard significance levels.

Moving into the more recent time period, the SNAP caseload nearly

doubled between July 2000 and December 2011, increasing by 96.6 percent.

Overall, SNAP policies account for one-fifth of the predicted caseload increase.

In contrast to the 1993 to 2000 time period, the SNAP reforms generally

expanded eligibility and reduced transaction costs and stigma. We estimate that

the SNAP policy changes that expanded eligibility explain 12.1 percent of the

                                                            15 Individual SNAP policy reforms, like the EBT payouts are often very imprecisely estimated, as expected from the regression results in Table 2, and we choose not to report the simulation results with them, although they are available upon request. Although there were some state specific changes in ABAWD policies, the data on their implementation have never been collected consistently, so we do not parameterize these changes in our empirical work. 

24  

dramatic increase, with the adoption of broad-based categorical eligibility

accounting for 4.3 percent of the increase. Reductions in transaction costs and

stigma explain 9.2 percent of the caseload increase. Each of these are precisely

estimated at standard levels. Because there was no change in our indicator

variable for welfare reform during this period, it plays no estimated direct role in

the expansion. Although we estimate the EITC to be a small positive factor in the

caseload explosion, the estimate is again very imprecise. The economy continues

to play a large role, but not as large as in the pre-2000 period, explaining only

39.0 percent of the caseload increase.

Our simulation methodology allows us to estimate the role of SNAP

policies across states in this recent, 2000 to 2011, caseload increase. Returning to

the states highlighted in Figure 2, we are able to show see that because state

policies and unemployment rates varied so greatly, our estimates of the role of

SNAP policies on caseloads also varies greatly. For example, state SNAP

policies explain15 percent of the predicted caseload increase in Michigan and

27.9 percent in New York. Michigan had larger swings in unemployment rates

over this period and New York eliminated their fingerprinting requirement in

2007, while Michigan never had one. Note that our simulation results are based

on the coefficients for the entire United States, so that differences between states

reflect differences in which policies were implemented, not differential responses

to those policies. We include West Virginia and Oklahoma as the two states

where the SNAP policy explain the largest amount of their predicted state

caseload increases, 53.3 percent and 34 percent respectively. Economic

conditions account for more than 20 percent of the predicted caseload increases

over this time in all states, with California and Michigan at 60.2 percent and 47.5

percent respectively.

Given the severity of the Great Recession, our final set of simulations,

returning to Table 5, includes the years 2007 to 2011. In those four years, the

25  

SNAP household caseloads expanded by 56.6 percent and our specification

predicts at 50 percent increase. Perhaps not surprisingly, we estimate that the

economy explained 61.8 percent of the predicted increase. The pace of policy

changes had slowed substantially by 2007, so the SNAP policy reforms explain

only 8.7 percent of the predicted increase, which is swamped by the share

explained by the economic and seasonal factors of 61.8 percent.

VI. Discussion and Conclusion

In this paper, we examine the role of economic conditions and policy

changes in explaining both the 1994-2001 decline in the SNAP caseload and the

2001-2011 increase. We estimate a first-differenced equation explaining state-

level monthly SNAP caseloads for the period January 1990 to December 2011 in

both static and dynamic caseload models. Using more than twenty year time

series of state, monthly caseload data allows us to more precisely identify how

changes in the economy and state policy choices influence its SNAP caseload.

Our estimation results indicate that the recent policies intended to increase

access to SNAP, such as more simplified application processes, increased state

SNAP caseloads. Consistent with previous research, we find that the SNAP

caseload increased as eligibility certification periods increased in the 2000s. In

addition, States’ adoption of less restrictive income reporting policies and asset

tests contributed to the 2000-2011 SNAP caseload increase, although the

measured effects were smaller than that of shortening recertification periods.

Importantly, while the broad-based categorical eligibility policy was a subject of

heated debate in the Congressional deliberations over the Agricultural Act of

2014, we find that it explains less than 5 percent of the recent caseload increase.

The unprecedented freedom states recently acquired to shape SNAP

eligibility and administration is not solely responsible for the recent caseload

26  

changes. Other means-tested programs, including welfare reform and the EITC,

have influenced SNAP caseloads in expected ways, highlighting the importance

of examining the SNAP caseload in the context of the overall policy environment

affecting low-income households. Moreover, the results also indicate a strong

positive relationship between the state unemployment rate and the SNAP

caseload, even after controlling for a large number of policies that increased

access to SNAP in the 2000s. Therefore, although we see a divergence in the

trends in state unemployment rates and the state SNAP caseload, the strong effect

of the unemployment rate suggests that a sustained economic recovery will reduce

the SNAP caseload from its recent historically high levels.

27  

References

Bitler, Marianne and Hilary W. Hoynes. 2010. "The State of the Safety Net in the Post-Welfare Reform Era." Brookings Papers on Economic Activity, (Fall 2010).

Bitler, Marianne and Hilary W. Hoynes. 2011. “Immigrants, Welfare Reform, and

the U.S. Safety Net.” in Card, David and Steven Raphael (eds), Immigration, Poverty, and Socioeconomic Inequality, in press, Russell Sage Foundation, New York, NY

Bitler, Marianne and Hilary W. Hoynes. 2013. "The More Things Change, The

More They Stay The Same: The Safety Net, Living Arrangements, and Poverty in the Great Recession, NBER Working Paper 19449, (September 2013).

Borjas, George J. 2003. "Welfare Reform, Labor Supply, and Health Insurance in

the Immigrant Population.” Journal of Health Economics, 22(6): 933-958. Borjas, George J. 2004. “Food Insecurity and Public Assistance.” Journal of

Public Economics 88(7–8): 1421–443. California Association of Food Banks. 2010. “Finger Imaging: Pointing

California in the Wrong Direction”. Executive Summary. July 6, 2010. http://www.cafoodbanks.org/docs/Finger%20Imaging%20Report%20Executive%20Summary.pdf

Cunnyngham, Karen E. “Reaching Those in Need: State Food Stamp Participation

Rates in 2011.” Alexandria, VA: U.S. Department of Agriculture, Food and Nutrition Service, February 2014.

Currie, Janet and Jeffrey Grogger. 2001. "Explaining Recent Declines in Food

Stamp Program Participation." Brookings-Wharton Papers on Urban Affairs: 203-244.

Driscoll, John C. and Aart C. Kraay. 1998. “Consistent Covariance Matrix

Estimation with Spatially Dependent Panel Data.” The Review and Economics and Statistics 80: 549-560.

Eissa, Nada and Jeffrey B. Liebman. 1996. “Labor Supply Response to the Earned

Income Tax Credit,” Quarterly Journal of Economics 111(2): 605-637.

28  

Eligon, John. 2012. “Cuomo Pushing City to End Food Stamp-Fingerprinting”

New York Times. May 17. http://www.nytimes.com/2012/05/18/nyregion/cuomo-seeks-to-end-fingerprinting-for-food-stamps-in-nyc.html

Eslami, Esa. “Trends in Supplemental Nutrition Assistance Program Participation

Rates: Fiscal Year 2010 to Fiscal Year 2012.” Alexandria, VA: Food and Nutrition Service, U.S. Department of Agriculture, July 2014.

Figlio, David N, Craig Gundersen and James P. Ziliak, 2000. “The Effects of the

Macroeconomy and Welfare Reform on Food Stamp Caseloads.” American Journal of Agricultural Economics, vol. 82, no. 3, August: 635-41

Fix, Michael E. and Jeffrey S. Passel 1999. “Trends in Noncitizens’ and Citizens’

Use of Public Benefits Following Welfare Reform: 1994–1997.” Washington, D.C.: The Urban Institute.

Ganong, Peter and Jeffrey B. Liebman. 2013. “The Decline, Rebound, and

Further Rise in SNAP Enrollment: Disentangling Business Cycle Fluctuations and Policy Changes,” NBER Working Paper 19363, August.

Goetz, Kaomi. 2009. “Fingerprinting for Food Stamps Under Scrutiny”

National Public Radio. December 18. http://www.wbur.org/npr/121560340/fingerprinting-for-food-stamps-under-scrutiny

Goodman-Bacon, Andrew and Leslie McGranahan 2007. “How Do EITC

Recipients Spend Their Refunds?” Economic Perspectives, Federal Reserve Bank of Chicago, Vol. 31, First Quarter 2007.

Hanratty, Maria J. 2006. "Has the Food Stamp Program Become More

Accessible? Impacts of Recent Changes in Reporting Requirements and Asset Eligibility Limits." Journal of Policy Analysis and Management 25 (3):603-621.

Hanson, Kenneth and Victor Oliveira. 2007. “The 2005 Gulf Coast Hurricanes'

Effect on Food Stamp Program Caseloads and Benefits Issued.” Economic

29  

Research Report Number 37, USDA Economic Research Service. http://www.ers.usda.gov/publications/err37/err37fm.pdf

Heflin, Coleen. 2004. “Who Exits the Food Stamp Program after Welfare

Reform?” IRP Discussion Paper 1279-04. Kabbani, Nader S., and Parke E. Wilde. 2003. "Short Recertification Periods in

the U.S. Food Stamp Program." Journal of Human Resources 38 (Special Issue on Income Volatility and Implications for Food Assistance Programs):1112-1138.

King, Miriam, Steven Ruggles, J. Trent Alexander, Sarah Flood, Katie Genadek,

Matthew B. Schroeder, Brandon Trampe, and Rebecca Vick. Integrated Public Use Microdata Series, Current Population Survey: Version 3.0. [Machine-readable database]. Minneapolis: University of Minnesota, 2010.

Klerman, Jacob Alex and Caroline Danielson. 2011. “The Transformation of the

Supplemental Nutrition Assistance Program” Journal of Policy Analysis and Management, 30(4): 863-888).

McKernan, Signe-Mary and Caroline Ratcliffe. 2003. “Employment Factors

Influencing Food Stamp Program Participation,” E-FAN-03-012, Washington, DC: Economic Research Service, U.S. Department of Agriculture.

Meyer, Bruce D., Wallace K. C. Mok, and James X. Sullivan. 2009. "The Under-

Reporting of Transfers in Household Surveys: Its Nature and Consequences." National Bureau of Economic Research Working Paper Series No. 15181.

Mikelson, Kelly S. and Robert I. Lerman. 2004. “The Relationship between the

EITC and Food Stamp Program Participation Among Households with Children.” Urban Institute. http://www.urban.org/UploadedPDF/411033_EITC.pdf

Ratcliffe, Caroline and Signe-Mary McKernan and Kenneth Finegold. 2008.

“Effect of Food Stamp and TANF Policies on Food Stamp Program Participation.” Social Service Review, 82(2), June: 291-334.

30  

Ratcliffe, Caroline, Signe-Mary McKernan, and Sisi Zhang. 2011. "How Much Does the Supplemental Nutrition Assistance Program Reduce Food Insecurity?" American Journal of Agricultural Economics 93 (4):1082-1098.

Ribar, David C, Marilyn Edelhoch, and Qiduan Liu. 2008. “Watching the Clocks:

The Role of Food Stamp Recertification and TANF Time Limits in Caseload Dynamics” Journal of Human Resources, 43(1): 208-39.

Ribar, David C., Marilyn Edelhoch, and Qiduan Liu. 2010. “Food Stamp

Participation among Adult-Only Households. Southern Economic Journal. 77(2): 244-270.

Richburg-Hayes and Lashawn and Isaac Kwakye. 2005. Food Stamp Caseload

Dynamics: A Study in Four Big Cities, MDRC Technical Report. Rosenbaum, Dorothy. 2000. “Improving Access to Food Stamps: New Reporting

Options Can Reduce Administrative Burdens and Error Rates.” Washington, DC: Center on Budget and Policy Priorities, September.

Rowe, Audrey. 2012. Budget Hearing before the House Committee on

Appropriations Subcommittee on Agriculture, Rural Development, Food and Drug Administration and Related Agencies, February 28.

Schwabish, Jonathan. 2012. “Downloading Benefits: The Impact of Online Food

Stamp Applications on Participation.” APPAM Conference Paper. Available at: https://appam.confex.com/appam/2012/webprogram/ExtendedAbstract/Paper2890/Schwabish%20Online%20SNAP%20Paper_10_12_2012_APPAM.pdf

Tanner, Michael. 2013. "It's Time to Rein In the Growth of Food Stamps."

Commentary, CQ Roll Call, November 13, 2013. http://www.rollcall.com/news/its_time_to_rein_in_the_growth_of_food_stamps_commentary-229047-1.html

Heritage Foundation, “Reforming the Swelling Food Stamps Program,” Factsheet No. 129, October 28, 2013: http://www.heritage.org/research/factsheets/2013/10/reforming-the-swelling-food-stamps-program

31  

Tschoepe Gary J. John J. Hindera. 1998. Explaining State Food Stamp Caseload Variance: Does State Administration Matter? Administration & Society, 30(1): 53-61.

U.S. Department of Agriculture, Food and Nutrition Service. 2006. “An Analysis of Food Stamp Benefit Redemption Patterns,” http://www.fns.usda.gov/analysis-food-stamp-benefit-redemption-patterns.

U.S. General Accounting Office. 2002. “Food Stamp Program: States' Use of Options and Waivers to Improve Program Administration and Promote Access,” GAO-02-409, Washington, DC.

Wallace, Geoffrey, and Rebecca M. Blank. 1999. "What Goes up Must Come Down: Explaining Recent Trends in Public Assistance Caseloads." In Economic Conditions and Welfare Reform, edited by Sheldon Danziger. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research.

Ziliak, James P, Craig Gundersen, and David N. Figlio. 2003. “Food Stamp

Caseloads over the Business Cycle” Southern Economic Journal, 69(4): 903-19.

Ziliak, James P. 2013. “Who Uses SNAP, and When? The Role of Economy,

Policy, and Demographics” University of Kentucky Center for Poverty Research Working Paper. (Forthcoming, Five Decades of Food Stamps, 2015)

Ziliak, James P., David N. Figlio; Elizabeth E. Davis; Laura S. Connolly. 2000.

“Accounting for the Decline in AFDC Caseloads: Welfare Reform or the Economy?” The Journal of Human Resources 35(3):570-586.

32  

Figure 1. SNAP Participants, Unemployment Rate, and Poverty Rate, 1980 – 2013

Source: USDA, Food and Nutrition Service (SNAP participants), U.S. Department of Commerce, U.S. Census Bureau (poverty rate), and U.S. Department of Labor Bureau of Labor Statistics (unemployment rate).

0

2

4

6

8

10

12

14

16

18

20

0

5

10

15

20

25

30

35

40

45

50

1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013

Une

mpl

oym

ent R

ate,

Pov

erty

Rat

e

Mil

lion

s of

peo

ple

SNAP participants Poverty rate Unemployment rate

33  

Figure 2: Monthly SNAP Caseloads, Selected States, January 1990-December 2011

2

4

6

8

10

12

14

16

0.00

0.50

1.00

1.50

2.00

Janu

ary-

90Ja

nuar

y-91

Janu

ary-

92Ja

nuar

y-93

Janu

ary-

94Ja

nuar

y-95

Janu

ary-

96Ja

nuar

y-97

Janu

ary-

98Ja

nuar

y-99

Janu

ary-

00Ja

nuar

y-01

Janu

ary-

02Ja

nuar

y-03

Janu

ary-

04Ja

nuar

y-05

Janu

ary-

06Ja

nuar

y-07

Janu

ary-

08Ja

nuar

y-09

Janu

ary-

10Ja

nuar

y-11

Une

mpl

oym

ent R

ate

SN

AP

Cas

eloa

d, in

Mil

lion

s Michigan

I di id l

2

3

4

5

6

7

8

9

10

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

Janu

ary-

90Ja

nuar

y-91

Janu

ary-

92Ja

nuar

y-93

Janu

ary-

94Ja

nuar

y-95

Janu

ary-

96Ja

nuar

y-97

Janu

ary-

98Ja

nuar

y-99

Janu

ary-

00Ja

nuar

y-01

Janu

ary-

02Ja

nuar

y-03

Janu

ary-

04Ja

nuar

y-05

Janu

ary-

06Ja

nuar

y-07

Janu

ary-

08Ja

nuar

y-09

Janu

ary-

10Ja

nuar

y-11

Une

mpl

oym

ent R

ate

SN

AP

Cas

eloa

d, in

Mil

lion

s New York

34  

Monthly Unemployment Rate Individual SNAP Caseload Household SNAP Caseload

2

4

6

8

10

12

14

16

0.00

0.50

1.00

1.50

2.00

Jan-

90Ja

n-91

Jan-

92Ja

n-93

Jan-

94Ja

n-95

Jan-

96Ja

n-97

Jan-

98Ja

n-99

Jan-

00Ja

n-01

Jan-

02Ja

n-03

Jan-

04Ja

n-05

Jan-

06Ja

n-07

Jan-

08Ja

n-09

Jan-

10Ja

n-11

Une

mpl

oym

ent R

ate

SN

AP

Cas

eloa

d, in

Mil

lion

s Pennsylvania

2

4

6

8

10

12

14

16

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

4.50

Jan-

90Ja

n-91

Jan-

92Ja

n-93

Jan-

94Ja

n-95

Jan-

96Ja

n-97

Jan-

98Ja

n-99

Jan-

00Ja

n-01

Jan-

02Ja

n-03

Jan-

04Ja

n-05

Jan-

06Ja

n-07

Jan-

08Ja

n-09

Jan-

10Ja

n-11

Une

mpl

oym

ent R

ate

SN

AP

Cas

eloa

d, in

Mil

lion

s Texas

35  

Table 1: Summary Statistics of Covariates, Selected Years

Policies Affecting Eligibility 1990 1993 1996 1999 2002 2005 2008 2011

Number of States Exempting at Least 1 but not all Vehicles from SNAP Asset Test 0 0 3 3 14 13 13 5

Number of States Exempting All Vehicles from SNAP Asset Test 0 0 0 0 21 29 35 46

Number of States with Transitional SNAP Benefits for TANF leavers 0 0 0 0 4 16 20 21

Number of States with Broad-Based Categorical Eligibility 0 0 0 0 9 11 18 41

Number of States with Eligibility Restrictions for Nonelderly Noncitizens 0 0 0 43 43 45 45 48

Policies Affecting Transaction Costs

Number of States with at least 25% of Working Households with Short Certification Periods (1-3 Months)

5 3 14 23 16 2 1 0

Mean State Proportion of Working Households with Short Recertification Periods (1-3 Months)

0.105 0.062 0.175 0.312 0.186 0.041 0.022 0.007

Number of States with Simplified Reporting 0 0 0 0 22 44 48 50

Number of States with Online Application Availability 0 0 0 0 2 8 18 33

Policies Affecting Stigma

Mean Proportion of State Benefits Issued via Electronic Benefits Transfer (EBT) 0.001 0.028 0.135 0.667 0.859 1.000 1.000 1.000

Number of States requiring Fingerprinting during Application 0 0 1 3 4 4 3 3

Policies Affecting Outreach

Number of States with a Federally Funded Radio or TV Ad 0 0 0 0 0 37 29 10

Other Low-Income Policies and Economic Conditions

Number of States with either AFDC Waiver or TANF Implemented 0 7 40 51 51 51 51 51

Mean Real State and Federal EITC, in Thousands (2010 Dollars) 1.553 2.145 4.449 5.104 5.123 5.078 5.097 5.256

Mean State Unemployment Rate 0.054 0.063 0.051 0.041 0.054 0.049 0.053 0.081

Caseload Variables

Household Caseload per Total Households 0.081 0.107 0.099 0.074 0.080 0.102 0.112 0.178

Individual Caseload per Capita 0.080 0.100 0.090 0.066 0.072 0.091 0.098 0.146

Sources: Bureau of Labor Statistics, USDA Food and Nutrition Service. US Health and Human Services. http://aspe.hhs.gov/hsp/Waiver-Policies99/Table_A.PDF http://www.acf.hhs.gov/programs/ofa/annualreport6/chapter12/chap12.htm#4, and University of Kentucky Center for Poverty Research.

36  

Table 2. Estimates of the Long-Run Determinant of State Policy Options on Households Per Capita Receiving Food Stamps Static Dynamic 12 lags 24 lags 12 lags 24 lags Long-run effect of: (1) (2) (3) (4) Policies Affecting Eligibility Exempt At Least One But Not All Vehicles from SNAP Asset Test

-0.015 0.022 -0.024 0.059 p=0.024 p<0.001 p=0.020 p=0.001

Exempts All Vehicles from SNAP Asset Test

-0.009 0.020 -0.012 0.067 p=0.182 p=0.021 p=0.557 p=0.031

Transitional SNAP Benefits to TANF Leavers

0.027 0.043 0.042 0.064 p=0.029 p=0.127 p=0.011 p=0.059

Broad-based Categorical Eligibility 0.059 0.054 0.095 0.059 p<0.001 p<0.001 p<0.001 p<0.001

Eligibility Restrictions for Non-elderly Noncitizens

-0.098 -0.160 -0.156 -0.240 p<0.001 p<0.001 p<0.001 p<0.001

Policies Affecting Transaction Costs Proportion of Working Households with Short (1-3 Month) Certification Period

-0.1254 -0.133 -0.164 -0.125 P<0.001 p<0.001 p=0.005 p=0.002

Simplified Reporting Policy 0.027 0.018 0.049 0.052 p=0.028 p<0.001 p=0.010 p<0.001

State Online Application Availability 0.012 0.023 0.019 0.026 p=0.330 p=0.042 p=0.528 p=0.233 Policies Affecting Stigma Proportion of State Benefits Issued via EBT

-0.028 -0.006 0.012 0.041 p=0.250 p=0.002 p=0.241 p<0.001

Fingerprinting Requirement -0.064 -0.084 -0.075 -0.039 p=0.007 p=0.007 p=0.026 p=0.030

Policies Affecting Outreach Federally Funded TV and Radio Ad Campaign

-0.005 -0.026 0.026 -0.024 p=0.021 p=0.011 p=0.055 p=0.081

Other Low-Income Policies and Economic Conditions Implementation of AFDC Waiver or TANF

-0.042 -0.048 -0.055 -0.061 p<0.001 p<0.001 p=0.003 p<0.001

Real Maximum State and Federal EITC Payout($1,000)

0.062 -0.260 0.021 -0.481 p<0.001 p<0.001 p=0.027 p=0.021

State Unemployment Rate 4.506 5.789 7.311 10.074 p<0.001 p<0.001 p<0.001 p<0.001

Lagged Dependent Variable - - -0.392 p<0.001

-0.350 p<0.001

Note: All variables are first differenced and estimates use Driscoll-Kraay standard errors. Reported estimates are the long-run effects of each covariate. The long-run effects for each explanatory variable are given by

L

j1

for static models and

L

L

j

1

1

1

for dynamic models. Note that θ is the coefficient estimate of the lagged caseload. Reported p-values represent the statistical significance of the F-test on all the lagged policy coefficients. See text for further details.

37  

Table 3. Estimates of the Long-Run Determinant of State Policy Options on Individuals per Capita Receiving Food Stamps Static Dynamic Long-run effect of: 12 lags 24 lags 12 lags 24 lags (1) (2) (3) (4) Policies Affecting Eligibility Exempt At Least One But Not All Vehicles from SNAP Asset Test

0.003 0.038 0.008 0.091 p=0.077 p=0.037 p=0.249 p=0.029

Exempts All Vehicles from SNAP Asset Test

0.011 0.032 0.018 0.084 p=0.227 p=0.032 p=0.380 p=0.002

Transitional SNAP Benefits to TANF Leavers

0.033 0.046 0.052 0.064 p<0.001 p=0.001 p=0.007 p=0.048

Broad-based Categorical Eligibility

0.062 0.064 0.122 0.105 p<0.001 p<0.001 p<0.001 p=0.001

Eligibility Restrictions for Non-elderly Noncitizens

-0.086 -0.148 -0.154 -0.248 p<0.001 p<0.001 p<0.001 p<0.001

Policies Affecting Transaction Costs Proportion of Working Households with Short (1-3 Month) Certification Period

-0.145 -0.152 -0.235 -0.179 p<0.001 p<0.001 p<0.001 p<0.001

Simplified Reporting Policy

0.027 0.021 0.058 0.068 p<0.001 p<0.001 p<0.001 p<0.001

State Online Application Availability

0.009 0.022 0.016 0.026 p=0.522 p=0.019 p=0.566 p=0.125

Policies Affecting Stigma Proportion of State Benefits Issued via EBT

-0.043 -0.012 -0.013 0.019 p=0.132 p=0.028 p=0.366 p=0.007

Fingerprinting Requirement

-0.068 -0.104 -0.105 -0.100 p<0.001 p=0.001 p=0.013 p=0.037

Policies Affecting Outreach Federally Funded TV and Radio Ad Campaign

-0.001 0.005 0.034 0.015 p=0.019 p=0.026 p=0.258 p=0.648

Other Low-Income Policies and Economic Conditions Implementation of AFDC Waiver or TANF -0.049 -0.054 -0.072 -0.081

p=0.002 p<0.001 p<0.001 p<0.001 Real Maximum State and Federal EITC Payout($1,000)

0.039 -0.311 -0.025 -0.658 p<0.001 p<0.001 p=0.045 p=0.059

State Unemployment Rate 4.531 5.751 8.556 11.396 p<0.001 p<0.001 p<0.001 p<0.001

Lagged Dependent Variable - - -0.664 p<0.001

-0.565 p<0.001

Note: All variables are first differenced and estimates use Driscoll-Kraay standard errors. Reported estimates are the long-run effects of each covariate. The long-run effects for each explanatory variable are given by

L

j1

for static models and

L

L

j

1

1

1

for dynamic models. Note that θ is the coefficient estimate of the lagged caseload. Reported p-values represent the statistical significance of the F-test on all the lagged policy coefficients. See text for further details.

38  

Table 4. Estimates of the Long-Run Effects of Restrictions on Eligibility for Nonelderly Noncitizens by Size of Potentially Eligible Population No Lagged

Dependent Variable With Lagged

Dependent Variable 12 lags 24 lags 12 lags 24 lags

Household Caseload

Ineligibility for Noncitizens-0.092

p<0.001 -0.149

p<0.001 -0.146

p<0.001 -0.222

p<0.001

Ineligibility for Noncitizens*Large Population not Native-born Citizen

-0.052 p<0.001

-0.075 p<0.001

-0.084 p<0.001

-0.132 p<0.001

Individual Caseload

Ineligibility for Noncitizens-0.078

p<0.001 -0.137

p<0.001 -0.142

p<0.001 -0.232

p<0.001

Ineligibility for Noncitizens*Large Population not Native-born Citizen

-0.064 p<0.001

-0.075 p<0.001

-0.105 p<0.001

-0.119 p<0.001

Note: States with a large population of citizens that are not native-born are determined by the states with the largest absolute size of the population that are estimated to be native-born citizens from the IPUMS-CPS: California, New York, Florida, Texas, New Jersey, and Illinois. See text for further details. All Estimates controlling for all other state policies from Tables 2 and 3. Reported p-values represent the statistical significance of the F-test on the coefficients.

39  

Table 5. Simulation Results of the Long-Run Determinants of State Policy Options on Household Receiving Food Stamps

Percent of Predicted Change Explained by:

Time Period

Actual % change in household caseload

Predicted Change with All

Observable Factors

All Observable

Factors

SNAP Policies

All

SNAP

Policies Affecting Eligibility

SNAP Policies

Affecting Transaction Costs and

Stigma Welfare Reform EITC

Economic and

Seasonal Factors

03/93-07/00 -50.0% -40.4% 100% 38.1% 36.5% 1.4% 10.0% 21.7% 60.0% (8.1 ) (0.0) (16.6 ) (13.6 ) (5.7 ) (6.9 ) (14.1 ) (17.8 ) 07/00-12/11 96.6% 87.0% 100% 21.2% 12.1% 9.2% 0.0% 1.2% 39.0% (8.9 ) (0.0 ) (5.1) (3.1) (3.2 ) - (1.0 ) (5.7) 12/07-12/11 56.6% 50.0% 100% 8.7% 6.4% 1.9% 0.0% -0.8% 61.8% (3.6 ) (0.0 ) (3.3) (2.1) (2.2) - (1.1 ) (6.7 )

Notes: Author’s calculations based on column 4 of Table 2. We generate standard errors (in parentheses) from a bootstrapping method described in the text.

40  

Table 6. Simulation Results of the Long-Run Determinants of State Policy Options on Household Receiving Food Stamps Selected States, 2000-11 Percent of Predicted Change Explained by: State

Actual % change in household

caseload

Predicted Change with

All Observable

FactorsSNAP

Policies All

SNAP Policies

Affecting Eligibility

SNAP Policies Affecting Transaction

Costs and Stigma

Economic and

Seasonal Factors

United States 96.6%

86.9 (8.6)

21.2% (5.1)

12.1% (3.1)

9.2% (3.2)

39.0% (5.7)

CA

90.0

74.4 (16.2)

27.1 (13.1)

27.4 (11.3)

0.0 (7.9)

60.2 (20.0)

FL

124.8

112.7 (49.7)

16.7 (12.9)

8.3 (9.0)

8.6 (5.8)

42.2 (45.9)

MI

119.6

116.8 (15.5)

15.7 (4.3)

5.4 (2.2)

10.2 (3.2)

47.5 (7.9)

NY

76.8

77.9 (17.7)

27.9 (14.5)

17.8 (8.3)

10.2 (8.8)

28.7 (8.4)

PA

80.2

73.6 (14.3)

26.4 (11.7)

19.0 (8.3)

7.7 (5.4)

34.3 (7.0)

TX

107.8

84.2 (44.0)

18.5 (33.1)

5.2 (10.2)

13.6 (28.5)

27.9 (57.0)

WV

54.6

48.3 (14.3)

53.3 (27.1)

19.3 (9.6)

34.4 (19.1)

21.8 (8.5)

OK

87.9

74.8 (18.5)

34.0 9.4)

19.0 (8.3)

7.7 (5.4)

34.3 (7.0)

Notes: Author’s calculations based on column 4 of Table 2. We generate standard errors (in parentheses) from a bootstrapping method described in the text

41  

Appendix Table 1: Variation in State EITC adoption and Timing of Federal EITC Payments

Portion of Federal EITC paid out in Each Month, by Year:

Year Number of States with

EITCs January February March April

May-December

1990 5 0.003 0.215 0.441 0.171 0.169 1991 6 0.006 0.254 0.428 0.155 0.157 1992 6 0.002 0.340 0.379 0.140 0.139 1993 6 0.023 0.433 0.324 0.117 0.103 1994 7 0.020 0.452 0.299 0.126 0.102 1995 7 0.013 0.205 0.292 0.217 0.274 1996 7 0.009 0.325 0.376 0.144 0.146 1997 9 0.010 0.379 0.351 0.134 0.125 1998 10 0.041 0.456 0.301 0.108 0.094 1999 11 0.037 0.468 0.295 0.110 0.090 2000 14 0.001 0.548 0.275 0.093 0.083 2001 15 0.036 0.539 0.264 0.093 0.069 2002 15 0.030 0.547 0.253 0.092 0.077 2003 15 0.043 0.549 0.246 0.087 0.075 2004 16 0.059 0.383 0.428 0.076 0.053 2005 19 0.083 0.597 0.192 0.078 0.051 2006 19 0.083 0.595 0.197 0.081 0.045 2007 21 0.092 0.584 0.194 0.080 0.051 2008 23 0.081 0.592 0.205 0.081 0.041 2009 23 0.121 0.597 0.175 0.069 0.037 2010 24 0.122 0.574 0.158 0.062 0.021 2011 25 0.114 0.559 0.183 0.083 0.061

Sources: Data on states with Earned Income Tax Credits (EITCs) from the University of Kentucky Center for Poverty Research. Data on timing of federal EITC payouts (“Payment where earned income credit exceeds liability for tax”) are from the Department of Treasury’s Monthly Treasury Statement for 2006 and 2007. Ryan Edwards generously provided his data for the 1990 through 2005 period.