The Role of Food Assistance in Helping Food Insecure Households Escape Hunger^†

Every year, the federal government spends a substantial sum on food assistance—totaling $41.8 billion in 2003 (Oliveira). As a result, there is a great deal of interest in assessing the role that food assistance programs play in helping low-income families maintain nutritional adequacy. One outcome of interest is household food security, a measure developed by the U.S. Department of Agriculture (USDA) in the 1990s which indicates whether all members of a household have access, at all times, to enough food for an active and healthy life. The relationship between food assistance and food security has been difficult to estimate because the program participation decision is endogenous. Food insecure households are more likely to self-select into food assistance programs, resulting in a positive association that must be controlled in order to assess the effect of program participation on food insecurity and hunger. Furthermore, the positive association resulting from self-selection has been found to dominate the (presumably) negative program effect (Gundersen and Oliveira).

In this paper, we study one aspect of the relationship between food security and food assistance. For households that experienced some degree of hunger during the course of a year, we examine whether food assistance use was associated with lower odds of food insecurity during the last thirty days of that year. By focusing on changes in food security status, we partially circumvent the self-selection problem. We limit our analysis to the Food Stamp Program (FSP) and the National School Lunch Program (NSLP).

Previous Research

There is a growing body of research that attempts to study the relationship between food assistance and food security, relying on statistical methods to control for selection bias. Some studies use a two-stage procedure in which predicted program participation replaces actual participation in the food security equation (Gundersen and Oliveira; Huffman and Jensen; Kabbani and Yazbeck). While these studies have succeeded in removing the positive association between food insecurity and food assistance, they have had limited success in identifying a significant program effect.

Borjas used changes in program eligibility rules across states to estimate the effect of program participation on food insecurity among immigrant households. His analysis provides strong evidence that public assistance use reduces food insecurity. Nord and Romig used seasonal variation in food insecurity between spring and summer months to examine the effects of the NSLP and the Summer Food Service Program. They found evidence that both programs reduce hunger among households with school-age children.

Others have studied the association between changes in program participation and changes in food security. Ribar and Hamrick found that FSP participation was associated with lower rates of exit from food insufficiency (evidence of self-selection), but was not significantly associated with concurrent levels of food insufficiency. Hofferth found that leaving the FSP was associated with higher odds of remaining food insecure or becoming food insecure (evidence of a program effect). However, she found no significant association between entering the FSP and remaining or becoming food insecure. Thus, studying changes over time only partly overcame the selection problem.

Empirical Methodology

This paper studies the relationship between food assistance use and changes in household food security, using a previously unexploited longitudinal dimension of food security. The Food Security Supplements of the Current Population Survey (FSS-CPS), which are used by USDA to estimate annual prevalence rates of hunger and food insecurity in the United States (Nord, Andrews, and Carlson), include questions that allow the construction of USDA's standard twelve months food security scale as well as a thirty days food security scale. Using the twelve months scale, we identify households that experienced hunger during the course of a year. For these households, we identify factors associated with avoiding food insecurity during the last thirty days of that year. In other words, for households that experienced hunger during the year, we estimate the following equation:

(1)

where S₃₀ = 0 indicates that a household was food secure during the thirty days before the interview and S₃₀ = 1 indicates that the household was food insecure during the thirty days before the interview. The food assistance program variable (FAP) refers alternatively to participation in the FSP and the NSLP during the thirty days before the interview. We limit the analysis, as much as possible, to the eligible population for each program. This means limiting the sample to households with incomes under the income cut-offs for each program. For the NSLP, this also means limiting the analysis to households with school-age children. We also consider a “dose-response” model, suggested by Hamilton and Rossi, in which we use average household FSP benefits per person to estimate the program effect, restricting the sample to FSP participants.

The thirty days scale is constructed using questions that represent more severe levels of food insecurity (Nord). By focusing on whether households that experienced hunger during the year avoid food insecurity (rather than hunger) during the last thirty days of the year, we take a conservative approach to defining what it means to “escape” hunger. This allows us to focus on households that really were able to achieve a significant improvement in their food security status during the year.

There are several limitations to using cross-sectional data to identify changes in food security status over time. First, households are more likely to recall recent incidences of food insecurity compared to events earlier in the year. We do not anticipate that this issue will significantly bias our results. Second, the FSS-CPS only allows us to identify whether a household participated in the FSP or the NSLP during the previous month. Some households may have participated for the entire year. This limitation can be partly addressed in future work by merging the April FSS-CPS with data from the March Demographic Survey, which contains program participation information for the previous year. Third, we are only able to identify whether households persist in or transition out of food insecurity. For some households, a single incidence of hunger may have occurred in the last month of the year, concurrent with the thirty days measure. These households would be dropped from our analysis if we had longitudinal data. We expect this limitation to bias the coefficients on program participation downward, suggesting weaker program associations than may actually be the case.

Fourth, focusing on changes in food security status and controlling for observable differences across households only partially addresses the problem of selection. In our study, we do not control for unobservable differences that may be jointly correlated with program participation and food security. Finally, it is important to emphasize that the coefficients in equation (1) are not estimates of a typical program effect. They estimate the association between program participation and escaping hunger for a very specific and policy-relevant population—households that experienced hunger during the year.

Data

This paper uses data from the April 1995, April 1997, April 1999, and April 2001 Food Security Supplements of the Current Population Survey (FSS-CPS). The CPS is a monthly representative survey of around 50,000 U.S. households conducted by the Census Bureau. The FSS-CPS was also administered in September 1996, August 1998, September 2000, December 2001, and thereafter in December of every year. Our study only uses data from the April supplements because food security prevalence estimates exhibit significant seasonal variation between the spring and fall/summer supplements. We also exclude survey rotation 8 in 1995 and rotations with experimental questions: rotations 4 and 8 in 1997 and rotation 8 in 1999 and 2001.

The FSS-CPS include adult- and child-referenced questions that are combined to create annual and thirty days scales, which are used to classify households into three levels of food security: food secure, food insecure without hunger, and food insecure with hunger. Recent studies suggest that child hunger covers a separate dimension not adequately captured by standard food security measures (Nord and Bickel; Wilde). We therefore follow Nord and Romig by developing modified annual and thirty days measures based only on the adult-referenced items common to households with and without children.

The FSS-CPS contain information about household participation in several food assistance programs, including the FSP and NSLP, as well as information on food stamp benefit amounts, which we use in one empirical specification. We limit our analysis to the FSP and NSLP because it is comparatively simple to identify eligible nonparticipants for these two programs in the CPS. Briefly, the FSP provides low-income families with coupons or electronic benefit cards that allow eligible recipients to purchase food at local stores. To be eligible, most households must have gross cash income from all sources under 130% of federal poverty guidelines and assets below $2,000. The NSLP provides daily free and reduced-price meals to children in school. Households must have incomes less than 185% of poverty guidelines to qualify for reduced-price meals and incomes less than 130% of poverty guidelines to qualify for free meals.

One limitation of using FSS-CPS is that income is only available in broad ranges. In our analysis, we use the mid-points of these ranges as estimates of household income. Since we use different income-to-poverty-threshold cutoffs (130% for the FSP and 185% for the NSLP), we estimate separate regression models for each program. The CPS does not distinguish between NSLP participants that receive free and reduced-price meals. We address this issue by separating households into those that are income-eligible for free meals and those that are only income-eligible for reduced-price meals.

Our analysis controls for a number of demographic characteristics of household heads (as defined in Kabbani and Yazbeck), including age, race, gender, level of educational attainment, employment status, immigration status, whether they were disabled, and whether they were over sixty-five years old. Other household-level variables in our analysis include household size, composition, geographic location, homeownership, and income. The regression analysis is conducted without using household weights. However, using weights does not alter the main results. Finally, we do not correct for the complex sample design of the CPS, but we acknowledge that our standard errors will be biased downward by focusing on results that are significant at the 5% level.

Results

By restricting our analysis to households with incomes less than 130% of the poverty threshold for the FSP and less than 185% of the poverty threshold for the NSLP, and by focusing only on households that experienced hunger during the year, our sample size reduces to a maximum of 2,505 observations. We find that, in the thirty days prior to the survey, 81% of households with incomes less than 130% of the poverty threshold (that experienced hunger during the year) were food insecure. Around 42% of these households received food stamps, with an average benefit amount of $57 per person. We also find that 75% of households with school-age children (with incomes less than 185% of the poverty threshold that experienced hunger during the year) received free or reduced-price meals. This percentage increases to 80% for households with incomes less than 130% of the poverty threshold.

The first column of table 1 presents logistic regression results for the empirical model with no program participation variables included. The results are presented as odds ratios. An odds ratio of less than one indicates that the variable is associated with lower odds of being food insecure during the previous thirty days. Only a few of the control variables are significantly associated with food insecurity in our model. Households with an employed head are less likely to be food insecure (more likely to escape hunger) compared to households with an unemployed head. The number of children less than five years old and between the ages of five and eighteen are also positively associated with lower odds of food insecurity. One possible explanation for this finding is that households with children are more likely to be eligible for food assistance. Households with children are able to receive food stamp benefits indefinitely, as long as they meet program eligibility requirements, and the NSLP is only available to school-age children.

The second column suggests that participating in the FSP is associated with lower odds of being food insecure, but this finding is not statistically significant. The third column suggests that food stamp benefit amounts are associated with lower odds of food insecurity in the last thirty days for households that experienced hunger during the year. A $10 per person increase in the food stamp benefit amount is associated with a 12% reduction in the odds of a household being food insecure during the last thirty days.

Column 4 finds that, for households with school-age children that experienced hunger during the course of a year, those that participated in the NSLP were 33% less likely to be food insecure during the last thirty days of that year compared to households that did not participate. To further distinguish between households eligible for free meals and those eligible for reduced-price meals, the last column of table 1 restricts our analysis to households with incomes less than 130% of the poverty threshold. The odds ratio on participation in the NSLP suggests an even stronger program association. By comparison, limiting the analysis to households with incomes between 130% and 185% of the poverty threshold produces a statistically insignificant odds ratio of only 0.81 (results not shown). Accordingly, most of the program association observed for the NSLP is taking place among households eligible for free lunches.

In conclusion, this paper finds evidence that both the FSP and the NSLP play important roles in helping households escape hunger. While the results for participation in the FSP were not statistically significant, a dose-response analysis suggests that higher FSP benefit amounts are strongly associated with lower odds of food insecurity for households that experienced hunger during the year. The results for the NSLP suggest that participation in the program is associated with lower odds of food insecurity for households with school-age children that experienced hunger during the year. This association appears to be related mostly to households that were eligible for free meals.