Statistical Model Will Help Estimate Effects of Receiving Food Assistance
January 26, 2010
For the U.S. Food and Nutrition Service Abt Associates developed and tested a model that could potentially be used to design studies to estimate the effects of receiving food assistance under the Supplemental Nutrition Assistance Program (SNAP), formerly known as the Food Stamp program. In such studies, it will be necessary to take account of underlying differences between people who receive SNAP benefits and those who do not receive benefits but are eligible to do so. Otherwise, effects due to such differences might be falsely attributed to SNAP participation. This model will help make the necessary statistical adjustments.
“Understanding SNAP’s effects is more important today than ever, because of a dramatic increase in the demand for SNAP due to the downturn in the U.S. economy combined with rising food prices.”
The “gold standard” of evaluation research is the experimental approach in which the treatment group receives the intervention — in this case, SNAP food assistance — and the control group does not. In the case of SNAP, it would be both unethical and infeasible to deny food assistance to people who were eligible for it; therefore it is necessary to explore ways to make statistical adjustments.
Using Propensity Score Matching to Make Statistical Adjustments
Results from this project suggest that the necessary statistical adjustments could in fact be developed and applied through a method called propensity score matching. This method involves examining in detail the characteristics and circumstances of households who are likely to receive SNAP benefits, a task that the current model has accomplished.
The model that Abt Associates developed was able to predict with over 75 percent accuracy whether a SNAP-eligible household would receive food assistance from SNAP. Because the researchers included a large number of previously unexplored variables in the model — which they identified during an extensive literature review — they believe the other 25 percent consists of random differences between the SNAP participants and eligible non-participants, rather than resulting from systematic differences. This is important, because if the differences were systematic it would mean that the estimated effects of SNAP based on this model will be incorrect.
The researchers recommend that the U.S. Food and Nutrition Service proceed to test the propensity score approach for suitability in a large-scale national study. They state that “It will be essential to allay the concerns of the research community regarding the feasibility and validity of a non-experimental approach to estimating SNAP impacts before engaging in such a major undertaking. Such a test might be done using extant data, if any can be found with suitable measures of both participation factors and nutritional outcomes. Alternatively, FNS could collect new data modeled on parts of the Survey of Income and Program Participation supplemented with outcomes data.”
SNAP Expenditures Rise by More Than 60 Percent in One Year
As the agency responsible for administering SNAP, the Food and Nutrition Service (FNS) has always had a large stake in understanding the program’s impacts on participants’ food expenditures, household food supplies, individual dietary intake, and food security.
Understanding SNAP’s effects is more important today than ever, because of a dramatic increase in the demand for SNAP due to the downturn in the U.S. economy combined with rising food prices. In June 2009 the program served 15.9 million households at a cost of $4.7 billion. This is an increase of 23 percent in caseload and 61 percent in expenditures from a mere 12 months before.
