Association for Public Policy Analysis & Management

Food insecurity is defined as the social and economic condition of having limited access to enough food to lead a healthy, active life. In 2014, 17.4 million (14.0 percent) households in the United States were food insecure at some point during the year (Coleman-Jensen et al., 2015). These numbers have remained at relatively high levels since the onset of the Great Recession in 2008, where food insecurity rates increased from 11.1 to 14.6 percent between 2007 and 2008 (Coleman-Jensen et al, 2015). Given the high prevalence of food insecurity and the negative health outcomes associated with it (Gundersen, Kreider, and Pepper, 2011), food insecurity is considered one of the major nutrition-related public health issues in the United States. Consequently, household food insecurity is receiving more interest from the research community.

            Many researchers use food insecurity measures from pre-existing survey data without critically evaluating how these measures are generated. Since 1995, the Current Population Survey (CPS) has included the Household Food Security Survey Module (HFSSM), a series of 18 items about the conditions and behaviors characterizing households having difficulty meeting their basic food needs. Responses to these items are combined using a Rasch measurement model (Rasch, 1960) to calibrate the U.S. food security scale, and to obtain estimates of the severity of household food insecurity. Households are then classified into ordinal food security status categories (i.e., food secure, low food security, very low food security, etc.) based on their scale score (Bickel et al., 2000).

            When examining food insecurity, researchers typically rely on analyzing ordinal or even binary indicators of food insecurity. The choice of a binary measure of food insecurity may be useful for communicating findings in some situations, but continuous measures of severity of food insecurity offer more accuracy and precision in many research situations (Rabbitt, 2013). For example, two households may differ in only one score point on the continuous scale, but be classified into two different food security status categories.  The use of continuous measures of food insecurity are preferable under these circumstances. This paper proposes and implements an alternative approach for estimating a continuous indicator of food insecurity based on a Joint Maximum Likelihood (JML) Rasch model using data from the 2001-2015 Current Population Survey Food Security Supplement (CPS-FSS). Estimating severity of food insecurity estimates using this methodology results in greater prevision in estimates of food insecurity, and also provides an opportunity to examine the quality of these estimates using model-data fit statistics. An illustrative example of the use of these estimates in empirical research is undertaken focusing on the Supplemental Nutrition Assistance Program (SNAP, formerly the Food Stamp Program) and food insecurity.