The food insecurity scales were defined in the late 1990s. Unlike NAEP, the questions on the HFSSM remain the same year after year. Nevertheless, Ohls et al. (1999) discuss some indications that the scales were somewhat different for some of the years investigated. The source of this variability remains uncertain and could be due to several sources, for example, the poor fit of the Rasch model, technical aspects of the data collection, and different interpretations of the HFSSM questions in different years. The possibility that scale drift might occur should be examined on an ongoing manner to the extent possible. Methods of detecting differential item functioning may be used to investigate it with the data from different years being treated as the multiple groups.
The current approach to IRT modeling used by USDA is to create dichotomous/binary questions out of the several types of questions on the HFSSM, and then to use the Rasch model, which is designed for dichotomous questions. This approach has the potential of not using all of the information that is available in the battery of food insecurity questions. In addition, because two pairs of the questions are each two parts of a single question, the assumption of conditional independence in equation (5) is violated with unclear consequences. This section briefly outlines how this practice could be modified using ordered polytomous items.
In the current HFSSM, three types of questions are asked of either all households or households with children. First, there are dichotomous/binary questions with a yes/no response set. Second, there are questions with a trichotomous ordered response set (never, sometimes, often). Third, there are two-part questions that include a frequency follow-up to an initial question. The initial questions have a yes/no response set, and the frequency follow-up, if the initial answer is yes, has the trichotomous response set of (1–2 months, some months but not every month, or almost every month). For all households there is one dichotomous, three trichotomous, and four two-part questions. For households with children there is, in addition, one dichotomous, three trichotomous, and three two-part questions. Hence, for all households there are 8 different measures of food insecurity and for households with children there are an additional 7 measures, for a total of 15. These enumerations include only the questions that ask about “in the