placed on only one of them, the standard group, and the other groups have means and variances that are relative to those of the standard group.

USE OF ESTIMATED IRT MODELS FOR MEASURING FOOD INSECURITY

This section examines how USDA currently uses an IRT model to measure food insecurity and its prevalence in the United States, describing its classification system for households without children, and then interpreting this in terms of the material developed earlier. For households with children similar comments apply, so this case is not considered further here.

The responses to the 10 “adult” food insecurity questions are dichotomized and then the number of affirming responses to these 10 questions is used to classify the households into food insecurity levels. Households affirming two or fewer questions are classified as food secure; those with three to five affirmations are classified as food insecure without hunger; and those with six or more affirmations are classified as food insecure with hunger.

Basing the levels of food insecurity on the number of affirming responses to the dichotomized questions greatly simplifies the task of classifying households. For example, using the 10 dichotomized questions for households without children, the number of possible affirmations range from 0 to 10. On the one hand, using these same 10 questions, and, accounting for the fact that two pairs of them allow only three possible responses for each pair rather than four, there are 2632 = 576 different possible patterns of affirming and nonaffirming responses. If the questions were not dichotomized or if missing data were also taken into account, then the number of possible response patterns across the items could be much greater.

On the other hand, this simplification has also led to criticism of the food insecurity categories. Bavier (2004) noted that households that did not affirm the one question that specifically asks “… were you ever hungry …” would be, on the basis of affirming six or more other questions, classified by USDA as food insecure with hunger. Furthermore, households that did affirm the “hunger” questions but affirmed only four additional questions would not be classified as food insecure with hunger. It can also be argued that the location of the “were you ever hungry” question on the Rasch scale is a more reasonable cut point for the “food insecure with hunger” category. To the right of that point, the Rasch model predicts that the respondents are more likely than not to affirm the hunger question, but to the left they are less likely than not to affirm it. The “were you ever hungry” question is the seventh least likely to be affirmed of the 10 household adult questions. This suggests that the criterion for the category should be seven or more affirmations rather than only six. Of course, the criticism of Bavier men-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement