rity at some time during the past year (or other period), and therefore is considered food insecure, may in fact be food secure at the time of the interview.” In such a case, the respondent’s beliefs about the stability or level of change in periods of food insecurity or hunger may supplement the respondent’s memory as the answer is constructed (see, for example, Ross and Conway, 1986). Other heuristics that supplement or replace memory when answers are constructed are described in Schwarz (1994) and Tourangeau, Rips, and Rasinski (2000).
The items currently included in the HFSSM vary in how the reference period is specified and where in the question the reference is located. In addition, the reference period confusion is exacerbated by the manner in which the questions are actually administered.
Items in the HFSSM use several formats to record the frequency of the various psychological states or behaviors. Six of the questions in the HFSSM require respondents to rate how often (i.e., often, sometimes, or never) the behavior or psychological state in question was true for them (or in some sort of aggregation across adults in the household). Three follow-up questions ask respondents to assess whether the behavior occurred “almost every month, some months but not every month, or in only 1 or 2 months.” Several of the questions in the FSS use the response options “often true, sometimes true, never true.” For the remaining items, respondents answer using a dichotomous yes/no format. When responses to scale items are converted into numerical scores, both “often” and “sometimes” responses are collapsed. The problems in deciding how often a description of an event is “true” are exacerbated when it is a complex event (so that part of the event described may have happened and part not) and a compound event (which is aggregated over multiple actors, who may have experienced the event with different frequencies). The use of vague quantifiers (often, sometimes) further confounds the interpretation of the response options.
One of the critical concerns in designing questions is ensuring that respondents interpret terms consistently. Cognitive testing on low-income respondents in upstate New York who were mostly white and black found that most respondents understood the terms of “hungry” or “not eating enough” as intended (“hunger” as a severe problem of decreased food quantity and “not eating enough” as less severe), however, some respondents also associated reduced quality with “not eating enough” (Alaimo, Olson, and Frongillo, 1999). As noted above, respondents may not understand the