IRT to measure food insecurity through people experiences Seminar

We introduce methods to compile a new indicator of food insecurity based on the Food Insecurity Experience-based Scale (FIES), an 8-item questionnaire that elicits information on individuals' experiences and behaviors resulting from inadequate access to food. Compared to the other indicators of food insecurity, the FIES provides a measure with stronger theoretical and statistical underpinnings, and supports measurement of food insecurity at different degrees of severity. Classification of individuals against meaningful thresholds of severity allows national and sub-national estimates of prevalence rates of moderate and severe food insecurity. The questionnaire is currently being administered in more than 150 countries worldwide through the Gallup World Poll, translated in all major languages in each country. The samples represent the adult population (15+ years) in each country.

A single-parameter logistic model is estimated through conditional maximum likelihood, extended to allow the inclusion of sampling weights. The performance of the measurement scale is assessed through an analysis of (mis)fit statistics, computable at item and individual level. Fit statistic errors, conditional correlation and sampling independence are also defined and checked. An innovative technique is used to treat extreme raw score parameters and errors depending on the proportion of respondents falling in these classes. Raw scores and corresponding measurement errors are used to draw food insecurity severity profiles, along which thresholds are set to classify respondents in moderate or severe food insecurity classes.

Although effort has been made to linguistically and culturally adapt the questions to harmonize their meaning when used in different countries and contexts, discrepancies in the way respondents, in different countries, interpret the questionnaire can still persist; an equalization technique is introduced to ensure that prevalence rates are comparable across countries. The technique entails an iterative process that leads to the identification of "common" and "unique" items across countries, and to the construction of a global reference scale. Such reference scale is obtained by considering only the items identified as "common" in each country.

We illustrate the provisional results on 33 countries and critically discuss some methodological developments.