Validation of the food access survey tool to assess household food insecurity in rural Bangladesh

Participating women (n = 11,992) were asked to recall the frequencies of nine food insecurity experiences over the past 6 months on a 5-point Likert scale. The Rasch partial credit model was used to study the item category severity and differential item functioning (DIF) by literacy status, respondents’ age, land ownership and household sizes. Principal component analysis (PCA), non-parametric methods, and cumulative ordinal logistic regression models were applied to examine the Rasch model assumptions, namely unidimensionality, monotonicity and measurement invariance (non-DIF).

All items demonstrated good model fit with acceptable values of fit statistics (infit). PCA as well as other indices (Cronbach’s alpha = 0.85, scalability coefficient = 0.48) indicated that all items fit in a single statistical dimension. The ordered responses of nine items displayed monotonic increasing item category severity as expected theoretically. All nine items were flagged with statistically significant DIF between key demographic—and socioeconomic subgroups (p < 0.001); however, none of the detected DIF was considered practically significant given small effect sizes (variance explained by group membership and interaction term < 1 %). The total summed score over the polytomous FAST was inversely associated with household wealth, dietary diversity score and maternal body mass index, demonstrating external construct validity.