We tested the properties of the 18 Household Food Security Survey (HFSS) items, and the validity of the resulting food security classifications, in an English-speaking middle-income country.
Survey of primary school children in Trinidad and Tobago. Parents completed the HFSS. Responses were analysed for the 10 adult-referenced items and the eight child-referenced items. Item response theory models were fitted. Item calibrations and subject scores from a one-parameter logistic (1PL) model were compared with those from either two-parameter logistic model (2PL) or a model for differential item functioning (DIF) by ethnicity.
There were 5219 eligible with 3858 (74%) completing at least one food security item. Adult item calibrations (standard error) in the 1PL model ranged from -4.082 (0.019) for the 'worried food would run out' item to 3.023 (0.042) for 'adults often do not eat for a whole day'. Child item calibrations ranged from -3.715 (0.025) for 'relied on a few kinds of low cost food' to 3.088 (0.039) for 'child didn't eat for a whole day'. Fitting either a 2PL model, which allowed discrimination parameters to vary between items, or a differential item functioning model, which allowed item calibrations to vary between ethnic groups, had little influence on interpretation. The classification based on the adult-referenced items showed that there were 19% of respondents who were food insecure without hunger, 10% food insecure with moderate hunger and 6% food insecure with severe hunger. The classification based on the child-referenced items showed that there were 23% of children who were food insecure without hunger and 9% food insecure with hunger. In both children and adults food insecurity showed a strong, graded association with lower monthly household income (P < 0.001).
These results support the use of 18 HFSS items to classify food security status of adults or children in an English-speaking country where food insecurity and hunger are more frequent overall than in the US.