Authors: R. Jerome Anderson
This research utilizes the Rasch model to generate linear measures of nutrition from data on foods fed to children, thus providing a gauge with which to measure levels of child nutrition across a population. Linear measures are standardized, interval units that allow comparisons of latent constructs over time and among differing socio-economic variables. In this research, the latent construct is child nutrition. The data were drawn from the 2007 and 2013 Liberian Demographic and Health Surveys (DHS). The DHS provides data on foods fed to children, grouped by differing nutritional values. The Rasch software constructs interval measures from the yes/no responses given to the questions regarding foods fed to children. The measures then rank the foods fed to children from most frequently fed to least frequently fed. An examination of the resulting patterns indicates the nutritional levels of the children’s diets. In addition to these patterns, the overall measure of each child’s nutrition is also inferred from the responses to the questions as to the foods given to each child. Those measures may then be aggregated by the various socio-economic variables in the DHS database, such as geographic location, family’s wealth status, or mother’s educational level. The resulting analysis shows patterns that health care workers and policy-makers may use to target needed interventions in specific locations or for specific sub-groups of the population. The Liberian data used for this study show a decline in nutrition levels between 2007 and 2013 across all Liberian counties, as well as across educational and wealth categories. Using a subset of the data controlled for appropriate statistical power, a statistically significant decline in the level of nutrition was shown between the two years, in spite of impressive (25%) per capita GDP growth during this period. The results of the analysis demonstrate the feasibility of using a linear measurement model to establish standardized units by which child nutrition levels across DHS surveys may be legitimately compared.