Comparative Study of Probabilistic Models for Dietary Intake Data
Ayona Chatterjee
California State University East Bay
Santosh Gummidipundi
California State University East Bay
Henry Lankin
California State University East Bay
Christine Marachi
California State University East Bay
An important area of food safety risk assessment involves monitoring intake of pesticides through the diet.Dietary data obtained on consumption of certain food products may have a large number of zeros. Distribution of intakes for products have a peak at zero which need to be accounted for while modelling such data. Most dietary intake data such as the NHANES provide consumption values for only two days. Also consumption of certain foods may be correlated. In this paper we compare two models to account for the issues above using a Bayesian framework; a propensity model and a latent Gaussian model. The propensity-model is a two-stage model which first assigns each individual a certain probability of consumption for a product and then we model the non-zero consumption. We also develop a latent Gaus- sian model for the data with the additional assumption that consumptions between foods may be correlated. We compare predicted values from our Bayesian models with those observed. We also discuss extending our models for predicting long-term consumption patterns to study chronic risk.