Hierarchical multiple informants models: A case study to determine the relative contributions of fast food restaurants vs convenience stores on childhood obesity

Based on a non-standard approach of generalized estimating equations, Pepe et al. (1999) and Horton et al. (1999) developed estimators for the association between univariate outcomes and multiple informant predictors. Their approach enables estimation of the marginal effect of each multiple informant predictor, and formal comparison among predictors in regard to the strength of their association with the outcome. We extend these multiple informant methods for hierarchical data structures to estimate and compare the strength of association among multiple correlated predictors while taking into account for the correlation within a cluster or a group. We applied the extended method to address two substantive questions regarding how features of the food environment near school affects child body mass index: 1) We investigate how the association between the number of fast food restaurants and child's BMIz varies across several buffer sizes from a school, and 2) compare the association of BMIz with two different features of the food environment (fast food restaurants and convenience stores). The newly developed methodology enhances the types of research questions that can be asked by investigators studying effects of environment on childhood obesity, although it can potentially be applied to other fields.