Adolescent obesity is now recognized as having a complex etiology. Ecological models are often used to explain this complexity, with different levels of influence at the individual, familial, peer, school, and neighborhood levels. Not only are there multiple levels of influence, but these different levels have bidirectional relationships of influence between them, further adding to the complexity. While the ecological framework is often used to conceptually explain obesity, there are few examples of statistically testing this type of multilevel model. This proposal will develop and examine new statistical models and methods incorporating latent variables, social networks and propensity scores that simultaneously incorporate multiple variables from multilevels of influence on adolescent obesity in order to inform a richer understanding of the phenomena. Using existing data from a large population-based longitudinal and serial cross-sectional survey of ethnically and socioeconomically diverse adolescents (Project EAT), statistical models and appropriate estimation methods for them which are computationally feasible will be developed. In addition to extensive individual level questionnaire data and measured heights and weights used to calculate BMI, adolescents also provide lists of nominated friends providing information about peer-networks. These data are all collected at school. Furthermore, school environmental and policy data are collected as well as the adolescent's residential neighborhood information. The Project EAT data source provides the richness that new, more completely developed hierarchical models could exploit in building our understanding about adolescent obesity. None of the analyses being proposed were part of the original analysis plan of Project EAT and each represents our attempt to go a step beyond what has been done.
The prevention and reduction of adolescent obesity requires a better understanding of the complex interplay among its multiple influences including neighborhoods, schools, family, peers and peer-network, and the adolescents'individual characteristics and behaviors. This proposal will develop and examine new statistical models and methods incorporating latent variables, social networks and propensity scores that simultaneously incorporate multiple variables from multilevels of influence on adolescent obesity in order to inform a richer understanding of the phenomena.
Showing the most recent 10 out of 19 publications
