GMM logistic regression model for obesity with time-dependent covariates


Rhodes, Rachael; Wilson, Jeffrey; & Fang, Di (2016). GMM logistic regression model for obesity with time-dependent covariates. 2016 Add Health Users Conference. Bethesda, MD.


The prevalence of obesity is a major public health concern with its causes confounded in the fields of biology, social-economics, and psychology. The progression of obesity through time calls for special attention as there is likely a causal effect between behaviors in a prior time period and the obese outcome later. The objective of this study is to identify the prevalence and risk factors of obesity based on Add Health data in the US while accounting for feedback between obesity and its contributing factors. To this end, we employ the Add Health data (public data). The outcome measures were height and weight that were calculated into the body mass index (BMI). The BMI is then translated into a binary variable indicating whether the respondent is obese. To facilitate comparison between models, we first employed a standard logistic regression that accounted for the correlation between present covariates and future outcomes as well as present outcomes and future covariate measures. Furthermore, we accounted for the correlation inherent from the repeated measures of the adolescent as well as the correlation realized on account of the feedback created between the responses at a particular time and the predictors at other times. We then explore an alternative approach using a generalized method of moments (GMM) with time-dependent covariates while relaxing some assumptions about the feedbacks. The GMM results showed that the variables hours spent watching television, participation in vigorous physical activities, moderate physical activities and low-impact physical activities, what the children thought about their health, gender, race and time point 2 and time point 3 are significant factors of obesity. Adolescents with disadvantaged socioeconomic and community status were at higher risk of obesity. Significant community- and household-level variations were also found. Our study contributes to the existing public health literature by providing a novel approach to evaluate obesity related drivers while accounting for the feedback of such factors through time. Failing to address such feedback effects can lead to biased estimation and unfounded implications. In this sense, our study presents a more robust interpretation of the issue on obesity utilizing the longitudinal survey provided by Add Health.


Reference Type

Conference proceeding

Book Title

2016 Add Health Users Conference


Rhodes, Rachael
Wilson, Jeffrey
Fang, Di

Year Published


City of Publication

Bethesda, MD

Reference ID