Brandon Pope‘s (2011 graduate of the lab) dissertation is about modeling incentives in healthcare systems. This paper is a critical component where he explores the dependencies in health behaviors (diet, exercise and smoking) modeling as binary decisions and explores modeling the dependence through either joint attraction functions or probabilistic dependence. We find some evidence of superior performance of the joint attraction function approach for our data.
Abstract
The prediction and control of distributed healthcare behaviors within a population such as smoking, diet, and physical activity are of great concern to those who pay for healthcare, including employers, insurers, and public policy makers given the significant effect on costs. In considering the selection of multiple health behaviors, the nature of dependence between behaviors must be considered because simplifying assumptions such as independence are untenable. Using data from the National Heart, Lung, and Blood Institute, we find strong evidence to reject the hypothesis of independence between the aforementioned behaviors, while finding some evidence of conditional independence. In this paper, several alternatives to the assumption of independence are presented, each of which signi cantly improves the ability to predict combined behavior. We present models of dependence through marginal probabilities and, taking inspiration from non-expected utility maximizing behavior, through attractions to behavioral alternatives. We find that consistently healthy (or unhealthy) combinations of behaviors are more likely to occur relative to the assumption of independence. We discuss how our results could be used in designing policies to curtail costs and improve health.