Three essays in the application of obesity economics to intervention evaluation and food access
Obesity remains one of the greatest public health concerns in the United States due to both its cost and complexity. To effectively identify the potential causes and evaluate the effectiveness of interventions to reduce obesity requires multiple dimensions of analysis to ensure the goals of analysis are being adequately addressed. This paper contains three papers that span the spectrum of obesity research, from potential causes in the local food environment, to potential programs to address obesity in the worksite.
The first chapter addresses the effectiveness evaluation of weight loss programs, which usually focuses on significant changes in weight. However, the underlying goal of weight loss is to reduce weight to reduce health risks. This requires a second dimension in the effectiveness evaluation, since the relationship between body mass index (BMI) and obesity health risks is non-linear. Severity can be used to address this dimension by using the squared depth of obesity, which can better detect changes in BMI that indicate changes in obesity risks. When used in the time-effect analysis it can identify important heterogeneous responses to the treatment, which can be used to direct future research.
The second chapter addresses the missing data dimension of the cost-effectiveness analysis (CEA) of weight loss programs. Most previous studies ignore missingness in the CEA of weight loss programs, but this could result in biased results. Comparing two sample selections reveals that the analysis is sensitive not only to the missingness mechanism, but also which outcome is considered, and the experimental design.
Finally, the third chapter considers potential causes of obesity in the local food environment. Although the spatial dimension of access is often overlooked, it could help better explain previous mixed findings. Access hypothesizes that living further from a supermarket increases the probability of obesity, which would result in spatial variation in the distribution of the outcomes relative to supermarkets. Spatial scanning statistics can detect this specific type of variation, which can then be formally tested for in conditional regression analysis.