Examining the effect of health behaviors on wages and healthcare utilization in models with endogeneity

This dissertation contains three essays on applied health economics. Although each essay is independent of the others, all three address the issue of estimating models where the relationship of interest is confounded by factors that are unobservable to the researcher. The first essay is an econometric simulation study while essays 2 and 3 address behavioral health topics. Essay 1 compares the accuracy and efficiency of parametric count data specifications paired with the Extended Olsen Model (EOM; Terza, 1998, 2009). The EOM is a nonlinear instrumental variables approach that allows for consistent estimation of model parameters when the data suffer from binary endogenous switching (e.g., endogenous sample selection or endogenous treatment). Count data models are ubiquitous in the health literature for estimating non-negative, discrete outcomes such as physician visits, hospital admissions, cigarettes smoked, etc. Essay 1 provides insight into the model selection process by informing practitioners which specification is likely to provide the most accurate parameter estimates under a variety of data configurations. Essay 1 also demonstrates the applicability of the Conway-Maxwell Poisson (CMP), a flexible count model developed in the field of industrial engineering that has yet to be utilized in the economic literature. In Essay 2 I apply a count version of the Extended Olsen Model to estimate the relationship between marijuana use disorder (MUD) and ER visits among US Medicaid recipients. This essay is the first in the literature to estimate the relationship between marijuana consumption and the demand for ER visits in isolation from other illicit drugs, thus providing an important addition to the ongoing policy regarding the potential relaxation of marijuana regulation. This study is also the first in the illicit substance literature to use an instrumental variables count data model to estimate the full distribution of ER visits, thus accounting for unobserved factors that may be jointly correlated between individual propensity for MUD and demand for ER visits. I fail to find a positive relationship between MUD and ER visits, instead uncovering suggestive, but inconclusive, evidence that MUD and ER visits may rather be negatively correlated. Essay 3 considers the relationship between wages and obesity. Although prior literature has firmly established a negative relationship between wages and obesity, it is equivocal with regard to the underlying pathway(s) through which obesity results in lower wages. Using firm-level data that gives me unique access to proxies for productivity and discrimination against obese individuals, I find that inputs to productivity, particularly health, are important confounders of the wage-obesity relationship. I fail to find any evidence of discrimination against obese employees, but I do find that among females the negative relationship between wages and obesity exists only among mothers