Employer provided health insurance status and job separation of obese workers

This dissertation analyzes, in relation to Employer provided health insurance (EPHI), the behaviors of employers and obese workers. Chapter 1 examines how EPHI status varies over time after job separation, depending on obesity. Estimation results from the weighted probit model using the 1996-2005 Medical Expenditure Panel Survey (MEPS) show that the probability of having EPHI after voluntary job change is not significantly different for obese and non-obese workers. However, obese involuntary job separators tend to have 13-18 percentage points lower probability of having EPHI than do job stayers, compared to their non-obese counterparts. We also found that status of other fringe benefits is not closely associated with obese workers' involuntary job loss. This result could be evidence that employers are reluctant to hire obese workers due to their higher health care costs. Chapter 2 examines whether obese workers are more likely to be job-locked due to EPHI, and whether they are more likely to be separated involuntarily from their jobs with EPHI. According to estimation results using the 1996-2005 MEPS, a negative interaction effect between obesity and EPHI on voluntary job change appears, particularly for the single or never-married group. This negative interaction effect may be caused not because obese workers with EPHI are less likely to change their jobs, but because obese workers without EPHI are more likely to change their jobs. We do not find evidence that either obese male or female workers are more likely than non-obese workers to leave their jobs involuntarily when they have EPHI. For other fringe benefits, we obtain similar results. We interpret this result as evidence that obese workers are not regarded as the target of different treatment in relation to EPHI. Chapter 3 suggests a new technique to compute average marginal and interaction effects, and their standard errors after the estimation of non-linear models with interaction terms, which was used in Chapters 1 and 2. The new technique using bootstrap replications can be applied generally to any non-linear model.