| This dissertation investigates three separate questions relevant to labor economics and econometrics. Chapter 1 focuses on workers' compensation and whether there is evidence supportive of employees' use of workers' compensation to cover off the job injuries. Using administrative data from a large nation-wide temporary staffng agency, I find that difficult-to-diagnose injuries are disproportionately likely to occur on Monday, possibly due to workers being injured off of the job during the weekend and claiming the injury occurred at work on Monday. I also find this phenomena is particularly strong for claims for compensation of lost-wages, rather than medical-only injuries. Focusing on recent reforms in California which increased the level of scrutiny placed on workers' compensation claims, I find a relative decrease in Monday injuries in California only for the difficult-to-diagnose injuries. Taken as a whole, the evidence is consistent with off-the-job injuries composing a fraction of the claims reported on Mondays. Chapter 2 studies the relationship between instructional days and student performance on standardized tests. To identify quasi-random variation in instructional days which is independent of school inputs, I examine snow-days in Colorado and Maryland and test-date shifts in Minnesota. Both sources of instructional day variation suggest more instructional time improves student performance. Chapter 3 focuses on testing for latent regime-switching, an ongoing challenge in econometrics. We take advantage of recent advances by Cho and White (2007) which proved the existence of a limiting distribution for the likelihood ratio statistic. However, we find the approximation method of Cho and White (2007) requires the specification of the alternative parameter space. If the true alternative --- something which is ex-ante unknown to the researcher --- lies outside of the pre-specified space, the power of their test can fall to zero for distant alternatives. We find subsampling provides critical values with reasonable size which also negate the drawback of pre-specified coefficient intervals. The power gains are large relative to other established tests for unobserved heterogeneity. |
