Three essays in health economics

The first chapter of this thesis studies the relationship between obesity and school performance among children aged 16 to 17 years using data from the National Longitudinal Survey of Children and Youth (NLSCY). OLS, quantile regression and IV estimates of obesity all indicate there is a negative relationship between obesity and school performance. Quantile regression estimates indicate being obese has a strong negative impact among children at the 75th percentile for the full sample. Being obese seems to have an impact on school performance among high achievers but not among lower achievers. In addition, IV estimate of obesity indicates a strong negative effect of obesity on academic performance. The second chapter studies the effect of retirement on various measures of health among Canadians using data from the Canadian Community Health Survey (CCHS). Both OLS regressions and a fuzzy regression discontinuity design are used to capture the effect of retirement on health. OLS estimates suggest retirement is associated with a more physically active and less stressful life among retirees in the full sample and the female subsample. On the other hand, 2SLS results from regression discontinuity design indicate retirement has a negative impact on mental health among retirees in the full sample. Retirement causes a decrease of 1.48 point (around 148% of one standard deviation) in the standardized mental health score, and the estimate is robust across different bandwidths. The third chapter uses a linear regression model to study peer effects on adolescent weight status using data from Add Health. OLS estimates suggest the adolescent’s own BMI is positively related to average peers’ BMI for the female subsample. Since OLS estimates would be inconsistent if omitted variable bias exists, the linear regression model is re-estimated under different degree of correlation between the main explanatory variable and unobservable factors. Estimates under those relative correlation restrictions suggest OLS estimates are quite robust when omitted variable bias exists. The sign of estimated peer effects would only change if the correlation between the main explanatory variable and unobservable factors is as much as three times the correlation between the main explanatory variable and other controls.