Overview

Abstract (English)

A number of studies have examined whether education has a causal impact on health, with some finding a positive effect and others no significant effect. Evidence on the relationship between education and health in the Canadian context is however scarce. The first two chapters in this thesis address this gap by presenting econometric analyses of the effect of education on health using Canadian Census data. Chapter 1 uses data on changes to the compulsory schooling laws in Canada as instruments for educational attainment. Using the Canadian Census master dataset for the ten provinces, the results in Chapter 1 indicate that education does indeed have a causal impact on mortality (a robust indicator for health). Increasing pre-tertiary education by an additional year reduces the 5-year adult mortality rate by between 2% and 3.5%. These findings suggest that traditional methods of evaluating the benefits of education may in fact understate its true importance. In Chapter 2, the instrument used for education is potential eligibility for the university education benefits under the Canadian Veterans Rehabilitation Act. Because the VRA affected large numbers of anglophone Ontario-born men and extremely few francophone Quebec-born men, the latter is used as the control group for the treatment group consisting of the cohorts of Ontario men of university age at the end of World War II. Using the Canadian Census Public Use Microdata files for Quebec and Ontario, the results from the main analysis in Chapter 2 suffer from large standard errors due to measurement error in the dependent variable and a sample size that is too small. An alternative analysis using administrative data on mortality rates yields results suggesting that an additional year of tertiary schooling reduces the 5-year mortality rate by 1.9%. Inspired by the measurement error problem in Chapter 2, Chapter 3 is an empirical investigation of the impact of measurement errors on estimation results. The main contribution of this chapter to the literature relates to implications for statistical inference of random versus non-random measurement errors in the dependent variable, based on Monte Carlo simulation results.

Abstract (French)

Please note that abstracts only appear in the language of the publication and might not have a translation.