The effects of perceived racial discrimination on cardiometabolic risk markers in the elderly US population

This thesis leverages state-of-the-art machine learning techniques (Bayesian Causal Forests and Bayesian Weighted Estimators) to estimate the causal effect of perceived racial discrimination on the cardiometabolic risk markers systolic blood pressure, BMI and waist circumference in the elderly United States population from data from the Health and Retirement Study. In doing so, this thesis also investigates the consequences of consideration of sampling weights that ensure representativeness for a population. Posterior evidence is found for a lower waist circumference of 0.35 and a decrease of 0.21 inch on average as a result of perceived racial discrimination for the elderly United States population, respectively in a cross-sectional (2016) and longitudinal (2012 and 2016) analysis. Furthermore, this thesis finds evidence that blacks and African Americans have particularly strong effects on BMI (i.e., 0.80 point higher in the cross-sectional analysis and a 0.41 point decrease) and waist circumference (a 1.48 mmHG decrease in systolic blood pressure in the longitudinal analysis) compared to others. This thesis also finds substantial evidence suggesting that those who have the most adverse health effects as a result of racial discrimination are generally the more vulnerable people in the elderly United States population. Compared to previous findings in the health literature and on common interventions targeting cardiometabolic health, particularly the estimated effects on BMI and waist circumference suggest that perceived racial discrimination has serious effects (both beneficial and adverse) on cardiometabolic health. This thesis further concludes that sampling weights can have severe consequences for the uncertainty of estimated effects. These findings suggest that researchers should address potential heterogeneity of effects when estimating causal effects, and must be careful in extrapolating findings from small scale studies to larger populations.