Generalized Boosted Models to Measure Racial Effects at Different Quantiles in Observational Studies

Abstract

In this paper, we consider the estimation problem of treatment effect at different quantiles in observational studies with longitudinal data. The research motivation is from the NHLBI (National Heart, Lung, and Blood Institute) Growth and Health Study (NGHS), a longitudinal cohort study that aims to discuss the effects of race on cardiovascular risk factors. Because the true propensity score model is unknown, a nonparametric generalized boosted models (GBM) method is adopted to obtain the propensity score estimator. Combining the ideas of quantile regression and inverse probability weighting, a GBM-based quantile weighting estimation method is developed for the quantile treatment effect and applied in NGHS data to measure the racial effects at different quantiles. The results indicate that the racial effect varies with different quantile levels and may not equal to zero. Under various parameter configurations, some simulation studies are conducted to assess the effectiveness and advantages of our proposed estimation method compared with the existing approaches.