|
Abstract:
|
Estimation of causal effects for continuous exposures is of high interest in many applied fields. In order to estimate robust effects of continuous exposures, attention must be paid to how one is handling bias from observed pretreatment confounders. In this study, we will compare the relative performance of three different methods (generalized boosted models [GBM], covariate balancing propensity score [CBPS], and entropy balancing [EB]) for introducing high-quality weights for estimation of the causal effects of continuous treatments. We will do so within the context of an applied study examining the causal effects of post-traumatic stress symptoms on substance use outcomes among youth in enrolled in an evidence-based substance use treatment program. In doing so, we will share practical guidance on the utility of using multiple estimation methods in the first phase of a study and discuss the implications of using different metrics for assessing balance. We will highlight how the weights from each method can be used to estimate the dose-response function through weighting and discuss the needed assumptions underlying these methods.
|
