Abstract:
|
There is a growing interest in estimating causal effects with multiple treatments using observational data, as the randomized experiment may not always feasible for ethical or practical reasons. One of the most significant issues for deriving an accurate causal effect from observational data is that the treatment effects are always confounded with background covariates. We formulate a robust method that can extend the binary treatment group to more than two groups, reduce model imbalance, confounder effects, and lower bias, thereby improving statistical power and increasing accuracy for causal estimation.
|