Abstract:
|
To unbiasedly estimate a treatment effect on an outcome unconfoundedness is often assumed. If there is sufficient knowledge of the underlying causal structure then existing confounder selection criteria can be used to select subsets of the observed pre-treatment covariates, X, sufficient for unconfoundedness, if such subsets exists. Here, estimation of these target subsets is considered when the underlying causal structure is unknown. The proposed method is to model the causal structure by a probabilistic graphical model, identify the graph from observed data and select the target subsets given the estimated graph. Using Markov and Bayesian networks the approach is evaluated by simulation. Several common target subsets are investigated and the selected subsets are compared with respect to accuracy in estimating the average treatment effect. The proposed method is used with existing software that easily can handle high-dimensional data, in terms of large samples and large number of variables. The results from the simulation study suggest that this approach is suitable when the sample size is relatively large (>1000) and that certain target subsets yield better results than others.
|