Estimating average causal effect using backdoor

Effect estimation with backdoor amounts to estimating a conditional probability distribution. Given an action A, an outcome Y and set of backdoor variables W, the causal effect is identified as, \(\sum_wE[Y|A,W=w]P(W=w)\).

We can use any estimator that can output conditional expectation. DoWhy supports the following three types of estimators for average causal effect.

• Regression-based methods
• Distance-based matching
• Propensity-based methods
• Do-sampler