Effect Inference from Two-Group Data with Sampling Bias