Dealing with Limited Overlap in Estimation of Average Treatment Effects