Dealing with small sample bias in post-crisis samples