Covariance analysis for case-control studies with small blocks.