Single marker tests (Cochran-Armitage test) were performed at each SNP on the genotyping chip where information were simulated from the relevant HapMap panel. Multi-marker tests of association were performed in an identical fashion with the marker being formed by the multi-marker haplotypes known to tag HapMap variation. To avoid over estimation of power, multi-marker tags chosen to tag the current putative disease SNP in the simulations were excluded from the test set. Tests at imputed SNPs took account of the uncertainty in genotypes through a missing data likelihood as described in [14].
