We generated a sample of 2,000 diploid cases and 2,000 diploid controls from the panel as follows. For the controls, we sampled haplotypes uniformly from the panel (without replacement) and combined them in pairs. For the cases, we sampled haplotypes according to the genotype frequencies at the causal SNP as dictated by the disease model. Specifically, we first simulated genotypes at the causal SNP by sampling with probabilities proportional to:
