The impact of imperfect LD has been well characterized for multiplicative models, both in terms of effect sizes and power [Chapman et al., 2003; Pritchard and Przeworski, 2001; Zondervan and Cardon, 2004]. Measuring the LD using the squared correlation (r2), a well-known rule of thumb is that a sample size of roughly N/r2 is required at a marker in order to have the same power to detect an association as a study with sample size N that types the causal SNP [Pritchard and Przeworski, 2001]. Here we derive a similar result, showing that a sample size of about N/r4 is required to maintain equivalent power to detect a deviation from a multiplicative model.
