Table I shows the distribution of the three outcomes for simulations across different disease models and RRs. We see that much of the time when we detect association, the deviation test will also give the correct outcome, even at the smaller effect sizes. This is despite the distortion effect observed above. The reason for this is that the LD between the causal and hit SNPs is often quite high, and thus will not suffer from much distortion. Figure 6A shows a typical LD distribution for a set of simulations—most of the time the hit SNP is at the extremes of the LD spectrum. Correspondingly, Figure 6B shows the distribution of outcomes for a given amount of LD, and Figure 6C shows the outcome of the deviation test among detected associations only. We see that, as the LD decreases, the relative amount of distortion among detected associations gradually increases. The overall proportion of associations detected without deviation may seem slightly small (i.e. the yellow bars in Fig. 6B), but note that this is in a sense “competing” with the no-association outcome
