Table 2 compares the results of Evans et al to the analytic AUC for the diseases without strong MHC effects, using P<0.8 to select SNPs into the score, as that threshold generally gave the highest AUC. At that threshold, the choice of has little bearing on the results unless it is very close to 1, so it is set to 0. Also shown is the maximum AUC possible for each disease, obtained by letting the sample size grow to infinity. Bearing in mind that those authors noticed inflation in AUC for null SNPs, it is again clear that their modest results are compatible with the study design, and more encouraging results might be obtained from a larger sample. The calculations also confirm their observation that the allele count estimator is consistently less accurate than logistic regression; however while the two estimators give similar results at this sample size, more considerable differences emerge in the limit of large samples.
