Results of GMDR analyses.

Each testing set corresponds to 1/5 of the data. The same individuals are in each testing set across models and within type of analysis (unadjusted or adjusted). The individuals are not necessarily the same in the testing sets across type of analysis because of the smaller number of individuals that were available in the adjusted analyses compared to the unadjusted analysis (see the text for details). The average is the average over the five testing sets and the P-value corresponds to χ2 tests of fitting the models to the testing sets or the whole data.