In addition, the continuous net reclassification index (cNRI), with 5%% risk reclassification as cut-point between models, was calculated based on logistic models.
Root mean square deviations (RMSDs) of all heavy atoms between the models were calculated using the compare_structures program in the RAMP modeling suite.
Following each superimposition, the mean absolute distances between the 2 models were calculated at 4 regions: anterior cranial base, forehead, left and right zygomatic arches.
As no obvious sex differences between the estimates emerged, models were calculated for the sexes combined.
Differences in Beta coefficients between men and women in main effect models were calculated as the differences between Beta-coefficients divided by the square root of the sum of the squared standard errors for each variable.
Degrees of freedom for the test between global and local clock models were calculated by considering that for a binary tree, a local clock model has n-1+r free parameters, where r is the number of unconstrained relative evolutionary rates [41].
Empirical 95% confidence intervals for the differences in C-indexes between the two sets of models were calculated using the 2.5th and 97.5th percentiles of the differences.
In addition, average quadratic deviations between p optimal and sub-optimal models,, were calculated to indicate changes in the estimates of genetic correlations while moving away from the optimal model [ 11]. was defined as (6) where t is the number of traits and r ij,m is the estimated genetic correlation between traits i and j from an analysis fitting m PC.
Two models were calculated.
The significance of a difference in C-index between different models was calculated using the method described by Haibe-Kains et al. [12].
