Table 2.

Average Kendall's τ and Pearson's r for different versions of the MQA method using Benchmark A. Note that while Kendall's τ is improved by using non-contact constraints, Pearson's r is decreased. The inclusion of non-contacts decreases the linearity of the correlation, but improves the ranking of models. We have argued that we prefer Kendall's τ over Pearson's r, and so we consider the non-contacts to be beneficial to our MQA method.

Constraint Set	τ̅	r̅
All	0.570	0.825
Best fraction	0.575	0.833
Optimized	0.582	0.838
Opt+noncontacts	0.589	0.827
Opt+models	0.622	0.866