Fig. 3.

Recalls of Bayes classifiers for first through sixth-degree relatives. Results are from classifiers trained on (a) exact and (b) inferred segments with features n, r, or (r, n). The recalls for both (a) and (b) are calculated using the uniform distribution of 3,000 pairs per degree and averaged over 80 independent runs. For each degree, the lower subplot shows the corresponding significant ($P<{10}^{-4}$) change in recall between classifiers (r, n) and r [positive values have greater recall in the (r, n) classifier]. Error bars indicate one standard error.