Mean AUC versus models with various attributes removed.

<p>The AUC for voting before (red lines) and after (blue lines) an answer is accepted, as well as accepting an answer (green lines) averaged over two through twenty answers, weighted by the amount of test data, in (a) technical, (b) non-technical, and (c) meta boards. See main text for descriptions of models. Error bars are standard deviations of the mean values. We remove attributes and measure the drop in mean AUC to determine the relative importance of various attributes. The statistical significance of the drop in AUC values compared to the full model is as follows: “***”, p<0.001; “**”, p<0.0028 (Bonferroni correction for 18 variables with <i>α</i> = 0.05 [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0173610#pone.0173610.ref042" target="_blank">42</a>]); “–”, not significant.