Bootstrap distributions of P 2 for Spinosaurus sensitivity cases 4–7.

Bootstrap analysis was used with 2000 trials to predict P 2 , the posterior probability of Spinosaurus belonging to the class of “subaqueous foragers.” Each bootstrap trial contains the results of 100 random trees, so there are a total of 200,000 predictions. Histograms show the distribution of P 2 for the Spinosaurus sensitivity analysis variations 4–7 of Table 8 . Vertical gray lines and numbers along the top of each chart show the medians and their 95% CI, as determined by the BCa bootstrap confidence integral algorithm. Vertical red lines show the 2.5%, 50%, and 97.5% quantiles of the bootstrap distributions of P 2 . In a case where the bootstrap distribution has the same median as the original dataset prior to bootstrapping, there would be no bias. In general, however, bootstrapping can introduce bias. The BCa bootstrap algorithm adjusts the bias and also corrects for nonconstant variance. As a result, the BCa 95% CI does not always line up with the quantiles of the bootstrap distribution ( i . e ., gray lines and red lines may not overlap). (TIF)