Do Resampling Metrics Predict Optimality-Based Support in Parsimony Analyses? An Empirical Evaluation Using Phylogenetic and Phylogenomic Datasets

Goodman-Bremer (GB) is a direct (=optimality-based) measure of support in parsimony analyses, calculated for each clade as the difference between the optimal tree cost and the next-best tree cost without a given clade. In contrast, although bootstrap (BS) and jackknife (JK) are more common in phylogenetics, both present some shortcomings because they (1) are not proportional to optimality, (2) may give high values for clades absent in the optimal tree, and (3) can be underestimated or overestimated in some scenarios. Despite these disadvantages, if resampling and optimality-based metrics are correlated to each other, resampling estimation could be an indirect measure of support. Here we tested the correlation between GB, BS and JK using empirical datasets. Our results revealed a weak correlation between resampling metrics and GB in parsimony analyses. However, BS and JK either underestimate or overestimate GB in most nodes, corroborating a previous study testing the same in a maximum likelihood framework. These results are robust when we vary pseudoreplicate numbers. Our findings highlight that neither bootstrap nor jackknife are good predictors of GB. Thus, while GB is adequate to measure support of clades, resampling metrics should be cautiously interpreted and its widespread use as support rethought.