Fig. 4

Logistic regression coefficients to explain dataset stability. a Filled circles represent logistic model coefficients that are significant at \(P\le 0.01\), and error bars represent the 99% confidence intervals. Values in the top left of each panel represent the AUC from an associated ROC (receiver operating characteristic) analysis. Note that effect sizes for “mean number of sites” and “number of sequences” are significant but extremely close to zero. Dataset sizes in each logistic regressions are equivalent to the number of datasets explored: 236 PANDIT datasets, 1000 Drosophila datasets, and 1000 Euteleostomi datasets, for each of AA and NT datatypes. b Distributions of predictor variables used in each logistic regression