Fig. 4.

Support for the S2024 topology relative to support for J2014 for various taxon samples with and without the outlier regions. Note that the S2024 topology is always favored unless taxon sampling is reduced and the outlier regions of chromosome 4 are included (mean support above zero indicates that ASTRAL will recover S2024). y-axis: the difference between the quartet score of S2024 and J2014 trees, showing median, mean, and SE over 63,430 locus trees from the S2024 study. x-axis: the numbers of taxa (1 to 6) removed from S2024 locus trees for each group shown as a heatmap. Owing to the presence of outliers, the median (which is more robust to outliers) and mean (sensitive to outliers) diverge. Reducing the number of doves and cuckoos reduces support for S2024 compared to J2014, eventually leading to the recovery of J2014 when at least four of the five doves or five out of the seven cuckoos are removed. Excluding the outlier region of chromosome 4 makes the median and mean similar and leads to the S2024 regardless of taxon sampling.