The results of the regressions on the randomisation scenarios.

The 3-letter scenario name as in 7. For each of tone1, tone2 and tone counts, I show the percent of models fitted on permuted data that, compared to the model fitted on the original data, have a better AIC, a smaller (i.e., more negative) regression coefficient β for ASPM-D, and a smaller β for MCPH1-D, respectively. The lower these percentages, the more “special” is the relationship on the original dataset relative to the structure that is destroyed in each scenario. In bold are the values < 5%.