[tmule]

Without defining truth and correct, the rejoinder makes little sense, your data science credentials notwithstanding.

What seems like a plausible interpretation to my un-credentialed, low-IQ self is: 1. Within dimension comparisons in this instance may suggest that “like” populations are similar across gender, suggesting that there is no bias in how tech compensates women and men. 2. The distribution of said dimensions differing between men and women may reflect exogenous effects that are not controllable by tech since they’re upstream of the tech hiring process.

[xenocyon]

Regarding Simpson's paradox, you can see a good treatment here [0] describing whether splitting a group is correct or not (based on causality); see Section 2.2 specifically.

[0] http://ftp.cs.ucla.edu/pub/stat_ser/r414.pdf

[jldugger]

Well, that's definitely going on the reading backlog. Thanks!