Saying it's a direct A/B comparison is seriously overstating it. Transitioning is itself a huge confounder, and if it were true that time before/after were exactly comparable, that would debunk one of the main justifications for allowing sex-changes in the first place!

Of course, the sample size is small here. And there’s no perfect agreement on cause-and-effect. Chris Edwards, a trans advertising executive, says that post-transition, he was given greater levels of responsibility—but he thinks it’s because the testosterone he took changed his behavior. He became less timid and more outspoken—and was seen, at work, as more of a leader. Indeed, some suggest that transmen might experience these workplace benefits partly because, post-transition, they are happier and more comfortable, and that this confidence leads to greater workplace success. But if that’s the case, one would expect that transwomen, armed with this same newfound confidence, would see benefits. The opposite seems to be true.

Note the willful incomprehension of the author about the possible effects of things like testosterone. 'Opposite seems to be true' my ass. But I suppose materialism and individual differences should never be allowed to get in the way of a good story about endemic sexism and racism...

(Sadly, this is only the second most infuriating statistical argument I've seen today. The first is a linear regression in the Washington Post about whippings vs productivity for slaves, in which they claim it shows whipping works. Aside from the usual correlation!=causality problem, their scatterplot clearly shows that there is not such a small positive correlation: their model does not fit the data because most slaves were never whipped so it's not Gaussian but more like a zero-inflated model, and in the population that was whipped a non-zero number of times, more whippings correlate dramatically with decreased cotton production. At a guess, male slaves were much more likely to act out or run away or get into fights or refuse to produce, and would be whipped for it. It borders on malpractice to present this graph baldly without including sex as a covariate or better yet doing a mixture model - certainly any model diagnostics would flag this regression as bogus. The author's bio says he's a professor at Columbia who "studies the roots of poverty and violence in developing countries, especially Africa"; all I can think is that if that's what passes for analysis for him, then no wonder Africa remains poor and violent.)