I'm not quite sure how to interpret this study. The decrease from 2017 to 2019 is so large (2.4% to 1.6% with expected error of about 0.3%)--especially if you consider the AMAB subset--that it seems likely to either conflict with other studies on the stability of transgender identity, or indicate that there were substantial flaws in sample selection.

I would treat these results with a large degree of skepticism; large sample size with poor methodology means that you get a precise result, but not necessarily an accurate one, because what you're measuring isn't necessarily what you want to be measuring.

(Another commenter noted the short baseline, which is also a problem.)

(This is a comment only about the study, not about your observations in general--that all seems sensible enough, or at least I lack the experience to know whether it's sensible or not.)