Unlike personality classifications (which are essentially witchcraft with better marketing), gender is objectively verifiable (yeah yeah gender studies students, I passed the class too, thanks for your comment) and so makes for a much more productive area of research. Trivial algorithms (~60 lines of awk last time I did it) are very capable of pegging gender with 80%+ accuracy for some corpora.
You're speaking of "sex," not "gender" -- the former is biological and there are two of them; the latter is cultural or social and there could be arbitrarily many.
I say this while being a conservative; the semantic distinction is a useful one, although attempting to blur the distinction between a (biological) sex and a (cultural) gender never does any good.
Why couldn't patio11 be speaking of gender? In fact, I think a study measuring gender would identify with even more accuracy, although data collection would be harder (you would need the authors to report how they identified). Anyways, the linked study uses the word "gender" - it's probably conflating the two, as often happens.
Which gender is. Your gender is how you identify, so unless you are lying to a researcher who asks you what gender you identify as, it's objectively verifiable.
You need a larger sample size before being impressed or unimpressed. If it's 74% sure of something, it should be wrong a quarter of the time; if not, its confidence estimate was too low.
