5 ms·
If you read all the studies, you'll generally find M/F variance ratio to be about 1.15. If you study sufficiently many subgroups you'll get a result which diffe
by yummyfajitas 10y ago
If you read all the studies, you'll generally find M/F variance ratio to be about 1.15. If you study sufficiently many subgroups you'll get a result which differs from this. But that's completely expected even if no subgroup differs at all.
Here's some discussion of that issue in a completely different context:
https://www.chrisstucchio.com/blog/2015/ab_testing_segments_and_goals.html https://www.chrisstucchio.com/blog/2015/ab_testing_segments_...
https://www.chrisstucchio.com/pubs/slides/crunch_conf_2015/slides.html https://www.chrisstucchio.com/pubs/slides/crunch_conf_2015/s...
- gone35 10y agoNot when the subgroups are large enough, and the observed differences are wide enough! (This is what degree-of-freedom corrections are for, anyway: if the effect persists despite the corrections, it holds with high probability.) I shouldn't even have to point this out...
- yummyfajitas 10y agoDid you read the study you are talking about? For Asians, n=219. Second, the number you are citing isn't ratio of variance, it's ratio above the 99'th percentile. That ratio looks like 105 men and 114 women. So we've got a deviation from what's expected of about 18 people. Note that the article didn't do any statistical test on this number (with or without a degree-of-freedom correction). That's unsurprising, given the small sample size. tl;dr; the subgroups are not large enough and the observed differences are not wide enough. Feel free to run a statistical test to prove me wrong.
- gone35 10y agoOh I thought the SI had some kind of analysis.. A priori though, I do think the differences point at misspecification. Too busy/lazy to formalize this claim though, I admit..