For the last 30 years over 40% of graduating math majors where where women. That drops to 15% of tenure track mathematicians being women, but again that's a long way to sub 2%.

This is the exactly the pattern you would expect to see though if the variance of mathematical ability is higher in men than in women, even if mean ability is precisely identical.

The sex ratio gets progressively more extreme as you go further out in the tails.

Not really, you get a fat tail effect at extreme ability reducing the differences. A more likely cause are highly capable women bailing on the field.

EX: Women live significantly longer on average and the oldest women lived 6 years longer than the oldest man. Yet, the 16th oldest person was a man and 6% of oldest 100 people where men. And 6% of the top 100 living people are men https://en.m.wikipedia.org/wiki/List_of_oldest_living_people

I'm not sure what you mean—why would you assume there is a fat tail expect in the first place, and not something approximating a normal distribution?

And even if you presume something like a Pareto distribution, the likelihood ratio between two distributions grow through the tails if their variance is not identical.

edit: I see you bring up longevity, but I don't see why this is relevant to a discussion about variance in mathematical ability or intelligence? See: https://www.sciencedirect.com/science/article/pii/S019188690...

Based on a wide range of ability testing we see fat tails (edit: more black swans than expected), it would be more surprising if they where skinny.

Granted, we can't measure very high ability very well due to sampling bias. I am simply saying even if there is a modest bias that's not enough it would have to be huge to account for these numbers.

So, I am bringing up something else with the kind of distribution we are talking about which has more accurate data. Women live ~ 5% longer both looking at the average lifespans and oldest examples which is a very significant difference. Yet, the oldest population has more men in it than you would think.

Edit: Math: 6 year longer lifespan + 50% risk of death per year = you would expect ~1% of top 100 oldest people to be men.

Life span here is interesting.

If it’s affected by things like work place deaths because men are more likely to take on dangerous jobs, e.g. sea fisherman, military service, etc. would that overlap with the section of the population likely to be working on pure mathematics?

Suicide, another cause of that difference in average life span, would overlap though I guess.

It would be worth properly investigating as I suspect there’s a lot of complexity hidden here.

> Based on a wide range of ability testing we see fat tails

What kind of ability? Mathematical? IQ?

IQ, memory etc, in uncalibrated tests you see something close to a bell curve in raw scores but more people score very high 180+ than you would get from an actual bell curve. So, many tests are given a max score or compressed at the high end.