[streptomycin]

Even if mathematical ability is totally determinative, and being a programmer required top 0.3-0.5% of mathematical ability, we would expect to see ratios of maybe 65/35 in the programming world, not 90/10 or 95/5.

That's true. But he described multiple factors that bias men towards programming. Even if individual factors (like mathematical ability) only cause 65/35 differences in isolation, combining a few of them together results in pretty extreme distributions.

[rayiner]

The difference in mathematical ability is about the starkest difference between the aptitudes of men and women. You expect a 65/35 difference if mathematical aptitude is the only thing that matters, and if a top 0.5% level of mathematical aptitude is required for programming. If mathematical aptitude is say only 40% of what makes a good programmer, and the factors in the other 60% skew less than mathematical aptitude, then you'd expect a more even ratio than 65/35. Thus, consideration of other factors dilutes the analysis.

[streptomycin]

Let's say being a programmer requires X, Y, and Z. X is 65/35 male/female, Y is 60/40, Z is 55/45.

The odds of finding a woman with X, Y, and Z is 0.35 * .4 * 0.45, about 5%.

The odds of finding a man with X, Y, and Z is 0.65 * .6 * 0.55, about 20%.

All else equal, you'd find that about 80% of programmers are male.

Obviously my example is contrived, but if you think I'm wildly wrong, please respond with something math-based, even if it's equally contrived. Then we'll at least know what the other is really getting at.

[rayiner]

Your math obviously doesn't make any sense, since say a 65/35 representation of men versus women among people who score a perfect Math SAT doesn't mean there is a 65% probability of finding a man with that characteristic and a 35% probability of finding a woman with that characteristic.

I was thinking more like this: consider giving everyone a goodness score. At first, let's say the goodness score is entirely based on mathematical ability, which skews strongly in favor of men. Then, let's make the goodness score based 50% on mathematical ability, and 50% based on an independent criterion that skews in favor of men, but less strongly. There should be more women who achieve a certain goodness score under the second set of criteria than the first. That's what I mean when I say that considering additional criteria is dilutive.

Besides that, all your metrics skew in favor of men. But out of the various metrics that go into being a good programmer, I think mathematical ability is the only one that skews in favor of men. Studies show that female students get higher grades, because they are more attentive over longer periods, have more patience and impulse control. Also, in terms of programming teams, I think being social and communicative is far more of an asset than being competitive.