TechCrunch messes up the math on sexism
crazybear.posterous.com
crazybear.posterous.com
- in mathematical ability
- among schoolchildren
- on a standardized test
and generalizes them to:
- many different types of abilities
- among the Y Combinator applicant pool
- on the Y Combinator application process
Not only that, but by its end, the article is postulating a model in which literally one in a million people have sufficient aptitude to be accepted by Y Combinator - whether they're interested in it or not!
Even if you grant the author all of those assumptions, plus the risk aversion thing, you're only down to 13% women, three times what TechCrunch says Y Combinator actually accepts. I guess you could add another independent 99.9th percentile ability requirement, but then you're talking about one in a billion people being Y Combinator worthy. Or you could try to find a different, more tilted risk aversion statistic -- but at that point I think we'd be cherry-picking citations to fit a conclusion.
So, sure, if you accept a whole raft of dubious assumptions, you can explain the 4% acceptance rate by aptitude alone. But enough dubious assumptions can explain almost anything.
Why do you feel this is unreasonable? Do you think there is some huge pool of people good enough for YC who just don't get in? Maybe the true numbers are 1 in 100,000, but I'd be surprised if they are much higher than that.
In any case, I'm not attempting to claim my assumptions are correct. The only point I'm making is that small differences in underlying probability distributions can have large effects in the composition of people accepted into a highly selective program.
I.e., Eric Ries is making the fallacy of the excluded middle: "either aptitude/preference differences are huge, or else they don't explain much." This is simply mathematically incorrect.
I actually agree with you here. This is why I think things like sexist/marginalizing presentations at conferences are a big deal: a hostile environment seems to me a lot more likely to achieve this than possible differences at the bare fringe of the aptitude spectrum.
Yes, I believe that there are probably more than 300 people in the United States who could potentially be accepted to Y Combinator. I'd believe there are a lot more than 3000. We're talking about aptitude here, remember, so you have to consider people who could potentially get in, but don't apply. Do you really believe that the vast majority of the people who could even conceivably get in to YC, given some set of life choices, do in fact apply?
The only point I'm making is that small differences in underlying probability distributions can have large effects in the composition of people accepted into a highly selective program.
Yeah, if we're willing to grant you all kinds of crazy concessions, you can cobble together an explanation. But you're asking for a lot to make YC selective and sensitive enough for your example to be reasonable. The YC selection process has to be consistent and incredibly accurate. Most of the people who are even capable of getting in must apply. Acumen in various areas has to be minimally or negatively correlated, and to even be considered for YC, one person has to have all of the relevant skills.
And this is just to get close to the gender disparity that YC sees.
Sure, I agree that normal distributions with minor differences in variance are very different at the tails. But to get this to explain away the gender disparity in YC selectees? Epicycles on top of epicycles.
(Also: if this is really about the tails of the aptitude distribution, going by the paper you cited and your same arguments, we'd expect to see a lot more Asian women than Asian men in YC. I suspect this is not the case.)
If anything, I suspect your risk aversion point is closer to the mark, but that opens up an entirely new cultural and political can of worms.
> people who are in the top 2.5% in mathematical ability alone, i.e. people with the capability to be decent programmers
Decent? Top 2.5% in mathematical ability can be "decent" programmers? Most programmers don't do rocket science. You don't need to be a world-class software engineer to make most of the web apps that come out each day. CRUD operations on a database, ETL code, implementing someone else's graph algorithm... these do not require over 2 sigmas in raw math skill to do. A "decent" programmer can do these things just fine.
When you're coming from such a warped worldview, it's not too surprising that he thinks only 1 in 1000 people have the math skills alone to write code for a Y Combinator startup and that successful applicants are literally 1 in a million.
Dunning-Kruger strikes again.
This is a counterpoint to Eric Ries, who claims:
"Here in Silicon Valley, we’re looking for the absolute best and brightest, the people far out on the tail end of aptitude...How far apart would the two populations have to be to explain YC’s historical admission rate of 4% women? It would have to be really extreme."
The only point is to do the math and demonstrate that small differences can add up to a lot if you have highly selective filters. I chose mathematical ability as an example because a) it's fairly well studied and b) Eric Ries cites a slideshare presentation which also uses it as an example. (However, the data used in the presentation are about 20 years older than mine.)
"Small differences in the distributions of men and women don't allow you to predict much about any individual trait of a randomly selected individual. But when you limit yourself to the far tails of probability distributions, they do. "
So, he's saying that by being more selective for a field that men have greater variance in, you're more likely to find more men at the far tails.
I'm not saying either of those articles is right, but clearly none have shown why that difference exist.
What percentage of applicants to YC are female?
Acceptance rate in terms of applicant pool for each gender? (ie "10% of all male applicants are accepted vs 14% of all female applicants" -- and I chose to skew it that way intentionally because I am betting if there is a gender based selection bias it is probably in that direction)
These statistical analyses are meaningless outside of that context. If it just so happens that only 4% of the applicants are female, then why should the gender balance of final selections be any different from that?