>"So when a team lacks diversity, that’s a bad sign. What are the odds that the decisions that were made to create that team were really meritocratic? "
By what kind of a priori logic does the author presume that any given meritocratic filter will happen to have an equal pass rate along arbitrarily chosen orthogonal characteristics such as race and gender?
For any given test for any kind of ability, I would be shocked if the population selected happened to perfectly fit the hiring-brochure rainbow. It's just not very likely. It is very possible that a more meritocratic selection procedure results in less diversity.
In fact, that's precisely what happened when California passed a referendum in the late 90s to forbid considering race in admissions to the University of California. The process became less biased and the results became less diverse. My apologies to the author's preconceived ideas.
>"Demographic diversity is an indicator. It’s a reasonable inference that a group that is homogeneous in appearance was probably chosen by a biased selector. Even if men have an innate advantage at software development, the gap would have to be massive in order to explain why startup after startup has an all-male team."
Innate biological gender differences are not the only cause of differential average programming ability between gender populations. Sociological factors matter too.
While some of these sociological forces may be unjust and we might want to address them, that does not change the fact that by the time they reach adulthood the population of qualified programmers has many more males than females. Thus, a lack of diversity can emerge from a just meritocratic process.
> By what kind of a priori logic does the author presume
> that any given meritocratic filter will happen to have
> an equal pass rate along arbitrarily chosen orthogonal
> characteristics such as race and gender?
I don't think that's a correct reading of the argument. The right question to ask is "What would have to be true about the underlying population to support the observed result that a particular startup is 100% male?"
Let's assume that there is significant "differential average programming ability between gender populations" and that this has a combination of biological and sociological causes. Even so, I think we can agree that the two populations are probably normally distributed around each average, right?
Now, is it safe to assume a roughly equal standard deviation for both curves? I'm not aware of any data to suggest otherwise.
So now we can be specific: how large would that differential have to be before we find ourselves in the part of the curve that has many men but basically no women? I haven't done the math, as it seems intuitive to me that this is going to yield an absurd result, namely, that men would have to have an overwhelming advantage.
At that point, it seems more reasonable to me to infer that we have a selection bias in our filter, than that the curves are so far apart in reality. My personal, albeit anecdotal, experience has given me no reason to doubt that conclusion, either.
So, I don't think the argument depends on an equal pass rate, and I think it stands up even if you believe in substantial gender differences, regardless of cause.
Doing a startup is not just about programming ability. Determination, desire, competiveness, a willingness to take risks, etc, are all far more important. Of the long course of human existence, something like 40% of men have reproduced, while 80% of women manage to reproduce ( http://tierneylab.blogs.nytimes.com/2007/08/20/is-there-anyt... ). Thus for a woman to reproduce she basically has to not take risks, and not screw up. But the proper reproductive strategy for a man is to take major risks so he can be the top dog with 5 wives, rather than the median man with no wives. The result is that men are by nature far greater risk takers. Startups require a lot of risk and a lot of sacrifice, it's not at all surprising to me that it is so male dominated. All the most risky activities, throughout the entire course of human history have been male dominated.
And also, being good at programming is not just about having the smart genes. It's also about liking programming. I'm a nerd, I like spending most of time dealing with abstract logical concepts rather than with people. Most women are not like this. We have several women on our team in QA and product management. They have undergraduate degrees in engineering from top schools. But they don't program. Why? Because they don't like it. Whenever I talk with women about career plans, and mention programming, the usual response is something like, "meh". On the flip side, I'm pretty sure I would hate working in PR.
Men and woman are different. And you know what? That's OK. I don't see what this obsession is with putting all of society in a blender until every job as a perfect distribution of every demographic component. Celebrate and embraces differences!
Isn't it maybe a bit of a stretch to go from the hypothetical reproduction strategies of our stone-age ancestors to the gender balance of startup founders?
A much simpler and more informative approach would be to compare the gender ratios of startups in less techy/programming fields. If what you say is correct, the predominance of men should be more or less equal in these.
It's kind of embarrassing to me that whenever these issues come up on forums like this, people immediately start getting into amateur evolutionary psychology. This is basically code for "I like things the way they are; end of discussion."
And a lifetime of observations tell me that men in general are greater risk takers and more competitive. I cannot prove my thesis with 100% rigor, but it's the best explanation of the facts that I've got.
It's kind of embarrassing to me that whenever these issues come up on forums like this, people immediately start getting into amateur evolutionary psychology.
There's never going to be perfect data on either side that proves any of this. We'll never know for sure how much of the male/female divide is genetic or environmental. Neither sociology nor psychology is a Popperian science. So all we can do is combine knowledge of what science might apply, personal observations, stories from others, personal experimentation, readings from histories, etc, and make a judgement. In a word, we use Phronesis - http://en.wikipedia.org/wiki/Phronesis
I said that the inference to reproduction strategies was hypothetical, not the numbers.
As you say, there simply isn't any good evidence available for any position in this domain. For this reason, I don't see the point in making these highly speculative connections between reproduction strategies and startup gender balance. It only serves to cloud the issue by offering a pseuoscientific defense of the status quo (and by going off on a hell of a tangent).
The whole thing is a complete distraction from considering the issue in a practical way. If we actually got significant numbers of women in these positions, we could find out whether or not they were inherently unsuited to them.
Now, is it safe to assume a roughly equal standard deviation for both curves?
No.
Hyde, J. S., Lindberg, S. M., Linn, M. C., Ellis, A. B., & Williams, C. C. (2008). Gender Similarities
Characterize Math Performance. Science, 321(5888), 494-495.
I don't have access to the underlying study - the WSJ summary doesn't contain the actual data. Would you be willing to share it?
My claim is not that the SD's are the same, but that they are roughly the same. In other words, I don't think the differences in SD are large enough to affect the underlying argument. If the data proves otherwise, I'm eager to learn why.
Basic point is that SD for men in math ability alone is 1.11-1.21x as large as for women.
If you break out your table of normal distributions, you'll see this difference is amplified at the top of the distribution (i.e., for very smart people).
And this addresses only base mathematical ability. It completely ignores risk aversion, persistence and other factors which go into startup formation. A factor of 2 here and a factor if 3 there add up quite quickly.
I simply can't imagine what relevant factor would cause a difference in the width of these curves. I know that, in general, men have a higher standard deviation for most activities and attributes than women, but these effects are minor. So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? If so, I'd actually be curious to learn more about it.
My argument here is simply the best reasoning I can muster given the data that I'm aware of. You are welcome to poke holes in the reasoning or present alternate data. I'm eager to learn more in either case.
"So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? "
What underlying argument? You are presenting a political position unsubstantiated by any facts. Arguing about the distribution of standard deviations about two imaginary curves is like arguing about something in Alice in Wonderland. (Note that i didn't make an argument either way about the std dev in my comment).
Speaking against political correctness will probably get me downvoted, but here goes anyway,
From your blog post, I gather you think "diversity" of genders/races makes better dev teams. All I am saying is "prove it". With data, not hypotheses.
I have yet to see any real evidence for this, whether in terms of sustained commercial success or even in a statistical sense.
Show us these supposed benefits from race/gender diversity as applied to software development and startups. Is that too much to ask?
So because you can't imagine it, it must not exist? Even though you can't imagine how the standard deviations could differ, the evidence of a difference is extremely strong.
And a 10% higher standard deviation is much less minor than you think if you are looking at an ability cutoff. If you have two populations of equal size with equal medians, one of which has a 10% higher standard deviation, then about 99.5% of the top 1% across both groups belongs to the group with the higher standard deviation. It is that extreme because the standard bell curve has tails that drop off very rapidly, so one curve has pretty much ended while the other curve is still going on.
By what kind of a priori logic does the author presume that any given meritocratic filter will happen to have an equal pass rate along arbitrarily chosen orthogonal characteristics such as race and gender?
For any given test for any kind of ability, I would be shocked if the population selected happened to perfectly fit the hiring-brochure rainbow. It's just not very likely. It is very possible that a more meritocratic selection procedure results in less diversity.
In fact, that's precisely what happened when California passed a referendum in the late 90s to forbid considering race in admissions to the University of California. The process became less biased and the results became less diverse. My apologies to the author's preconceived ideas.
>"Demographic diversity is an indicator. It’s a reasonable inference that a group that is homogeneous in appearance was probably chosen by a biased selector. Even if men have an innate advantage at software development, the gap would have to be massive in order to explain why startup after startup has an all-male team."
Innate biological gender differences are not the only cause of differential average programming ability between gender populations. Sociological factors matter too.
While some of these sociological forces may be unjust and we might want to address them, that does not change the fact that by the time they reach adulthood the population of qualified programmers has many more males than females. Thus, a lack of diversity can emerge from a just meritocratic process.
The thinking here is just sloppy.