[TeMPOraL]

Wish I could upvote twice. Very important insight, and one that is constantly missed in gender discussions. One shouldn't expect any change downstream to suddenly cancel out the effects of something high upstream. BTW. the huge skew at college level seems to be a result of a skew in as early as high-school, or maybe even earlier. [0] has an interesting discussion on the topic.

[0] - http://slatestarcodex.com/2015/01/24/perceptions-of-required...

[PeterisP]

You can get a reversal effect by self-selection.

Let's assume a simplified model of reality that has exactly two biased filters - an "upstream" bias that causes the pool of qualified candidates to contain only (as an assumed example) 10% women, and a "downstream" bias that causes qualified women candidates to disproportionally not get chosen or get worse offers, that results in a majority of companies hiring only (again, assumed example) 5% women.

If your company gets publicly known for hiring fairly, avoiding the second filter, then it may actually result in an effect that would "cancel out the effects of something high upstream" - simply because qualified women candidates may preferentially choose to apply at your organization, and your pool of candidates may contain significantly more qualified women developers than the national average, and thus also the people you hire would contain more qualified women developers than the national average.

A company with a reputation "if you're of group X, you'll hate it here" can have a perfectly fair hiring process, but still won't get much of group X simply because they will avoid that organization. An organization like Ku klux klan doesn't really need to do racial discrimination when hiring as most black people simply won't apply.

A fair hiring process would result in a proportion of women employees that generally matches the proportion of qualified women applicants, but the proportion of women among qualified applicants may vary significantly between different companies.

[sokoloff]

It doesn't even need that feedback-to-candidates loop to work.

Imagine that the female population of developers is 10%, that they exhibit in every way a performance distribution equal to males in job and interviewing performance, but that every company except yours is half as likely to hire a female candidate as the straight odds would suggest.

The candidate pool as experienced by all companies would consist of more than 10% females (as they would need to apply to twice as many places on average) and the average quality of the female candidate may well be higher than the average male candidate because of the adverse selection at play. (Qualified female candidates are being preferentially passed up in favor of inferior male candidates, leaving the residual female candidate pool more talent-rich than the male pool.)

[TeMPOraL]

Ok, I haven't considered that. That could indeed explain the results. Thanks.

[pron]

It is not missed at all. But the reality of discrimination at a young age is no excuse for not fighting it in software companies, where women constantly report hostile working conditions.

The article you linked to is quite good, but it ignores the very real decline over the past few decades[1] in women participation in software relative to other professions.

[1]: http://www.npr.org/blogs/money/2014/10/21/357629765/when-wom...

[lmm]

So these two things don't seem to fit together. The linked article seems to provide pretty solid evidence that women are just as successful as men when you control by numerical SAT score. But some[1] women do indeed report hostile working conditions.

How can this be? Is it:

* Working conditions are hostile to women, but this doesn't actually affect anything in terms of income, number of women in the industry and so on?

* More cynical variation: any woman who makes it to graduation in a STEM field has already taken a lot of hostility; those who hostility can affect are driven out earlier?

* Women and men actually experience equal amounts of hostility, and simply interpret it differently?

* Women are better at their jobs (or somehow have it easier) in a way that doesn't show up in numerical SAT scores, and this effect is exactly equal and opposite to that of a hostile work environment?

* Something else I haven't thought of?

[1] though by no means all, I remember lorettahe's post here a year or so ago

[belorn]

It seems that any work environment with a homogeneous group over 90% has reports of hostile working conditions for minorities. Reading reports about women entering a workplace dominated by men, men entering a a work place dominated by women, or blacks entering a predominated white work place, I constantly hear the same kind of abuse.

"By being the only Y in the work place, people expect me to represent the whole Y group just because I am Y."

"Because I am Y, everyone assume {common fear about group Y} about me".

"People think something is wrong with me because I applied to a X dominated work place and not one of the Y dominated ones."

It seems to me that the hostile working environment is the result of human nature when confronted with a minority. When the number of women in the industry increases, then the hostile working conditions will likely go away as quickly it initial started.

[pron]

Probably, but I'd like to point out that this subject has been studied for the past forty years or so, so we can do better than conjecture. While it is true that minorities often feel left out in homogenous groups, there is a big difference between cases where that minority is simply a numeric difference and those where that minority has less power[1] in society. We now know that the discussion of sexism and racism is not about numbers and differences but about certain groups having more power than others, and that causes some very specific behaviors.

[1]: http://en.wikipedia.org/wiki/Power_(social_and_political)

[pron]

I don't think you can combine different statistics of such high dimensionality like that. Unless women math GRE scores are in decline since the eighties -- and they don't correlate with success in science or medicine -- they don't explain the decline in participation in this industry alone. Also, the correlation of women participation and GRE math scores can remain just as strong regardless of the actual participation rates: the high correlation does not explain sex differences (as in this illustration http://upload.wikimedia.org/wikipedia/commons/6/6b/Heritabil...)

There is also a danger in looking at statistics about human behavior that only take a snapshot in time, because the data itself changes all the time. For example, it is very possible that ever since women participation started to decline, there have been few role models for women, less desire to participate, and therefore less desire to excel in math.

It is as impossible to study social dynamics from a statistical snapshot as it is to study planetary motion from a still photograph of the sky.

[lmm]

> the correlation of women participation and GRE math scores can remain just as strong regardless of the actual participation rates: the high correlation does not explain sex differences (as in this illustration http://upload.wikimedia.org/wikipedia/commons/6/6b/Heritabil...)

I don't understand what you're claiming. The differences in numeric GRE scores absolutely do explain the post-graduation sex differences. Are you claiming that low numeric GRE scores and low workplace success might have some common cause? Sure, but that cause would necessarily be pre-graduation, meaning that's the place to tackle it, and efforts to e.g. make workplaces less hostile aren't going to make any difference.

[pron]

I am saying two things. One, that the differences highlighted in the article do not predict the observed women participation rates in SV startups (i.e. the correlation of GRE scores could stay the same, but the participation baseline still increase 10%). In fact, the numeric difference he mentions (19% men, 6% women passing the cutoff) is a 3x difference, while the participation difference is more than 6x. So at best, it can explain less than half the effect.

Two, and this is the more important point, I absolutely claim that low numeric GRE scores and low workplace success might have some common cause, as I explained in my description of the feedback. That cause is not pre-graduation -- its effect is. Low participation in the industry causes fewer women to be drawn to the field, to be less interested in math etc.

I just want to point out what it is that I don't claim: I don't say that there is no significant innate difference in math abilities between men and women. Maybe there is and maybe there isn't. But its existence -- if it exists -- cannot nearly account for the huge gender gap we see in SV, doubly so because participation rates have been dropping since the eighties.

[lmm]

> That cause is not pre-graduation -- its effect is. Low participation in the industry causes fewer women to be drawn to the field, to be less interested in math etc.

If it's just that kind of feedback loop, how would the fall have started? Mathematics was once almost exclusively a men's game and early computer science fell in with that; the kids of the '80s would have had more female role models in the computer industry (and certainly as you say more in biology or physics) than those of the '70s or '60s, so why would they have been less interested in numerate fields?

I'm not saying that the feedback loop you describe can't amplify things, but I think the underlying cause has to lie outside it.

I've been thinking more about your decline-since-the-80s point. The narrative I've heard most often is that '90s compsci classes contained women with lower GRE-type scores than men (either through explicit positive discrimination/affirmative action, or because there were plenty of applicants who passed the admission criteria and a roughly-even split were accepted), who went on to do less well in industry (in line with their GRE-type scores), and the crash of the early '00s prompted an adjustment to more natural levels. But that leaves a lot of unexplained questions.

Maybe Yvain's page is right about academia, but those results don't extend to SV where there is more outright discrimination? Maybe it's about men being more able or willing than women to move to SV? Maybe SV's standards are stricter than the GRE cutoff and the difference at the end of the bell curve is even starker?

I still feel like we must be missing something. The simplistic explanation of "it's all innate ability" doesn't fit, but neither does "it's all discrimination/hostility", nor even "it's a 50/50 split between the two". Something remains to be explained here.

[TeMPOraL]

Of course it's not an excuse. But you should not expect to get results that defy statistics.

[pron]

Defy what statistics? What statistics explain the constant decline in the last few decades? Are women getting dumber?

[TeMPOraL]

The ones that say, e.g. that if you have 10 men educated in software engineering per one woman and you change your hiring practices to be less discriminating, you shouldn't suddenly expect to have 50/50 gender balance among tech workers in your company. You should see that 10/1 proportion reflected in your staff, if you hire based on merit alone.

[pron]

I don't think so, because the rates are not immutable, and you're discounting causation going the other way. If the women who are hired are treated well (which would require their employers to undergo some training), the rate of women attaining software education is likely to rise. "Downstream" likely has a strong effect on "upstream".

[TeMPOraL]

I agree, but the issue is with a post where someone apparently claims to be getting such results right now.

[Gurkenmaster]

I don't think so. If they are treated well they won't complain and you will never hear about them.