They are matching the diversity of the schools' recruiting efforts. So yes, you could say they are (perhaps) unbiased vs their input, but their input has problems. In other words, their selection of schools to recruit from means they are outsourcing the distribution of your their candidate pool, and the GP poster is not happy with that aspect of the result.
If it doesn't match your society or customer base, and you think those things are important (I do), then it's bad.
(The reasons why I think it's bad are complex, so I won't belabor them here.)
I'm not disagreeing that the situation is bad. I just don't see how adding bias to your recruitment is a solution to the problem. If we had a bunch of women who couldn't get engineering jobs I'd understand, but we seem to have the opposite situation where everyone is competing for a limited resource. If you're proportionate to the pool, you're probably doing well.
> I just don't see how adding bias to your recruitment is a solution to the problem
The goal isn't to add bias. It's to seek out the source of a bias we think might exist and then remove it.
For example, if we find 5% of top-tier schools' CS grads are women while 20% of mid-tier schools' CS grads are, and find similar quality in both pools (keep in mind, when you interview graduates you get to select for quality), then it makes sense to recruit from both pools. By excluding the latter pool, even if not for discriminatory reasons, we're implicitly biasing our inputs.
> we find 5% of top-tier schools' CS grads are women while 20% of mid-tier schools' CS grads are
I agree with the sentiment, but isn't the situation exactly the opposite of that? Top-tier schools work to balance gender distributions, and many of them succeed well above industry averages. Other schools spend less effort, and get predictably thinner results.
Compare, say, RPI and MIT. (I know RPI is uncommonly skewed, but I wanted to compare tech schools to one another. Otherwise specific stats are swamped by the ratio of tech to non-tech majors.)
- RPI, overall: 32% female
- MIT, overall: 46% female
- RPI, computer science: 16% female
- MIT, computer science, 32% female
As I understood the stats, a company that pulls students at the same rate as their candidate schools will see worse gender diversity as they broaden their pool. (Among CS programs at 4 year US colleges; outside that I don't know.)
MIT's computer science program is 31.7% women. MIT's school of engineering is 39.87% women. I think even comparing tech school to tech school, non-tech majors have a significant impact. MIT especially has a surprising number of non tech majors.
Rather than try to counteract the bias with an opposing one on the input pool, the answer can be to try to find new less-biased input pools to draw from.
We have to be careful about using the phrase "less biased" because bias goes both directions. In fact almost all universities go out of their way to get women into their tech programs. I think what you're talking about is trying to find a candidate pool that is more biased towards minorities than average, which is fine, but let's use the correct language.
Generally speaking I do see the lack of diversity in tech as a problem and I applaud people making an effort to rectifying that. Unless the universities the OP was pulling from are really out of synch with the industry I have a hard time understanding why being proportionate is a problem. In fact, it seems like the right place to want to be.
> We have to be careful about using the phrase "less biased
Fine--we want more women. Research, and personal experience, shows gender-mixed teams outperform [1][2][3][4]. This is a business, not HR, decision.
We also want more cultural diversity. I don't like pitching in Latin America or the Middle East with zero cultural context on the home team. If you're a B2C company, you probably want socioeconomic diversity on your product teams (at the very least).
Randomly picking the last paper you cite for example, we learn that Spanish companies with more women on the board have better financial performance. Seems like an open and shut case! But then buried deep in the paper, we also learn that the Spanish government has passed lots of sexually discriminatory laws that financially advantage companies that achieve arbitrary gender quotas, most notably, only companies with a certain level of female presence on boards can bid for public contracts.
The Spanish economy isn't the best. Big chunks of the Spanish economy is propped up by government spending originating in ECB quantitative easing and it's been that way for a while. If companies with male-only boards are being discriminated against in public procurement contracts, that by itself is sufficient to explain part of the difference.
That's a common problem with papers that argue more women on boards = more financial performance. Invariably they are studying countries that artificially tip the playing field towards such companies already, making it difficult to disentangle the different factors.
If it's a business decision to hire more females, then who are we to demand top tech companies expand their college search range?
Could it not be a business decision to minimize risk?
(I, for one, am in favor of getting rid of all diversity quotas. Companies should be allowed to hire 100% black LGBT disabled women as much as they should be allowed to hire majority males out of elite universities.)
If you want to bring more women into your firm because you think they have some biological or sociological advantages over men in some areas then that's fine, do so. But look at the majority of posts in this thread, they're all about reducing gender bias. These other posts don't make sense in the context of an employer whose payroll is already proportional to the pipeline.
You're saying the opposite, that increasing gender bias will provide you with some economic benefit. At least what you're saying is rational and consistent.
"Bias" measures deviance from a parameter. Without agreement on the population being measured and the parameter in question, the term--and quibbling over it--is meaningless. You are defining the population as something close to "people who possess the relevant skills," i.e. the pipeline. Others are using "people who possess the relevant skills but aren't in the pipeline" or "people who could possess the relevant skills if factors out of their control were accounted for". Others, still, disagree with how you're measuring "possess the relevant skills". This is all reasonable disagreement.
In universities that don't consider race or gender in admissions (notably the UC system and Caltech), the minority rate tanks (except for Asian-American rates, which double).
It is matching society, however: CollegeBoard testing statistics reveal that in high school girls take AP Computer Science much less frequently than boys - at proportions fairly close to 1:3.
Given that anyone is free to take an AP exam provided they can spend $100, the only plausible explanation is that fewer girls choose to pursue computer science as a major.
Alternatively, you could suggest that there is nationwide collusion between CollegeBoard and amongst computer science teachers to limit girls from taking the APCS test - and that no one has found out or exposed it.
What I mean by that is there exist no gender barriers to take an AP exam - that fewer women are studying CS in college or working a technical role is not a result of a biased system, but of an individual decision that comes down to fewer girls DECIDING to learn computer science.
> If it doesn't match your society or customer base, and you think those things are important (I do), then it's bad.
Mismatch between company demographics and society/customer demographics is common in all industries. It is not a unique phenomenon. Before asking what we can do to change the discrepancy at tech companies, we must first ask why this is a problem in the first place when tens of thousands of companies in different industries have the same problem.
If it doesn't match your society or customer base, and you think those things are important (I do), then it's bad.
(The reasons why I think it's bad are complex, so I won't belabor them here.)