There is no evidence that Google "lowers the bar," and this is a common misconception about diversity-focused hiring practices. The idea isn't to lower the bar to hire more candidates for the mere sake of their ethic or gender diversity. The idea is to ensure that the hiring manager at least looks at some qualified, diverse candidates in his or her process.

The "lowering the bar" argument assumes there are not enough women or people of color in the population at large who are qualified for any given job. Google's approach probably assumes there are, and puts a burden on the hiring manager and recruiter to find and consider those candidates.

I don't work at Google and am wholly unfamiliar with their internal hiring practices. But in order to prove the author's specific "lowering the bar" claim, we'd need to see data not just on hiring numbers, but on success of the hired cohorts over time. (I'm willing to guess that Google does not try to hire unqualified candidates, period.)

"Lowering the bar" is a dangerous claim that offends the dignity and qualifications of every diverse employee at the company or ever hired by the company. A claim that bold would demand extraordinary evidence, which the author does not convincingly supply, and which I am taking a wild guess does not exist.

> The "lowering the bar" argument assumes there are not enough women or people of color in the population at large who are qualified for any given job.

No, the argument is that there are not enough women or people of color in the population at large who are qualified for any given job _in the quantities desired by Google and every other tech company_.

They are trying to create a workforce with roughly proportional amounts of women, men, blacks, etc. when men are awarded ~80% of CS degrees and blacks are awarded, what, 2% or so? You can't move the needle in employee demographics when that's the pool you're working with unless you fudge your system a bit. Hell, given how universities already do this[1], why wouldn't Google?

[1]: http://www.latimes.com/local/california/la-me-adv-asian-race...

> Lee's next slide shows three columns of numbers from a Princeton University study that tried to measure how race and ethnicity affect admissions by using SAT scores as a benchmark. It uses the term “bonus” to describe how many extra SAT points an applicant's race is worth. She points to the first column.

African Americans received a “bonus” of 230 points, Lee says.

She points to the second column.

“Hispanics received a bonus of 185 points.”

The last column draws gasps.

Asian Americans, Lee says, are penalized by 50 points — in other words, they had to do that much better to win admission.

How does a study about affirmative action policies in college admissions offer any specific evidence about Google's policies?

If universities can't increase minority enrollment without lowering the bar, what makes you think Google can? I would argue that university affirmative action is the inspiration for how Google and other tech companies are juicing their numbers.

Again, I'm not taking any issue with your assumptions, but assumptions are not direct or even circumstantial evidence. That's all I'm trying to say.

Your argument is legitimate. We don't know the exact mechanisms by which Google enacts their affirmative action hiring policy. We don't know if it's through quotas or scores or anything else. An affirmative action hiring policy, though, is built to hire more of a minority in some way or another. When the pool of workers of that minority is quite low(like in tech), I don't see how an organization can have such a diverse workforce without at some point choosing a weaker candidate that happens to be diverse.

If this isn't the case and Google does actually hire people based on their skill and not skin color or gender, then Google needs to emphasize that, or else the majority of people will look upon women and such as "diversity hires" and build a disdain for them, thus hurting the cause.

This assumes that talent and hiring is considered as a thresholded binary model, where you are either "good enough" or "not good enough" to work for the company, and the difference between two people who are "good enough" isn't important.

Of course, because of cultural and historical prejudices and biases, the pool of candidates that even apply to google is biased towards white males. Everything possible should be done to attract applications or seek out potential candidates from minority populations. But without suggesting any biological differences, it is still likely that because of cultural and historical prejudice, the distribution of talent among candidates will be skewed toward white males. Again, not because of any biological difference, but because they are more likely to have been encouraged or gone to rich universities or had parents who were engineers.

If we have a pool of 100 people and we want to hire the best five, then the bar is set by ranking the people by talent and qualification. If we want to hire everyone who is qualified, then wherever we set the bar, it is still likely that the population above the bar is skewed. So if we want to achieve a rebalancing or quota of employees from a certain sub-population, then it certainly does involve lowering the bar. We can argue over whether that is a good thing to want, or a fair thing, but I don't think you can logically argue that the bar isn't being lowered if you want to achieve a quota for a certain subpopulation. If you are suggesting that the population above the bar is not skewed, then you are suggesting that the terrible cultural stereotypes, biases and prejudices against women and minorities in technical fields has had no effect on their ability to achieve talent and qualification above the level of the bar, which I find implausible.

Yeah that's generally how it works at most of the majors. People at Google are evaluated by an independent panel and the decision to hire or not is independent of other candidates. So you really do get this kind of "good enough" or "not good enough" binary.

I believe the idea behind diversity initiatives is to ensure that the starting pool of 100 people is representative of larger demographics. Women make up ~50% of Carnegie Mellon's undergraduate computer science department (for instance). It's not like there's a dearth of qualified women.

That maybe the theory, but it often isn't how it works in practice. Unfortunately people who claims this publicly get fired. I don't know the numbers, but it would be interesting to see if the acceptance rate of a google job interview depends on gender or race.

I have a feeling that if race truly played a preferential role in hiring engineers, we would see more than 1% of tech roles being filled by Black engineers at Google. ;)

https://www.google.com/diversity/

Without knowing the distribution of the applicants, we can't say. Maybe 0.1% of all applicants are black, we don't know. In the same way, just saying 1% of tech roles are filled by black engineers doesn't imply that google's hiring practices are discriminatory.

Right. Which is why many female engineers do not want the appearance of preferential treatment in the hiring process.

I don't think Google ever lowers the bar for general hires, but they do have internships that are specifically targeted toward "underrepresented" people that are significantly easier to get into than regular internships. It's much easier to receive a return offer than a regular offer, so Google is effectively lowering the bar with that internship program

Apparently Google doesn't think that it has lowered the bar for women and minority candidates. So the memo's author has effectively raised a "Have you stopped beating your mother?" begging-type question. Should people get fired for poor rhetorical reasoning? Probably not, but this one happened to imply the inferiority of certain groups of employees.

Apparently Google fired the person. Apparently Google can't say that it has lowered the bar, as it will be PR disaster.

"Google doesn't think it has lowered the bar" is implied, not apparent.

You only said "probably not", so you are also implying it is probably true. Why didn't you just say "absolutely not"?

That's not how probably works. GP is only implying it's potentially true, presumably because they aren't sure they've fully considered every angle of the question and there may be weird edge cases.

Let's not Purity Test people for showing some humbleness in their beliefs. We could probably use more of that these days.

Or it's an implicit admission that the qualifications they look at are bused and they don't know how to fix that.

There's an underlying assumption in these discussions that implementing a meritocracy is easy and examining every candidate without regard to their race and sex is the default state of affairs. In this view, any changes you make to the process is necessarily de-optimizing for merit.

But what is "merit"? This isn't a field where you can quantify it. If you were hiring people to lift heavy objects or something, you could test them all and hire the ones with the best numbers and be confident you got the best people for the job. There's no way to implement something like that for the jobs Google has.

So we approximate. We look at degrees and open source projects and do in-person interviews. We pretend that this results in a purely objective evaluation of the candidate's ability to do the job, free of any bias... not because there's any reason to think that's true, but because we really want it to be.

Techies often bemoan whiteboard interviews for being unfair: the skill being tested isn't really the skill you'd use on the job, and it discriminates against people who get nervous when put on the spot, or whatever. It's not outrageous to think that maybe some parts of the interview process, without ever intending to, discriminate against women and minorities in a similar way.

That's an argument for digging into the inner workings of the hiring process more than an argument for looking at aggregate effect. If you're just looking at aggregate group statistics about candidate pools and hiring decisions, how would you ever know the difference between your hiring screen preferring a group and the candidate pool genuinely having more strong candidates in that group?

It seems more than a little quixotic if fixing every known flaw in the hiring process isn't enough - that the process must get a certain hiring outcome in order to be "fair".

And if they do dig into the inner workings of the hiring process and the results are still not representative of the general population? Do you just assume you've somehow managed to eliminate all bias and the results must reflect the population? Or do you assume that you haven't managed to 100% debug this highly difficult process?

It is very difficult to tell the difference between biased hiring and actual skew in the population. But when it's a trait that has no obvious connection to the job, it seems best to assume the process is biased, unless really good evidence exists to demonstrate a connection.

For example, let's say your interview process produces hires whose heights are substantially above average. If you're hiring for a basketball team then this would make perfect sense. If you're hiring Java programmers, it's highly likely that your interview process is screwed up. It's possible that the population of good Java programmers is unusually tall because of some biological factor, but this idea needs some major justification before you use it to make decisions.

Obviously, the first thing to do is to examine your interview process and eliminate any source of height bias you can find. But what if you do that and you're still hiring an abnormally large number of tall people? You could take this as sufficient evidence that tall people are better at this, but that's putting an awful lot of confidence in your interview program. It's more likely that you just haven't found all the sources of tall bias. If that's the case, then pushing for more short hires will improve the overall quality of your hires because you won't be artificially excluding good short ones.

>It's possible that the population of good Java programmers is unusually tall because of some biological factor, but this idea needs some major justification before you use it to make decisions.

But "we're somehow secretly and unknowingly discriminating in our hiring process" doesn't need such justification before using it to make decisions?

Correct, because bias is prevalent and almost impossible to avoid.

For example: does any part of your interview involve a subjective evaluation? If so, do any of the evaluators know the candidates' names? If so, congratulations, your process is probably biased! http://www.politifact.com/punditfact/statements/2015/mar/15/...

That's literally all it takes: some human judgment and some way to guess at the applicant's race. The discrimination can happen subconsciously. You and I would probably fall victim to it despite our best intentions.

Since the manifesto we're discussing is mostly about gender, it's worth mentioning that the same thing happens there: http://m.pnas.org/content/109/41/16474.abstract

Nobody knows how to screen job applicants for pure merit without bias. If it's a choice between "our process is imperfect just like everybody else's" and "we stumbled upon the perfect unbiased screening process and all bias in the output is due to inherent variation in the population," bet on the first one every time. It's about a million times more likely.

> Does Google's hiring practices "effectively lower the bar for 'diversity' candidates"?

He uses the phrase "effectively lower the bar" to mean something very different from what almost all of us think of as lowering the bar. The full quote is that one of Google's "discriminatory practices" is "Hiring practices which can effectively lower the bar for 'diversity' candidates by decreasing the false negative rate."

That is, the bar isn't lowered, they're just trying to fail to hire qualified candidates (from certain groups) less than they currently do.

Unless we have a reason to believe that false negatives are strongly correlated with the axis on which "the bar" is placed (and I don't think we do; false negatives are generally going to come from asking an out-of-left-field question, or the interviewee having a bad day, or something, which are pretty random), this doesn't lower the bar, "effectively" or otherwise.

So essentially, passing Google's hiring bar is a lottery, and Google is giving out more tickets to diversity candidates?

Google has a conceptual "bar", and sure, in theory there's a way to map every candidate onto that bar. Passing the bar isn't a lottery (it's just a deterministic mapping), but actually getting an offer is. Having gone through the interview process and received an offer recently, I'd certainly say it is.

I don't know that I'd agree with the second part; one thing you can reasonably conclude is correlated with false negatives is diversity (there are studies that people form hire/no-hire opinions within seconds of meeting a candidate, and studies that people have immediate biases when meeting people based on race/gender; there's also the simple fact that's put to good use in small companies that hiring people like yourselves, although mostly in ideological bias etc., is a good means for team cohesion). I assume Google has done more research on this than I have. Handing out "more tickets" may just be a way to cancel out the fact that so-called diversity candidates are getting worse tickets.

However, even if they were simply giving out more tickets to diversity candidates, the average level of qualification of these candidates doesn't change. You just have more of them. Maybe that's a problem because you're discriminating in their favor, but it certainly does not "effectively lower the bar" or impugn the qualifications of this over-represented group. They're all qualified. They're all on the right side of the "bar". There's no alleged increase in the false positive rate.

The fact that he chose to interpret this as "effectively lower[ing] the bar" is not just insulting to his colleagues and unprofessional, but also a sign that he doesn't understand math enough to pass a well-tuned Google interview and got one of the rare false-positive lottery tickets.

>Unless we have a reason to believe that false negatives are strongly correlated with the axis on which "the bar" is placed (and I don't think we do; false negatives are generally going to come from asking an out-of-left-field question, or the interviewee having a bad day, or something, which are pretty random), this doesn't lower the bar, "effectively" or otherwise.

I don't think that works. If the process is optimised for accuracy, then decreasing the false negative rate should increase the the false positive rate, right? Intuitively that's how it would work for most classification tasks. If you can decrease the false negative rate without increasing the false positive rate then you're improving accuracy. The bar must be correlated with the false negative rate.

Also, I'd mention that, even if, say, you have a noisy (presumably cheap) test that you give to everyone, and then you give an absolutely perfect (presumably expensive) test to minorities that fail the first test... It's probably the case that, of the people who failed the first test due to noise but would pass the second test, most of them are just barely above the "passing" bar. Because it seems reasonable to guess that people way above the bar have a smaller chance of failing the noisy test than people just barely above the bar. The policy would then lower the average ability level for hired minorities (though it wouldn't bring it below the "bar").

How large this effect is depends on just how noisy the first test is and on how widely above the bar people's abilities range.