Throwaway of a googler here.
These results could be explained by not promoting women as fast as men, because every year it is expected you will get a CoL and performance based raise if you exceeed expectations for your role, regardless if you exceed enough to convince a promo committee of a level change.
So this could mean that women stay at lower job titles longer.
It could also mean that they sign into the company with higher base salaries. I have seen that maybe happen once in my experience for the same experience level but I have no way of correcting that for their interview performance or how well they negotiated.
Basically I don’t think this data point is super useful on its own.
The company’s annual analysis only compares employees in the same job category, so the results do not reflect race or gender differences in hiring and promotion.
That was the thrust of Damore's claim and he got fired because it would imply that some Googlers didn't earn their positions which Google interpreted as "creating a hostile workplace environment". Apparently having dodgy hiring/compensation practices doesn't create a hostile workplace environment, but calling attention to them does. In any case, good on Google for taking steps to remedy things; I hope they continue to examine and correct their practices regardless of who are the beneficiaries of the status quo.
I find it unlikely that Google, specifically, has a supply shortage in any possible niche all the way down to "software developers who are also unicycling clowns".
Their diversity quotas are presumably expressed relatively, not absolutely. In other words, they don't say "we need 10000 'diverse' engineers", they say "we want our engineering organization to be 60% 'diverse'" (I'm obviously making up numbers here). Ergo, the shortage in question isn't an absolute number of 'diverse' engineers, it's a shortage of proportions in the pipeline. And they can't do too much to stray from the proportion because they're constrained on one hand by hiring laws (there is only so much you can do to avoid hiring white/asian guys without overtly violating the law) and on the other hand by a fiscal need to hire more engineers at a sane price point.
Note that 'diverse' is quoted here because the whole corporate diversity enterprise is utterly superficial or it wouldn't categorize individuals as 'diverse' or 'not diverse' and it wouldn't choose the most superficial criteria (race, gender, etc as opposed to viewpoint, perspective, or cultural diversity) to make those assessments.
You're just saying that the h1b minimum is the same as the domestic minimum. I'd be interested in seeing how actual salaries are actually being paid out. h1b is supposed to only be used when no domestic workers can be found; many tech companies are using it to increase the pool of available workers at a given price point.
Well of course it's abused but I would argue that if H1Bs are paid less, they should raise the rates of H1Bs and discourage their use as cheap labor. We shouldn't be adjusting for it in the data here.
Google doesn't have an "h1b salary", true. My position is that h1b's tend to lean towards the lower end of a published range for a role, creating a salary marketplace that benefits the employer (which is an abuse of the h1b system; it's to be used to fill roles where domestic talent isn't available, not where it's too expensive)
Aren't we both doing the same thing, only coming from different ideological perspectives?
I'd love you to prove me wrong. If Google has domestic talent available at $150k, and can pay $110k to an h1b, they are supposed to hire the domestic talent. That's how the system works (or rather, it does if a company doesn't abuse it). I'd love to find out they are instead having to pay the h1b $175k because no domestic talent is available. If Google is doing this, and is the statistical anomaly, then good on them. However, all the evidence I've seen (anecdotally and in writing) suggest otherwise (though at an industry-wide level, not at the Google level)
I know many, many H1Bs around Silicon Valley. Many are making more than me, a "domestic talent". My base salary is over 200k/year. I know one making $300k base at one company and $600k base salary at another. I know many others that are making the same as me, and the only ones that make less are the less talented ones which is right.
Seems like you're conflating Google with other companies. Even if the aggregate data says underpaying does happen on average across the H1-B program, it doesn't necessarily mean Google underpays their H1-B employees.
You can't just use a statistical average to make a conclusion about only one of the data points. That's stats 101.
I understand this makes a great headline, but it might be because they don't promote women as quickly as men, leaving the women 'overpaid' compared to the other men at their level.
A better comparison might be to trace the trajectories of men and women who started out in similar jobs and at similar pay levels, and the see how many women appear overpaid.
That would be useful information, but it would also need some way to judge how biased (or not) the promotions are.
There's also another likely contribution to higher pay for women: a shorter supply of them, and Google bidding to hire/keep them versus other companies.
This is less than $1,000 per employee that was adjusted. Given the salaries at Google, this is almost certainly a <1% difference and the pay bump is pretty much negligible. It's less than the cost of living adjustment that they probably get every year.
You're correct that the magnitude is insignificant - but it's still important because the study's conclusion contradicted the expectation that men are paid significantly more.
I suppose you have a point. I have believed (known?) for a while that this was simply not true so since the magnitude of this difference is what most would call not statistically significant, I treated it as worthless. Though, I imagine those who were still of the mindset that men were still paid much more would find this much more interesting.
Did the analysis in the NYT report that TechCrunch is cribbing from here account for Simpson's Paradox?
Google employs more men than women. Even if a higher percentage of women are underpaid, we could see a situation where more men get raises in an equalization.
This is not how they did their calculations. It's described here [1].
They were not comparing individuals against some set average and making sure each employee was then brought up to that average and then saying e.g. x% of men and y% of women were bumped up. What they did was compare groups against one another and ensure that there was cross-group consistency. E.g. - the regression formula to determine a man's compensation based on various inputs (such as e.g. performance ratings) should yield a mostly identical result for a women of the same inputs.
And then they contrasted groups against various groups. They found that if you were a women, you were paid disproportionately more than other employees. This is why in their announcement they speak of groups instead of individuals, "[T]he 2018 analysis flagged one particularly large job code (Level 4 Software Engineer) for adjustments. Within this job code, men were flagged for adjustments because they received less discretionary funds than women."
Remember the goal is not to have every individual be identical, but to ensure that there is no specific group bias. Any given man/woman/etc may be paid more than another employee for reasons that cannot be easily quantified, but when the entire group of all e.g. men or all women are disproportionately paid more than other groups of the same quantifiables then you either have inadequate quantifiables (e.g. - you are missing why certain people are paid more), or there is simply no good reason and you have a bias. In this case, they decided it was a bias.
So this could mean that women stay at lower job titles longer.
It could also mean that they sign into the company with higher base salaries. I have seen that maybe happen once in my experience for the same experience level but I have no way of correcting that for their interview performance or how well they negotiated.
Basically I don’t think this data point is super useful on its own.