I literally can’t remember the last time I saw a public discussion of discrimination in tech where someone didn’t assert that discrimination is impossible because of market forces. Here’s a quote from Marc Andreessen that sums up a common view1.
Let’s launch right into it. I think the critique that Silicon Valley companies are deliberately, systematically discriminatory is incorrect, and there are two reasons to believe that that’s the case. … No. 2, our companies are desperate for talent. Desperate. Our companies are dying for talent. They’re like lying on the beach gasping because they can’t get enough talented people in for these jobs. The motivation to go find talent wherever it is unbelievably high.
Marc Andreessen’s point is that the market is too competitive for discrimination to exist. But VC funded startups aren’t the first companies in the world to face a competitive hiring market. Consider the market for PhD economists from, say, 1958 to 1987. Alan Greenspan had this to say about how that market looked to his firm, Townsend-Greenspan.
Townsend-Greenspan was unusual for an economics firm in that the men worked for the women (we had about twenty-five employees in all). My hiring of women economists was not motivated by women’s liberation. It just made great business sense. I valued men and women equally, and found that because other employers did not, good women economists were less expensive than men. Hiring women … gave Townsend-Greenspan higher-quality work for the same money …
Not only did competition not end discrimination, there was enough discrimination that the act of not discriminating provided a significant competitive advantage for Townsend-Greenspan. And this is in finance, which is generally more cutthroat than tech. And not just any part of finance, but one where it’s PhD economists hiring other PhD economists. This is one of the industries where the people doing the hiring are the most likely to be familiar with both the theoretical models and the empirical research showing that discrimination opens up market opportunities by suppressing wages of some groups. But even that wasn’t enough to equalize wages between men and women when Greenspan took over Townsend-Greenspan in 1958 and it still wasn’t enough when Greenspan left to become chairman of the Fed in 1987. That’s the thing about discrimination. When it’s part of a deep-seated belief, it’s hard for people to tell that they’re discriminating.
It’s true that markets have impose a cost on firms that discriminate. In the long run, that can put pressure on firms that discriminate, but as we’ve seen, that timescale can easily be longer than half a century. As the saying goes, in the long run, we are all dead.
This isn’t unique to finance. You can see this in a lot of fields. The only problem is that it’s hard to separate out the effect of discrimination from confounding variables because it’s hard to get good data on employee performance v. compensation over time. Luckily, there’s one set of fields where that data is available: sports.
In baseball, Gwartney and Haworth (1974) found that teams that discriminated less against non-white players in the decade following de-segregation performed better. Studies of later decades using “classical” productivity metrics mostly found that salaries equalize. However, Swartz (2014), using newer and more accurate metrics for productivity, found that Latino players are significantly underpaid for their productivity level. Compensation isn’t the only way to discriminate – Jibou (1988) found that black players had higher exit rates from baseball after controlling for age and performance. This should sound familiar to anyone who’s wondered about exit rates in tech fields.
This slow effect of the market isn’t limited to baseball, and in fact it’s worse in other sports. A review article by Kahn (1991) notes that in basketball, the most recent studies (up to the date of the review) found an 11%-25% salary penalty for black players as well as a higher exit rate. Kahn also noted multiple studies showing discrimination against French-Canadians in hockey2.
In every field where there’s enough data to see if there might be discrimination, we find discrimination. Even in fields that are competitive by definition, like professional sports.
Studies on discrimination aren’t limited to empirical studies and data mining. There have been experiments showing discrimination at every level, from initial resume screening to phone screening to job offers to salary negotiation to workplace promotions3. And those studies are mostly in fields where there’s something resembling gender parity.
In fields where discrimination is weak enough that there’s gender parity or racial parity in entrance rates, we can see steadily decreasing levels of discrimination over the last two generations. Discrimination hasn’t been eliminated, but it’s much reduced.
And then we have computer science. The disparity in entrance rates is about what it was for medicine, law, and the physical sciences in the 70s. As it happens, the excuses for the gender disparity are the exact same excuses that were trotted out in the 70s to justify why women didn’t want to go into or couldn’t handle technical fields like medicine, economics, finance, and biology.
One argument that’s commonly made is that women are inherently less interested in the “harder” sciences, so you’d expect more women to go into biology or medicine than programming. There are two major reasons I don’t find that line of reasoning to be convincing. First, more people go into fields like math and chemical engineering than go into programming. I think it’s silly to rank sciences by how “hard science” they are, but if you ask people to rank these things, most people will put math above programming and if they know what’s involved in a chemical engineering degree, I think they’ll also put chemical engineering above programming. Second, if you look at other countries, they have wildly different proportions of people who study computer science for reasons that seem to mostly be cultural. There is, of course, genetic variation between countries, but I think that few people would argue that, between two countries, one where roughly 50% of people who study programming are women and one where roughly 20% of people who study programming are women, the difference is due to some kind of inherent or genetic bias. Given that we do see all of this variation, I don’t see any reason to think that the U.S. reflect the “true” rate that women want to study programming and that countries where (proportionally) many more women want to study programming have rates that are distorted from the “true” rate by cultural biases.
Putting aside theoretical arguments, I wonder how it is that I’ve had such a different lived experience than Andreessen. To him, these excuses must sound fresh and that complaints from women and minorities must not ring true. But to me, it’s just the opposite.
Just the other day, I was talking to John, a friend of mine who’s a solid programmer. It took him two years to find a job, which is shocking in today’s job market for someone my age, but sadly normal for someone like him, who’s twice my age.
You might wonder if it’s something about John besides his age, but when a Google coworker and I mock interviewed him he did fine. I did the standard interview training at Google and I interviewed for Google, and when I compare him to that bar, I’d say that his getting hired at Google would pretty much be a coin flip. Yes on a good day; no on a bad day. And when he interviewed at Google, he didn’t get an offer, but he passed the phone screen and after the on-site they strongly suggested that he apply again in a year, which is a good sign. That John managed to get so far in the process is a direct result of the fact that Google has a number of processes in place to fight both conscious and unconscious bias. Most places don’t. Most places wouldn’t even talk to John.
And even at Google, which makes an attempt to remove bias, the processes in place fail as often as not. It happens all the time. When I referred Mary to Google, she got rejected in the recruiter phone screen as not being technical enough and I saw William face increasing levels of ire from a manager because of a medical problem, which eventually caused him to quit.
William’s situation was obviously beyond the pale, but you might wonder if Mary really wasn’t technical enough. Well, Mary is one of the most impressive engineers I’ve ever met in any field. People mean different things when they say that, so let me provide a frame of reference: the other folks who fall into that category for me include an IBM Fellow, the person that IBM Fellow called the best engineer at IBM, a Math Olympiad medalist who’s now a professor at CMU, a former distinguished engineer at Sun who’s now one of Google’s top hardware engineers, who also co-designed a distributed key-value store in between hardware projects, and a few other similar folks.
So anyway, Mary gets on the phone with a Google recruiter. The recruiter makes some comments about how Mary has a degree in math and not CS, and might not be technical enough, and questions Mary’s programming experience: was it “algorithms” or “just coding”? It goes downhill from there.
Google has plenty of engineers without a CS degree, people with degrees in history, music, and the arts, and lots of engineers without any degree at all, not even a high school diploma. But somehow a math degree plus my internal referral mentioning that this was one of the best engineers I’ve ever seen resulted in the decision that Mary wasn’t technical enough.
You might say that, like the example with John, this is some kind of a fluke. Maybe. But from what I’ve seen, if Mary were a man and not a woman, the odds of a fluke would have been a whole lot lower.
This dynamic isn’t just limited to hiring. I notice it every time I read the comments on one of Anna’s blog post (or a post by most other female tech bloggers). As often as not, someone will question Anna’s technical chops. It’s not even that they find a “well, actually” in the current post (although that sometimes happens); it’s usually that they dig up some post from six months ago which, according to them, wasn’t technical enough.
I’m no more technical than Anna, but I have literally never had that happen to me. I’ve seen it happen to men, but only those who are extremely high profile (among the top N most well-known tech bloggers, like Steve Yegge or Jeff Atwood), or who are pushing an agenda that’s often condescended to (like dynamic languages). But it regularly happens to moderately well-known female bloggers like Anna.
Differential treatment of women and minorities isn’t limited to hiring and blogging. I’ve lost track of the number of times a woman has offhandedly mentioned to me that some guy assumed she was a recruiter, a front-end dev, a wife, a girlfriend, or a UX consultant. It happens everywhere. At conferences. At parties full of devs. At work. Everywhere. Not only has that never happened to me, the opposite regularly happens to me – if I’m hanging out with physics or math grad students, people assume I’m a fellow grad student.
When people bring up the market in discussions like these, they make it sound like it’s a force of nature. It’s not. It’s just a word that describes the collective actions of people under some circumstances. Mary’s situation didn’t automatically get fixed because it’s a free market. Mary’s rejection by the recruiter got undone when I complained to my engineering director, who put me in touch with an HR director who patiently listened to the story and overturned the decision4. The market is just humans. It’s humans all the way down.
We can fix this, if we stop assuming the market will fix it for us, and fix things ourselves.
Thanks to Leah Hanson, Kelley Eskridge, Lindsey Kuper, Nathan Kurz, Scott Feeney, Katerina Barone-Adesi, and Patrick Roberts for feedback on this post, and to Julia Evans for encouraging me to post this when I was on the fence about writing this up publicly.
Note that all names in this post are aliases, taken from a list common names in the U.S. as of 1880.
I won’t get into them here because they’re widely circulated and people who don’t want to see a bias will literally read the opposite result into studies they look at. There was a great invited talk at FSE this year, where the speaker noted that if you show a bunch of programmers random data, they will interpret that data as supporting their prior beliefs. Turns out, the more biased people are, the more objective they think they are, too.
This effect seems to be how people interpret Becker’s seminal work on discrimination as saying that markets are incompatible with discrimination. But what it really says is that markets impose a cost on discrimination, and that under certain market conditions, taste-based discrimination on average doesn’t mean there’s discrimination at the margin. In discussions among the broader tech community, I have never seen anyone make a case that we meet the conditions under which average taste-based discrimination doesn’t imply marginal taste-based discrimination. Nor have I ever seen people make the case that we only have taste-based discrimination or that we also meet the conditions for not having other forms of discrimination.[return]