Social Justice For The Highly-Demanding-Of-Rigor

My last two posts have led to a lot of anti-feminist activists getting linked to my blog, so this would be a more hilarious time than usual to write the next post in my series of arguments against Reactionary politics – about why fighting racism and sexism is necessary and important.

The Reactionary argument, as I understand it, is twofold.

First, that social justice advocates irresponsibly take some undesirable outcome in minority groups, like poverty, and then assume it is the result of racism or sexism without considering other possible explanations.

Second, that a disproportionate amount of time and energy is spent worrying about this, in a way that can only be explained through wasteful signaling cascades.

My counterargument is that although the first argument is true a depressingly large amount of the time, some people do more rigorous work and get the same result – that poor outcomes for minority groups are caused in large part by racism and sexism. And second, that these poor outcomes for minority groups are a major problem even by objective quantifiable standards.

Controlled Experiments On Prejudice

The most fun experiments on prejudice are Implicit Association Tests, which test people’s reaction times in linking together different concepts. If these concepts are socially important (for example, the concepts “white person”, “black person”, “good”, and “evil”) it can test how closely two different concepts are linked. The best way to get a feel for this is to take one yourself.

88% of white Americans (and 48% of black Americans!) show an implicit racial preference for whites on this test. How does that translate into the real world?

Some of the most interesting controlled experiments are detailed in an early ’90s review article in the Journal of Black Political Economy. A consortium of interested parties such as the Fair Employment Committee teamed up with recent university graduates. They laboriously paired white and black graduates by similar attractiveness, well-spokenness, age, gender, and qualifications (in some cases, the qualifications were faked to be as similar as possible), then sent them off job-hunting to the same companies.

In these sorts of experiments, 48% of white testers and 40% of black testers received interviews, a small and in fact nonsignificant difference. However, 47% of interviewed whites were offered jobs, compared to only 11% of interviewed blacks – a gigantic difference. Multiplying these two numbers together, we find that 23% of whites and 4% of blacks involved in the experiment got jobs – a difference of almost 6x. The whites also got a few other minor advantages – very slightly higher wages and slightly more likelihood of being informed of other open positions at the company.

Another good review article is in the Annals of the American Academy of Political and Social Sciences. It lists a few similar studies. In one such study, researchers, instead of training real applicants, send off fake resumes with extremely white-sounding or extremely black-sounding names; they find employers respond to the white-sounding names about 50% more often. But it also has some studies of in-person interview similar to the ones above. These studies, which are from the mid-2000s rather than the early 1990s, feature white:black success ratios of anywhere from 1.5x to 5x.

Along with labor discrimination, it’s harder for minorities to buy things. For example, when trying to buy a car, black men were asked to pay on average $1100 more than attribute-paired white men. Interestingly enough, black car salesmen, and black owned car dealerships, displayed this pattern to exactly the same degree as white-owned institutions.

The situation is roughtly similar in housing. In an experiment where researchers responded to Craigslist notices advertising apartments in Toronto, using names of various ethnicities, they found that Caucasian experimenters confused relative risk with odds ratios 100% of the t…ahem, sorry, they found black people experienced housing discrimination 5% of the time and Muslims 12% of the time, usually in the form of not receiving a response even when the white person was simultaneously invited to come on over. A similar study in Houston found an astronomical 80% discrimination rate for blacks, so either Houston is much worse than Toronto, someone’s not doing their studies properly, or I’m misinterpreting something.

Other experiments along the same lines include a cute little bus experiment in Sydney where someone got on bus, their travel card didn’t work, and they asked the driver to let them ride anyway. For whites (and Asians) it worked about 72% of the time; for Indians, about 50%, and for blacks, 36%. In a later survey, bus drivers (who were unaware the experiment was going on) claimed they would prefer to help black people over white people. Interestingly enough, although black bus drivers were a bit nicer to blacks than white bus drivers, they still let whites and Asians on more often.

We find much the same pattern with men and women. A famous study a few months ago found that faculty offered a female grad student a 12% lower salary than an identical male grad student (again interestingly, female faculty were more biased against female grad students than male faculty were).

A less perfect but more natural experiment is switching from an open application procedure where applicants’ genders are obvious to a blind procedure in which genders are unclear. If the percent of women hired increases (and perhaps if no similar increase is seen in competitors that don’t change procedure at the same time) this implies the institution was being unfairly biased before. When such a test was performed by the Journal of Behavioral Ecology starting in 2001, the percent of articles by female authors went up from about 29% to about 37%, about a 30% increase .[EDIT: This has since been found to be false] Symphony orchestras are another infamous example, and studies show that the switch from open to blind auditions explains between half and a third of the recent quintupling of the percent women in symphonies over the past thirty years.

What do these show and not show? They show that, even controlling for all other factors like different preferences, different negotiating strategies, different educational backgrounds, et cetera there is a large difference in the opportunities of minority and majority groups due solely to discrimination. This difference seems large enough to explain the proportion of the income gaps that people say it explains (usually around half of the gap for each minority group) and to give minorities large amounts of trouble throughout the rest of their lives.

One thing it does not show is that racism is just about straight white men being evil. Minorities seem just as willing to screw other minorities over and discriminate in favor of white men as the white men themselves are. A better model would be that ideas of certain races and genders being superior seem to percolate into people’s consciousnesses, regardless of what race those people themselves are, and shape their actions whether they mean for them to or not.

Economic Costs of Discrimination

A beautiful experiment by Gwartney and Haworth noticed that baseball formed a natural experiment about the costs of discrimination. During the post-Jackie-Robinson 1950s, some teams had integrated but others had not and remained white-only. G&H wondered whether this affected performance. They found that in fact the five teams quickest to integrate black players were five out of the six top performers in the league, and that every additional black player on a team resulted in an addition 3.75 wins. This was partially because black players outperformed whites on average, but also because there was more low-hanging fruit in the form of black talent which could be employed more cheaply.

What is true for baseball teams is probably also true for other companies, but harder to quantify. For such a popular field, I cannot for the life of me find any attempt to quantify the economic costs of racism. There seem to be some people in Australia working on it, but they have yet to publish any results. So let’s make some up (this, uh, ends the demanding-of-rigor part of this post).

One way to do this is to take people’s estimates of the purely-discriminatory pay gap for different groups – that is, how much less they earn than straight white men when all other factors anyone can think of (like education level, IQ, height, region of residence, whatever) are adjusted away. Then multiply this by the number of people in that group and their average wage, and we get part of the cost of racism per year.

One of the review articles above suggests the black pay gap is 15%; others suggest numbers around 10% for women. Asians and gays make a bit more than straight white men, and although Latinos make much less no one has bothered adjusting for confounders so I can’t include them.

Anyway, when I add all that up, I get $374 billion.

(one might argue that the companies these people work for gain this money as profit by paying employees less, so it all evens out. I don’t think that works. In at least some cases, the lower pay must be because they have lower-level jobs than their white male counterparts. But since we already agreed they have the same skills as their white male counterparts, this suggests their skills aren’t being used fully, which means the cost really is to the economy and not just to them. I have no idea whether this argument works in real life. Like I said, the highly-demanding-of-rigor probably should have stopped with the first half of this post)

Another claim is that companies lose $64 billion dollars to discrimination-related turnover yearly. The number is generated by taking the cost of replacing a lost employee with results of surveys about how many people leave due to discrimination or hostility at their former place of employment.

Suppose we arbitrarily and implausibly stop here because we’re tired. We’ve found costs of US racism equal to at least $438 billion per year.

That’s about the annual budget of the US military.

Note what it doesn’t include. It doesn’t include any non-monetary costs like people being unhappy. It doesn’t include the costs to the prison system of overprosecuting minorities. It doesn’t include the costs to the health system of minorities getting worse preventative health. It doesn’t include the amount the government spends fighting racism, or the amount people have to pay out in racism lawsuits. It doesn’t include people who are unemployed because of racism, because I took the data from the employment records. It doesn’t include any of the income gap due to racism anywhere other than at job – for example, racism that affects how much education people of different races end up with, or racism the person’s parent suffers that then screws up their families for several generations. It doesn’t even include Latinos because I couldn’t find any good numbers about them.

It seems very unlikely to me that the actual costs are less than $1 trillion/year in the US alone. But let’s stick with the $438 billion figure.

(another way to look at this is that these arbitrarily-stopped at costs of racism/sexism are about $2-3K per minority group member in the US, counting women as a “minority group”. This seems broadly reasonable, and is in fact still way less than the observed non-adjusted income gap)

What other things cost $438 billion dollars? According to the American Cancer Society, cancer costs $200 billion/year Heart disease costs $100 billion. Adding up all of the easy-to-calculate costs of 9/11 on this page, I get about $250 billion.

So in terms of purely economic, not-even-worrying-about-human-beings costs, the costs of racism and sexism that can be pretty plausibly attributed to discrimination alone are equivalent to about heart disease plus cancer plus half a 9/11 or so per year.

One good thing about the size of this number means that small successes in fighting racism and sexism are extremely valuable. For example, decreasing racism/sexism by 1% is a $4.4 billion gain per year to the economy, which is about equal to Facebook’s 2011 annual revenue.


Now none of this is meant to claim that the marginal blog on Tumblr complaining about the patriarchy has positive expected value or is anything other than a massive waste of everyone’s time.

But there are aspects of the social justice movement interested in testing what works and doing it.

As of yet, I don’t think most of them are aware of the pitfalls in claiming successful interventions – all of these “We found our intervention decreases expressions of prejudice on a seven point scale two weeks later!” things sound suspiciously like “Our drug increases ‘good cholesterol’ after three days, and we didn’t bother to check whether it actually prevents heart attacks but seriously how could it not?”

But we can’t blame them for their failure to be more rigorous than the hard sciences, and besides one day they might wise up.

With Implicit Association Tests, Ultimatum/Dictator games, and the like, I think there is a decent toolkit for people who want to wise up and seriously analyze anti-racism and anti-sexism strategies, and I bet when they are tested further some of the ones in that document will turn out to work longer-term. Maybe they could decrease racism by 1% a year and save us $4 billion or so.

The fact that racism in the sense of simple prejudice is a real problem that accounts for much of the disadvantage of minorities; that it has a huge negative effect even when you try to measure it objectively; and that it can be fought – seem to take some of the wind out of the Reactionary argument against social justice.

This entry was posted in Uncategorized and tagged . Bookmark the permalink.

93 Responses to Social Justice For The Highly-Demanding-Of-Rigor

  1. Eli says:

    Couldn’t employers just be hiring based on accurate stereotypes? If so, aren’t all the cited costs of discrimination really just costs of not having perfect information?

    Suppose it is true that black people just don’t care about money as much as white people. As an employer I probably don’t know this cognitively, but I have a sense, an impression based on my exposure to black and white people. Two resumes come in, or maybe two interviewees come in, one black and one white, and they give me all kinds of information I care about. But all that information is identical, but I still don’t know everything about them. Should it be just as likely that I hire the white one as the black one?

    I don’t think so, because I still have relevant information; one is white and one is black. That tells me something I care about. So maybe I’ll hire the black one because I can pay him less, or maybe I’ll hire the white one because his greed will give me more control over him, or maybe it cancels out. But the point is that it matters. Not discriminating based on relevant information is an economic cost.

    Now take a step back and recognize that races are different in many ways that matter to employers. You can adjust for confounders all you like, there is still risk being taken on because not all information is filled in. And so long as there are things I don’t know about candidates that race speaks to / correlates with, race will matter in my hiring decision. Adjusting for what employers do know doesn’t change that.

    Am I making an error?

  2. Not only that, but if you have prior knowledge of the distribution of ability among groups (hint: you do) and you think the resume is a noisy source of information about ability (hint: it is), then your posterior distribution for the two candidates won’t be the same, _even if you knew all the evidence cited on the resume was of the same strength for both candidates_ (which as you point out, it isn’t.)

  3. Rational says:

    You have two applicants with identical resumes, one with a white name, one with a black name. You also know that colleges grant preferential admission to black candidates with the same credentials, and often grant admission to black candidates who would not have been admitted if they were white. You know that other business also do this.

    So, you have two “identical” resumes, and you choose to hire the one with the white name. Was that because you’re racist, or because you’re making a rational decision that the two resumes are in fact not identical, and that the knowledge that a candidate is black, coupled with extrinsic knowledge about social institutions, leads one to conclude that there is more uncertainty regarding the black applicant?

    This would be social justice for the highly demanding of rigor.

  4. Valhar2000 says:

    I’ve taken the IAT and, frankly, I am not impressed by the methodology. I think the test I took measures my inability to adapt to a changing stimulus once I have identified a pattern, i.e. if the images they show require me to press the “e” key 5 times in a row and then suddenly they show me an image that requires I press “i” instead, I probably will press “e” again. This happens to me to me in many other situations that have absolutely no relation whatsoever to race.

    Does the IAT take this into account somehow?

    • Alexander Stanislaw says:

      Yes it accounts for that effect by randomizing which stimulus-word pairings are presented first.

  5. PatrickSMcNally says:

    “employers respond to the white-sounding names about 50% more often.”

    This is not as compelling as it may at first sound. How many applications did these employers receive? If an employer consistently faced a shortage of white male applicants along with an abundance of resumes from blacks or women applicants, then we should expect that an unbiased employer will respond to many of the latter applications. But if the job market is full of white males who are qualified and desperate to find work, then it is rational that a company will have to throw out many resumes without even looking at them. If the company has traditionally been made up of white male employees, then it makes perfect sense that this will show itself in the responses to job applications. There’s no valid reason for seriously attributing that to racism or sexism.

    In the 19th century it was common that European immigrants entering the northeast on the Atlantic shore would meet Anglo-Saxon bigotry. They dealt with it by moving to the south or west where northeast industry was moving to and they reaped rewards from that. After 1890 the end of the Western Frontier meant that colonial settler expansion was no longer an option. But instead the USA moved on to become the world’s greatest economic power after 1945. Since 1971 and the stagflation crisis the real growth has been stagnant. Industry has developed new techniques for putting people out of work and making the same items with fewer hires, even as competition from abroad has eroded the dominant position which the USA held in 1945.

    Under these conditions the older mode of integration which worked for Poles or Italians in the 19th century has proceeded at a much slower pace for black people, even though it clearly is occurring. People like Oprah Winfrey, Colin Powell, Wendy Williams, Herman Cain, Condoleezza Rice, Al Sharpton et al are testimony to the changes which have been ongoing for the last half-century. But the sustained economic stagnation which has set in signifies a much deeper problem.

    From the descriptions which you give, the tests which you’ve cited tend to blur the distinctions between racism or sexism in a proper sense versus the effects of economic drag. That does not shed as much light as it may at first appear to. Major social changes of a progressive kind have always required a vibrant economy in order to complete their success.

  6. Pingback: The Emerging Pro-Monarchy Movement: Neoreactionaries and the ‘Dark Enlightenment’ | pundit from another planet

  7. Pingback: Geeks for Monarchy: The Rise of the Neoreactionaries - go2pear

  8. Pingback: Geeks for Monarchy: The Rise of the Neoreactionaries | TechCrunch

  9. Pingback: Rational Bayesian Statistical Discrimination | ars libertatis

  10. Pingback: Fuck Your Theory – Ozy Frantz's Blog

  11. Pingback: Modern Social Justice is Nuisance | Bonum et Malum

  12. Pingback: Assorted Thoughts On The Definition of Racism – Ozy Frantz's Blog

  13. Michael Vassar says:

    Big numbers times small and difficult to calculate probabilities and magnitudes of effect. By default, we’re in Pascal’s Mugging territory here. That doesn’t mean that nothing can be done, but it does mean that a special high standard for intellectual integrity is necessary in order to claim to be doing any good in a situation like this, and that anyone who claims to be fighting these problems without rising to that special high standard of intellectual integrity is contributing to the problem, not fighting it, at the very least by creating noise that drowns out the signal of actual opportunities to do good.

    Call me when Givewell recognizes an efficient philanthropy opportunity in this department.

  14. Leonard says:

    I will second the comments of those above (Vladimir and Matt Simpson) advocating or at least mentioning rational discrimination.

    Matt gives an example of rationally discriminating against women. Let’s pump up the volume here: rational discrimination against blacks.

    You, Scott Alexander, know that blacks are much more criminal than whites. Across a wide variety of offenses, blacks are more criminal than whites by a factor of roughly 7. Of course in both cases it is small minority of people (mostly men — oh noes, sexism too!), who are the problem. But nonetheless, criminality has the unfortunate attribute that it can be extremely costly. Small risks of great harm must be considered by the rational man.

    I don’t know if you have some theory that explains the high observed rate of black criminality other than “(sputter) racism!”, but even allowing the use of “racism” as meaning “the invisible things whites do to make blacks act badly”, you know that such “racism” exists and is highly intractable. You cannot expect any given employer to be able to undo it, such that his black employees act exactly like his white employees.

    So: sitting before you are two resumes. One black, one white. You rationally compute your priors. Priors say that if you hire the white man, your chance of getting a criminal employee are about 1%. If you hire the black man, your chance is 7%. Which man do you hire?

    Not enough information! You might hire either one, depending on what employee criminality costs you! Perhaps you compute the cost to your company of a criminal employee as exactly $1 million. To an ideal employee (perhaps an East Asian), you are prepared to pay $30000/year. So, discounting for the criminality of the white man, you make him an offer of $20000/year. But discounting for the much higher criminality of the black man, you don’t even make him an offer, because your expected value from employing him is $-40000/year.

    Now maybe you act affirmatively, seeking out all sorts of information about your black applicant, and determine that his risk of being criminal is much lower than the average black person; perhaps it is just 2%. In this case, you offer $10000/year — assuming that fits within minimum wage. Oops, it doesn’t. Too bad.

    It is not just criminality that is a hidden attribute of job-seekers that must be estimated by the employer. Almost all relevant information is unknown. A resume is a pale shadow of a person. Other relevant attributes would be stuff like “conscientious”, “hard working”, and “won’t cause problems if I have to fire him”. In all of these cases, there is a stereotype of black underperformance. Such stereotypes are racism and as such unspeakably evil, of course, but this does not render them false. At least one is true: we know that fired white employees do not initiate highly costly race discrimination lawsuits.

    It seems to me that a simple model like I am suggesting here would explain most of the experiments you mention.

    Will you choose rationality and racism?

    • Army1987 says:

      “conscientious”, “hard working”

      Those two are probably largely screened off by stuff that does show on the resume.

  15. amuchmoreexotic says:

    I agree with the values of the social justice movement. The problem I have is that, as you touch on in your last section, 90% of the activity related to it is all about signalling what a good and just person you are, and often a lot of infighting between different groups who largely agree with each other but have found a way to construe something the other side said as Bad.

    Discussing actual interventions is often taboo. I remember a particular article on Feministing about how a male psychologist had showed that college rape could be reduced by an intervention involving showing men a video about male rape, which apparently increased their ability to empathise with women who’d been raped. The reaction of the commenters can be summed up as “THIS IS HOMOPHOBIC AND BAD”. This wasn’t just a suggestion that it would be good to measure whether the intervention increased homophobia, it was a total rejection of this researcher who was deemed to be a gay-hatin’ douche.

    I got the impression that reducing rape was less important than snarkily vilifying “rape-apologists” in a little corner of the internet, while the mainstream of society just goes on being rapey.

    • im says:

      Agreed, plus there is the contradiction of the Inclusiveness Wars and using ‘privilege’ as a cudgel. Love the goals, hate the community.

  16. Reactophile says:

    Any serious discussion of racism should be prefaced with the following disclaimer written in large bold letters:


    Please include such a disclaimer in previous posts. A failure to include such a statement amounts to a implicit accusation that modern whites have and continue to commit a staggering crime against humanity comparable to slavery or the Holocaust. If you don’t include such a disclaimer the discussion is just not going to go anywhere because we will know immediately that you are not arguing in good faith. We’re not going to talk to you, we’re going to start stockpiling weapons to protect ourselves from you when you try to launch your next pogrom.

    • Patrick (orthonormal) says:

      Did you actually read the post? Scott is arguing that your disclaimer is false. If you disagree, it behooves you to make a counterargument that addresses his post, not just a contrary assertion.

      • Erik says:

        Black people are less intelligent than white people [on average, in America, and other caveats] and have higher propensity for violence than white people and have lower future time orientation than white people. Stupidity, violence and short-sightedness contribute to underachievement. A context of affirmative action may also reverse the causality between black underachievement and white racism: if blacks are known to get credentials more easily because of quotas or the like, it is rational to prefer a white applicant to a black applicant with the same credentials. This is in addition to the factors already mentioned and commented on elsewhere in the thread, such as the increased risk of a lawsuit for firing a black employee.

        • Scott Alexander says:

          If I remember correctly, the studies included pairs of subjects without college degrees or any other affirmative-action-able credentials and found much the same pattern.

          We unfortunately have less information about whites/blacks than we do about men/women because it’s easier to put a fake female name on things than to put a fake black name on things (without confounding with the fact that the most obviously black names signal class as well as race). But we do know that reviewers will rate the same science article as less rigorous and less intelligent when it has a woman’s name on it, and are less likely to publish it in a journal.

          This seems like an example of discrimination that avoids even the few lingering possible problems you mentioned above (fireability, affirmative action). I grant that we can only prove it in the sexism case and not the racism case, but it would be very surprising if irrational discriminatory sexism were a real thing but irrational discriminatory racism wasn’t.

    • ozymandias42 says:

      Um. I’d say that the position of black people in America today is deplorable, but it’s not Holocaust-level evil. I find that claim rather bizarre, because observably African Americans are not currently the victims of a genocide.

      (I’m also pretty sure that there are tops five people who want to direct a pogrom against racists, and that’s just because you can find five people who want to direct a pogrom against anyone.)

      • gwern says:

        A quick consideration of life expectancies by race, annual prison rapes and murders, black overrepresentation in prison population, and integrating over a century certainly does suggest that the issue could reach Holocaust levels of evil…

    • amuchmoreexotic says:

      Yeah, people who argue that white racism is bad are exactly the type to launch pogroms, historically, aren’t they? Did you read and believe that “Liberal Fascism” book or something?

      • Scott Alexander says:

        Uh, to be fair, the history of communism has proved that “This privileged group is oppressing us, let’s kill them all” is actually a pretty successful way to organize a campaign of mass killing, especially when it’s true.

    • im says:

      Please tell me you aren’t actually afraid of a pogrom.

      • Reactophile says:

        I’m afraid of economic pogroms, and I categorically reject the idea that this is an irrational fear, because economic pogroms killed more people in the 20th century than the Holocaust.

    • Nick T says:

      This comment makes me think it would be valuable for someone to do a lot more thinking and writing about lines of retreat in the case that Standard Oppression Narratives turn out to be true.

    • Army1987 says:

      You might want to replace “not the cause” with something like “not the only, and probably not the main cause” — otherwise it’s way too strong a claim IMO.

  17. Matt Simpson says:

    Let’s consider two applicants for the same job. Both applicants are equally qualified. They each have degrees from top notch universities, several years of experience in their field, and a top notch record. No measurable differences in “ability to do the job.” However, based on some other credible information, the employer knows that applicant A has something like a 10% chance of needing to take a month or two off work (and this could happen repeatedly with some decreasing probability), and a 2% chance of leaving the workforce altogether. Applicant B on the other hand, has probabilities of about 5% and 1% respectively. Who is the employer more likely to hire? Who is the employer willing to pay more? The answer in both cases is clearly applicant B.

    Now my probabilities were completely made up, but I think the situation is pretty transparent: applicant A is female and applicant B is male. (The probabilities are made up, but are supposed to indicate something about what happens when they have children.) The point of this exercise is that gender and race potentially convey information beyond what you can learn from a resume or an in-person interview. Using this information is certainly “discrimination,” but I wouldn’t call it “racism” or “sexism” (as long as “racism” and “sexism” mean something evil).

    Some points about this:

    1) You really need to know *why* employers are discriminating in order to estimate the costs of discrimination. If employers are discriminating on the basis of their actual uncertainty about applicants, it’s misleading to say that “discrimination costs us $X” because then it seems like all we need to do is get employers to stop discriminating and we’ll save $X! But that’s not the case – you have to factor in the additional costs imposed by ignoring useful information. Now employers have to spend more resources hiring since the employees they hire quit more often in the hypothetical example above. I’m sure you can find more potential hidden costs if you think clearly about other cases of discrimination. In reality, it’s more accurate in this hypothetical situation to attribute the costs to uncertainty. If the employer knew with certainty that applicant A really was no different from applicant B, they wouldn’t discriminate – or if they could be compensated for the additional uncertainty (and possible cost) associated with applicant B, they wouldn’t discriminate. But the fact is, they don’t know and it’s (probably?) illegal for them to be compensated, so they do discriminate (to the extend they can get away with it under the law, of course).

    2) I’m not saying that the discrimination *isn’t* racism, or that the theory I outlined is correct. That’s an empirical question (that I don’t know the answer to). This is just the output of my model of reactionaries. In reality, it’s probably partially this “benign” discrimination and partially something more nefarious, but then the question is “how much?”

    3) This sort of effect might be plausibly called “structural racism/sexism,” but I’m unsure about that label because it seems to implicitly suggest, e.g., the employer’s beliefs in my example are incorrect. (That’s my understanding of structural X anyway, feel free to correct me if I’m wrong.)

    4) Another possible reactionary response is “yeah, your’re right, but while you’re worried about social justice, society is collapsing. PRIORITIES!” It’s basically the zombie apocalypse meme. I don’t know how to evaluate this claim.

    5) I really would like my paragraphs to have line breaks between the. HELP!

    • Matt Simpson says:

      So the preview of my comment didn’t show line breaks, but the actual comment does. Weird.

    • Nick T says:

      I’ve heard it claimed (don’t remember the source) that this sort of gender discrimination exists mostly because prospective employers aren’t allowed to ask “do you plan to take time off to have children?”. Oops.

  18. Andrew Hunter says:

    I wrote on arguments like the resume study last year. I don’t think they prove anywhere near what you think they do if you remember that a resume is a noisy data point and we have informative priors on the population.

  19. Emily says:

    Race is still predictive of cognitive ability to some degree even if you condition on education: check out the AFQT scores of NSLY97 respondents, and you’ll see that there are still meaningful differences in distributions by racial group, even within educational groups. Sex is still predictive of certain aspects of performance, even conditioning on your current education/job/etc. For instance, women are more likely to drop out science PhD programs and more likely to leave full-time work in many fields, so unless we think that women are getting admitted to those programs/fields with lower qualifications (which I don’t think is that likely), using the same indicators to predict these aspects of performance overpredicts female performance. So “identical” according to the stuff that any given study is measuring is not necessarily actually identical in terms of expected outcomes. Statistical discrimination on the basis of race or gender is, of course, totally illegal. But it’s not always irrational/bad for productivity.

    • Andrew Hunter says:

      > so unless we think that women are getting admitted to those
      > programs/fields with lower qualifications (which I don’t think is
      > that likely)

      Oh, they do. Affirmative action in the hard sciences goes well into grad school.

  20. Max says:

    I’ve been doing a lot more reading of nonfiction lately than I usually do, and without setting out to do so I realized I had ended up picking out bunch of books focusing on racial discrimination and history and such, and some of it has been pretty enlightening.

    I suspected that there might be some distinct historical reasons underlying the lower social and economic status of African Americans, relative to other historically discriminated-against groups, such as the Irish, and it seems that there are some pretty significant candidates for such, which I hadn’t been exposed to in my own prior historical education. A few specific historical elements seem likely to have kickstarted, or at least significantly exacerbated, the sort of cultural elements we see today which contribute to poor outcomes such as higher crime levels and lower education levels among African Americans.

    I’d sooner write a proper post on the topic than leave it buried way down in the comments on someone else’s blog, but I don’t have a blog of my own (nor do I particularly want one at the moment,) and I’m not sure that the subject would be that well received on Less Wrong.

    • g says:

      So, uh, were you going to share any of that enlightenment with the readers here, e.g. by mentioning those few specific historical elements?

  21. Federico says:

    Second, that a disproportionate amount of time and energy is spent worrying about this, in a way that can only be explained through wasteful signaling cascades.

    It can also be explained according to the model in which elites covet power and personal impact, and ordinary folk are susceptible to moralistic persuasion or coordination with high-status beliefs. Those employed by, representing or in service of the state can—when permitted to do so—usually maximise their psychic revenue through activity and expanded powers, rather than libertarian circumspection, whether or not this enhances the commonweal.

    Universal-suffrage democracy and a substrate of legitimate desire shape these activities in a particular way. It serves elite interests to have every unique, distinctive culture be undermined and subsumed into a homogeneous community in which the state is heavily involved.

    Democracy is ostensibly structured to allow periodic overthrow of elites and ideologies by new, diverse and independent social forces. Micromanagement of society, including its fonts of status and opinion, is the loophole by which an establishment can perpetuate itself under such conditions. Social inclusion and sensitisation is a suitable vector for the process.

    So the real reactionary view is that, much as non-elite progressives tend to be sincere, lust for power—and not empathy or any such virtue—is the main human drive to have shaped this ideology.

    First, that social justice advocates irresponsibly take some undesirable outcome in minority groups, like poverty, and then assume it is the result of racism or sexism without considering other possible explanations.

    Again it is necessary to distinguish between advocates such as yourself, and members of the elite. Elite belief (distinct from belief-in-belief) in their espoused ideas is dubious. Elizabeth Warren, for example, both advocates a massively increased minimum wage and pays her interns nothing.

    George Lucas, on the other hand, seems altogether too canny in his regard for low-income housing.

  22. Tom Hunt says:


    Firstly, I’m not entirely convinced of the applicability of the IATs to real-life results. To illustrate my misgivings, I’ll become rather personal: I do, in fact, become nervous on first seeing/meeting a black person. I strictly refuse to allow it to affect my interactions with them, and the effect disappears once I know any given black person, but it’s plausible that I broadcast this nervousness through some unconscious social cue, which could on the larger scale be considered part of “white privilege” or whatever. (I’m not certain what leads to this; it doesn’t seem to have any connection to actual fear for my personal safety. It probably has something to do with the fact that I was raised in an area where black people were very rare, and in all my various schools there might have been one per class year, if you were lucky.) However, even having taken this into account, when I took an IAT last year it told me that I had a “slight implicit preference” for black people. Take that as you will, but since I know it’s not actually true on the useful level it makes me inclined to doubt the applicability of IAT.

    Secondly, regarding employment statistics–I have no data about this, but I have heard that some companies are reluctant to hire black people, not out of any racism, but because if they then have to terminate that person’s employment for any reason (disciplinary problem, just not working out, whatever) it makes them a target for EEOC proceedings, which is friction a lot of businesses don’t need. This is of course an effect very difficult to quantify, but it seems plausible to me that it accounts for at least some of the observed disparity. (It would also, of course, generalize to any disadvantaged group under the aegis of the EEOC.)

  23. gwern says:

    > One of the review articles above suggests the black pay gap is 15%; others suggest numbers around 10% for women.

    Doesn’t the women pay gap largely go away once one controls for hours worked, danger of career, etc? If so, that casts serious doubt on your estimates of the costs: yes, the black pay gap or whatever could well go away after controlling for more underlying statistical generalities – since it just did for another group where the evidence seemed equally good. And this would quite satisfactorily explain why members of that group are ‘discriminating’ against fellow members.

    • I also believe that there’s no real ‘pay gap’ any more, but on the other hand it could well be that if I know I’m underpaid I’ll consequently work fewer hours (for my non-hourly position). It’s complicated.

    • Scott Alexander says:

      Most of it does (the naive unadjusted one is like 25-30%). The amount that doesn’t is still under dispute and is what I’m estimating at 10%.

      • JJJ says:

        Remember the God of the gaps? Well feminists are appealing to a sexism of the gaps:

        Economist: Once you control for enough variables, there is no pay gap between men and women.
        Feminist: Bullshit. My stats say that women are paid 25% of men.
        (1 year later)
        Economist: I’ve done some research and it turns out that women choose to work in industries that pay less. That explains 5% of the gap.
        Feminist: Ok, but there’s still a 20% gap.
        (1 year later)
        Economists: I’ve done more research. Women choose perks like flexible hours over more more pay. That explains another 5%
        Feminist: Whatever. There’s a still a 15% gap.
        (1 year later)
        Economist: I’ve done even more research. It turns out that women choose to leave the workforce earlier than men in order to have children. Women more eager to become a stay-at-home spouse than men. That explains another 5% of the gap.
        Feminist: There’s still a 10% gap! You can’t explain that, so it must be sexism!

        Every year we find more and more nonsexist explanations for why women are paid less than men. Every year the unexplained gap gets smaller and smaller. Is it reasonable to say that because we can’t yet explain this gap, this gap must be sexism? Or should we say that the gap is due to nonsexist factors we don’t understand yet, but will in the future?

        • Scott Alexander says:

          It’s reasonable to do controlled experiments that can ferret out exactly what percent of the gap is due to sexism. I’m saying those experiments have been done and the results are above.

          • gwern says:

            > It’s reasonable to do controlled experiments that can ferret out exactly what percent of the gap is due to sexism.

            Seems to me that by the descriptions, the experiments clearly do *not* eliminate the statistical bases and allow an inference of ‘sexism’.

            Example: suppose 5% of the pay gap is due to employers’ rationally calculating that women employees getting pregnant and dropping out causes substantial damages to their company via the health plan, reduced quality of work before dropping out, the disuption of replacing them and whatever skills or unique knowledge or relationships they have built up… The experimenters send in a large sample of male and female candidates with identical resumes and observe the offered salary disparity, because after all, the female candidates are indeed female and the male candidates are indeed male and it’s awful hard for a man to get pregnant. In what sense have you proven with your experiment that the gap is due to sexism rather than rational self-interest?

            You haven’t; you have merely demonstrated again that there is a gap, which no one denies. The debate is over the cause of the gap, which you have not randomized or manipulated and so you cannot claim to have disproven that causal relationship and provided evidence for the sexism relationship.

            You could only claim to have tested the pregnancy hypothesis if you had directly intervened on that, but that could be difficult (use aged women post-menopause? Stamp each resume with a note ‘by the way I am infertile and also I hate children so I would never hear my biological clock ticking and adopt children’?).

  24. Mary says:

    What experiment was that, with the names? The only one I’ve heard about — well, yes, statiscially the black names drew worse than the white ones. But the absolute worst was a white male name — Geoff. And the black and white names had a lot of overlap, so much so that a good number of people were arguing that the names were in fact being discriminated against for being, well, weird.

  25. Vladimir says:

    One thing it does not show is that racism is just about straight white men being evil. Minorities seem just as willing to screw other minorities over and discriminate in favor of white men as the white men themselves are. A better model would be that ideas of certain races and genders being superior seem to percolate into people’s consciousnesses, regardless of what race those people themselves are, and shape their actions whether they mean for them to or not.

    Another model, which you don’t even mention here, is that much, perhaps most, of that discrimination is in fact rational Bayesian statistical discrimination. If that is the case, then all these attempts to quantify the supposed “cost of discrimination” start from a completely wrong premise and end with nonsensical conclusions.

    Of course, the mainstream view is that statistical discrimination against certain protected categories is evil and unacceptable, period. This is in fact just a special case of the more general norm that Bayesian inference about people is evil and unacceptable when it involves probabilities conditioned on membership in these protected categories and the conclusion somehow reflects negatively on the subject.

    Therefore, in order to defend the mainstream perspective on discrimination, you have only two options:

    (1) Accept the full-blown mainstream anti-discrimination norm, and get into the odd position of a Rationalist(TM) who preaches that in many situations, Bayesian inference is a grievous and unpardonable sin.

    (2) Argue that somehow, by some happy contingency, it just happens that there are no (or very few) significant practical situations where a correct Bayesian inference of this problematic sort is possible. (But what are the chances that the universe just happens to be so conveniently aligned with the moral demands of a particular ideology? And moreover, note that as long as there are any such situations at all, you must either recant your Bayesianism or accept views about these particular situations that put you beyond the pale of the respectable opinion.)

    • Anonymous says:

      Unless knowing things like age, socioeconomic status, level of education, etc screens off race as a predictor, in which case the mainstream position is pretty reasonable because you’re basically never going to only know someone’s race. Even a simple email can give a lot of indications of level of education and so on.

      Even ignoring that, the mainstream position could be justified: it is entirely possible that an idealized Bayesian argument would show that an Asian candidate whom you know nothing about would be a better option than a black candidate who you know nothing about and it is still wrong to prefer the Asian candidate, because that preference does more harm than good.

      For a related example, I think most of us would concede that courts should not admit evidence obtained from torture even if that evidence is good Bayesian evidence. Just because something is Bayesian evidence it does not follow that it is always moral to allow it to affect your decision.

    • rumblestiltsken says:

      The problem is that there is a wealth of well-grounded evidence that most of the “statistical discrimination” inferences are wrong.

      Black people, women etc. are going to be just as effective in any job they are qualified for as the majority norm.

      In face, there is a reasonable amount of evidence that a diverse workplace is more productive and efficient.

  26. Federico says:

    One of the lacunae in the concept of social justice is that not all inequalities are equal. Businesses are also known to discriminate on the basis of physical attractiveness, height and age, but I hear very little about the economic or moral need to combat these attitudes.

    I would hesitate to use GDP as a measure of Americans’ physic revenue. Most of us prefer not to work 100 hours a week; likewise, a small businessman might conceivably trade utility for minor pecuniary gains when he neglects to hire certain ethnic groups or women.

    The bigger problem with these strident economic conclusions is that armed with this knowledge, the experimenters ought to be fabulously wealthy. Since their wisdom actually seems to go unused, one expects that all these discriminatory rules of thumb might for one reason or another, beyond the rather narrow purview of the studies, actually be profitable to landlords and businesses.

    Occam’s Razor also threatens the idea that all these people discriminate negatively against ethnic minorities like negroes and Muslims, but far less so or even positively towards ethnic minorities like Jews and Asians, just because they are reflexively bigoted in this very particular, complex way. It is more parsimonious to suppose that the former ethnic groups lack certain merits possessed by the latter ethnic groups.

    Finally, given that most Americans claim strongly to oppose ethnic discrimination, why are legal and coercive means necessary to redress the problem? Couldn’t progressive businesses, landlords and institutions voluntarily compensate or wield structural coercion (e.g. boycotts) to fight irrational discrimination?

  27. Sarah says:

    I would like someone to go back — really go back — and actually address the reactionaries.

    Why, for instance, do we want to not have a caste society?

    What is bad about bigotry? (Using only terms that an educated 19th century man would understand.)

    What are the disadvantages of a genetic aristocracy?

    What are the disadvantages of having strictly polarized gender roles in a society?

    Reactionaries, for the most part, get mileage out of the rhetorical shock value of simply proposing that rigid and unequal social roles, often based on sex, class, and race, would be better. Instead of being shocked, somebody should just address the claim directly.

    For instance, the US Confederates were, to my brief impression, not at all impressive thinkers, and it doesn’t seem that there were deep pro-slavery arguments. It’s mostly sophistry. Carlyle is interesting and should be addressed, but he *can* be addressed, he’s not some kind of thousand-year vampire. I haven’t really read the Nazis beyond excerpts, but it seems that even when you understand what they were thinking and live in their head a little (and it is very disturbing to realize that you could live in a Nazi’s head) it comes up empty. The way you can sympathize with Terence McKenna and try to steelman his stuff a little and almost see it if you squint, but no, actually it’s nuts.

    I’m not sure what you call “the belief that we ought to have fixed, unequal social roles for people, including sex roles and racial castes and aristocrats” but I think that’s the defining characteristic of the Reactionaries.

    And my honest view is not, really, Reactionary. I don’t have it all worked out yet, but it has something to do with recognizing people as people. That there is a thing in me which can reason, perceive, and communicate; there is also such a thing in you; and almost by symmetry or game theory or something, my little spark ought to be concerned with other people’s little sparks. To rigidify social roles to the point that I cannot treat someone in a different category as a person, that I’m always “saluting the uniform, not the man,” dealing with the caste, not the human inside it…is, as Mike would say, “a wrongness.”

    All men are not brothers. They’re distant cousins who fight all the time. It’s unreasonable to expect all men to be brothers. But I want to relate to people to the extent that they are my brother . Bump functions, not hard cutoffs (and long-tailed bump functions at that, because I don’t know everything.)

    • Berry says:

      I agree in full.

    • Oligopsony says:

      The Catholic reactionaries are quite intellectually formidable – in fact, as best I can tell, almost all the great reactionary thinkers had a Catholic or Jewish background (whether or not they cared about it,) make of that what you will. (Nietzsche was a rare exception.) Schmitt is a very formidable Nazi but admittedly doesn’t get you very far to Nazism itself – I am told the same is true of Heidegger, though I can’t understand him – and those who do get you as far as Nazism (like Rosenberg and Evola) are jokes.

      Third Positionists, Catholic reactionaries, and anarchists all have a sort of romantic James C. Scott critique of modernity floating around them, which obviously would be quite radical on its own, but I haven’t ever actually seen it without religion, race, or anarchy attached (although I have seen it with all three on occassion.) Hayek maybe? Nah, not remotely radical enough, and too market-friendly. There must be a reason this political space is unexplored.

      • Misha says:


      • Sarah says:

        If it involves reading and writing, you’ll get Jews, and, to a lesser extent, Catholics. It is what it is.

      • Federico says:

        Oligospony, in a spirit of friendly inquiry what do you understand by “reactionary”?

        I would call Hayek a conservative-libertarian. Skepticism towards democracy, of which Hayek speaks little, is essential to reaction. Most soi-disant reactionaries endorse Hayek’s economic beliefs, and in particular the Austrian business cycle theory.

        Does democracy work as described? Is it compatible with liberty? Does it facilitate Machiavellism? Has it corrupted our truth systems?

        “No, no, yes, yes” is the reactionary answer.

        • Federico says:


        • Mary says:

          In which case the meaning of “reaction” has drifted, and there is no particular reason why people would be current with the meaning. . . .

          Old meaning, Martin Luther King Jr. was reactionary. He wanted to bring back the Reconstruction, where Jim Crow laws were swatted down as fast as they could rise. (And it was that great Progressive, Woodrow Wilson, who ensured that that stopped.)

        • Federico says:

          Mary, that’s why I asked. I’m no expert on historical meanings of “reaction”, perhaps unlike Oligopsony.

          But since Scott has discussed the popular internet form quite a lot recently, it seems worth pointing out that most of these reactionaries agree with Hayek, as far as he goes.

          The only notion of anarchy that I would touch with a bargepole is David Friedman’s anarcho-capitalism, and then only due to an interest in polycentric law and political innovation. Nick Szabo points out that polycentric law is not synonymous with anarchy; also, anarcho-capitalism wouldn’t work.

          Friedman is in a sense Hayek-squared—he thinks that we should privatise everything. So the type of anarchy that might slightly appeal to neoreactionaries implies very pro-market views indeed.

          Internet-reactionaries have different ideas about race and religion to the mainstream, but are they the ones “attached” to these things, or just desirous of a legal system in which those not attached to one particular, self-consciously egalitarian and radically modern lifestyle can opt out?

          And I think the distinguishing belief of neoreaction, not shared with paleoconservatives or vanilla libertarians, is that progressivism is not only mistaken but that these mistakes are systematically manufactured by our incarnation of democracy.

        • im says:

          This is making me want more and more to describe my idea of ‘reconstruction’, which values being able to opt out of the self-consciously hypermodern lifestyle, but which doesn’t have all the racism.

    • komponisto says:

      Instead of being shocked … just address the claim directly.

      This should be a mantra.

      It really works, more than one would think it would.

  28. JJJ says:

    If women and minorities will do the same jobs for less money as whites and men, that implies that there is a lot of money to be saved on labor costs. You’re telling me that the greedy corporations are picking up a 374 billion dollar bill of the ground?

    Your baseball example only weakens your point. Exhibit A: Moneyball. Once team managers discovered that minority players could be recruited at a discount, the market corrected. And it corrected itself with no help from the government or social justice activists.

    Here’s an exercise I don’t have time to complete right now but might later. Find a big corporation. See how what it’s profit margin is. Then see how much it spends on labor costs. Estimate how much money it could save it replaced 10% of it’s privileged workforce with oppressed classes. Compare the savings to it’s profit margins. Now ask yourself if this corporation could compete with a less racist/sexist corporation that saved money on labor costs.

    • ozymandias42 says:

      …It’s illegal in the US to pay women less than men for the same job. So no. They can’t just hire women and pay them less, that’s *blatantly* illegal.

      • Vladimir says:

        Businesses would still have many plausibly deniable ways to tilt their hiring practices towards employing more women, while relying on these same alleged mechanisms that cause lower wages for women. Each marginal step in this direction would mean a marginal increase in profit.

        Even if this couldn’t be done in a plausibly deniable way in a big corporation with a thick HR bureaucracy, it wouldn’t be at all difficult for a small or medium-sized business.

        • Mary says:

          None of which would prevent their spending more than they have to on labor. The costs are indifferent to the character of those paying. If your employer is a miserable racist, sexist skinflint who would gladly fire every black and every woman if only he could hire men in their place — which he can’t, with the miserable salaries he offers, they go to his only mildly sexist and less stingy compeitors — he will still be saving on labor costs.

      • JJJ says:

        ???? I have no idea where you’re going with this Ozy. If businesses are unable to pay women less than men for the same job, then what on earth are feminists complaining about?

        • Ben L says:

          Because “same job” is very narrowly defined. When each employee often gets an individual offer, for example, there is a gap. You can also (as noted in the blog post) simply not hire women, if you are subtle enough. Lastly, yes, I *would* say corporations are just leaving money on the ground. Just not on purpose.

        • JJJ says:

          Ben L: Inasmuch as businesses discriminate against women, they are leaving free money on the table. If you hire a man for $10 when a woman would do the job for $9, you’re conning yourself out of a buck. Furthermore, even if most businesses were sexist, a business that put profits before sexism would be able to out-compete discriminating businesses and eventually drive them to bankruptcy. Evolve or die.

          Alternatively, inasmuch as businesses aren’t discriminating against women, feminists have no grounds for complaint.

        • Mary says:

          I add that historically weaseling around even legally required discrimination has occurred whenever it cost. The way Montgomery got the bus companies to enforce racial seating was to stop buses at random and arrest the driver if he had let anyone violate the law. Large portions of South Africa that were supposed to be white only often had a minority of whites, and since construction firms were supposed to be all white, they had a white front man whose job it was to deal with the government. And so on and so on.

        • ozymandias42 says:

          Okay. Let’s try this again.

          You have Job X which pays fifteen dollars an hour. Although Joe and Jane are equally qualified, you hire Joe. Jane goes to work Job Y for ten dollars an hour. Even though Jane would be willing to work Job X for twelve dollars an hour, you are not allowed to hire her at twelve dollars an hour while continuing to pay Joe fifteen dollars an hour, because that would be wage discrimination and get you sued. Repeat for a sufficiently high number of Joes and Janes and you have a gender wage gap.

          Obviously, this is not the ONLY factor involved in the gender wage gap. But even in this vastly oversimplified model “just hire women and pay them less!” doesn’t work as a solution.

        • JJJ says:

          Ozy: A company doesn’t have to pay women less than men in order to save money off labor costs. All it has to do is stop paying men more than women.

          You have Job X which pays fifteen dollars an hour. Although Joe and Jane are equally qualified, you hire Joe.

          If I’m the employer, then Job X pays whatever I want it to pay. If Joe and Jane are both equally qualified, the profit maximizing action is to give the job to whoever will do it for less. Now if there were a persistent wage gap in the industry, Jane would have fewer options and be willing to do the job for less. So I would hire Jane and Joe would stay unemployed.

        • JJJ says:


          Or let’s assume that I can’t pay less than $15/hr. (Maybe there’s a minimum wage or something.) In that case, I’m going to hire the most qualified person I can find willing to do Job X for $15/hr. And if, as feminists assume, women will work for less than equally qualified men, then we should expect the most qualified person willing to work at $15/hr to be a woman.
          Proof: Let’s take the most qualified woman willing to work for $15/hr. An equally qualified man would not be willing to work $15/hr, according to feminist assumptions. Therefore, all men willing to work for $15/hr must be less qualified than most qualified woman willing to work for $15/hr.

        • JJJ says:


          So if there was a pay gap, the cheapest person to do a job at a given level of quality will be a woman, and the most qualified person to do a job at a given price will be a woman. You can make more complicated models where businesses consider both the cost and the quality of applicants, but you’ll find that women are always the more profitable hire.
          So if there’s as long as there’s a pay gap, businesses will have a monetary incentive to hire women. This means women will get more job offers then men and get more bargaining power. So women will demand higher salaries, and the pay gap will get shrink. The cycle will repeat until the pay-gap disappears.
          In sum, a pay gap between men and women is unsustainable in a profit-driven economy.

        • Paul Torek says:

          Markets aren’t perfectly efficient. We do not have a purely profit-driven economy. These facts are as plain as day.

      • Eric Rall says:

        There are several strategies to take advantage of a discrimination-caused gender gap:

        1. If there’s a significant spread in salaries for the positions you’re hiring for, offer a salary that’s above the typical female salary for a the responsibilities, working conditions, and qualifications you’re looking for, but below the typical male salary. A disproportionate share of the pool of qualified applicants you get will be female and will be willing to work for less than you’d have to pay if you were discriminating against female applicants.

        2. If discrimination is a barrier to acquiring experience necessary to acquire and demonstrate the skills needed to qualify for senior positions, skew your recruiting towards overfilling junior positions and filling senior positions through internal promotion.

        3. If discrimination is an obstacle to getting the education needed to qualify for junior positions, start an apprenticeship and training program intended to substitute for the traditional education system.

        Three is tricky since it would work best if you could get applicants to sign traditional apprenticeship/indenture contracts where they have to work for you at below-market salaries for X years after graduation to compensate for you subsidizing their training until they become trained enough to be productuve, but in most fields, labor laws forbid this. The risk is that once you’ve trained your apprentices, they’ll get poached by other firms; this can be mitigated by focusing your training on firm-specific skills, or by counting on signalling issues and institutional inertia to lead to other firms undervaluing your apprenticeship graduates long enough for you to turn a profit.

      • Deiseach says:

        Ah, but women prefer part-time/flexi-time jobs because they still shoulder the greater part of the burden when it comes to looking after the house, kids, sick older relatives, etc. so that does make it easier to pay less since part-time work in general pays less. It also has a knock-on effect in bonuses, promotions and the like since if you’re not in the office forty hours a week interacting with the bosses and the rest of the staff the same way Joe is, Joe looks better when it comes to internal review and interview.

        And (at least in the Bad Old Days) it was possible for businesses to wriggle around equal pay requirements by saying that the woman was not doing the exact precise same job as a male colleague so she wasn’t entitled to the same level of pay, or give the male employee a job title that then included perks but not the female employee, or say it was on grounds of seniority etc.

        I was amused by the bias test, since it comes out that I appear to be racist and sexist – where white males are concerned, since they were at the bottom of my results at the end. Though I think that, as the test acknowledges, there are several factors which could influence that – for one, having to press keys with right finger/left finger and judging bias from that on speed of reaction time is a result that may be skewed by the likes of me, who still have trouble with right/left distinctions (seriously, when I was eight, the best way I could figure out which was my right hand was ‘it’s the one I bless myself with’ and to this day I often turn left when I mean to go right and vice versa). Plus, I would have liked more questions about what I thought of men (the ones about women in work/home were so obvious that I could have lied about them rather than give honest opinions, e.g. ‘yes I strongly agree women should go out to work’ – er, no I don’t, but I’m not going to say so because I know better than to express an unpopular opinion like that).

        So I think it’s an amusing test, but I wouldn’t take it as extremely valuable for judging bias – I don’t know what other tests of the kind are like?

    • Alex says:


      I suspect that I am misunderstanding you but it seems to me that your argument proves too much. It seems to prove that discrimination based on sex/race in the workforce is impossible in a free market.

      • gwern says:

        Why can’t it be done just by picking up the free lunches? In his autobiography, Alan Greenspan says he did exactly that: he hired underpriced women and made money from not being sexist. Similarly, I have read of a company or agency which does the same thing in highly sexist South Korea, and for other forms of discrimination, we have companies specializing in hiring discriminated-against people on the autism spectrum and making money off the difference.

        If you want to prove a market isn’t efficient, simply reliably make excess risk-adjused returns…

      • JJJ says:

        Not impossible, just expensive and unsustainable in the long run. But yeah, I’ll bite that bullet.

        Note that I am NOT claiming that women and minorities are paid the same as men and whites. Rather, there are hidden costs to hiring women and minorities (see Vladimir discussion of statistical discrimination) which means that employers cannot expect to get the same quality of work from an “equally qualified” minority employee.

        • Alex says:

          Thanks for clarifying.

          Would you mind expanding on the following?
          “which means that employers cannot expect to get the same quality of work from an “equally qualified” minority employee”

          Why not? It seems to me that statistical discrimination is useful when you are using one variable (race) as a proxy for another variable that you don’t know (for example having a science degree). However, this is not needed when hiring since you can get at these variables by looking at someone’s resume.

        • JJJ says:

          The resume doesn’t tell the whole story. It doesn’t tell you if the person will fit in with your workplace culture. It doesn’t tell if they have any friends involved in bad things. It doesn’t tell you if they money problems. Things like that.

        • Army1987 says:

          If the fact that people of the wrong race/gender won’t “fit in with your workplace culture” doesn’t count as racism/sexism (on the part of your other employees), I don’t know what does.

    • Scott Alexander says:

      If it happens, it’s possible.

      I think the controlled experiments show it happens.

      One possibility is that the CEOs of this company are aware of this problem, they would *like* to hire equally qualified minorities, their orders to their HR minions are “hire equally qualified minorities” but during interviews people subconsciously come up with negative assessments of a minority’s qualification and so think they’re doing the right thing by not hiring them. This would require active debiasing, and there’s enough fear around anything relating to race that I don’t think those debiasing programs would naturally spring up.

    • Fnord says:

      Your baseball example only weakens your point. Exhibit A: Moneyball. Once team managers discovered that minority players could be recruited at a discount, the market corrected. And it corrected itself with no help from the government or social justice activists.

      Yes, they eventually stopped, once the color barrier was already broken and, if what you say about Sabermetrics being important is true, with a measure of performance which is way more objective than is available to most companies.

      But did minorities who wanted to play baseball suddenly appear out of nowhere? On the contrary, baseball teams were leaving money on the ground for quite a long time.

    • Viliam Búr says:

      I find it completely credible that many corporations are crazy and harmful even to themselves. We don’t live in a perfectly fair market where the more efficient company would automatically win. Sometimes a company has a political leverage, and will get better options than its competitors. Even if it burns a part of this profit by internal stupidity, there is still a lot of money remaining.

      Another problem is that people often have to work in teams, and it is often difficult to measure an output of an individual. If you hire a member of an unpopular minority, the team’s output may decrease simply because some of the other team members will refuse to cooperate with this one. This could be fixed by creating a whole team of the minority members. But again, they would have to interact with the rest of the company.

      Also, I would expect a typical corporation to be greedy enough to desire saving every penny, but at the same time incompetent so it can only see the most straightforward methods, such as firing half of the people and reducing the wages of the remaining ones. Employing the most skilled members of an unpopular minority cheaply may be already too abstract for them. They would understand the “cheap” part, but they would have problem to tell a difference between a skilled and unskilled person.

      In a perfect world, Dilbert comics would not be so popular.

  29. Andrew Rettek says:

    Another thing that costs $438B is ending the recession. Ignoring any multiplier effects ( ask discriminated puerile out their new wealth under their mattresses), this would increase the US GDP to something like 5%, while reducing economic inequality.

    • Damien says:

      Where’d you get that figure? Obama’s stimulus was about $800 billion, spread over two years, and was immediately condemned as 1/3 the size it needed to be by Krugman and others. So that’s $2.4 trillion… of course, that’s outlays, offset by the benefits of avoiding the cost of a depressed economy. If I understand Keynes right, ending the depression should be socially profitable, and possibly profitable for the government. Lessee, 5% of GDP is $700 billion/year, so by now that’s $3.5 trillion out of output we’ve lost. Plus stimulus spending lowers later welfare/benefits spending, while lack of stimulus means decaying infrastructure and undereducated children…

      • Andrew Rettek says:

        Government stimulus and novel commercial activity are fundamentally different. With a stimulus you either have to tax the money from someone or borrow it, both of which have negative effects on the economy ( the extent of those is debated). Novel economic activity without distorted incentives would just be flatly added to the GDP. We don’t usually talk about such things because no one has a magic button to create it ( the closest we have is stimulus packages).