Why statistical discrimination is higher than is either socially optimal or Bayesian rational

Let’s say there is only a mild amount of statistical discrimination in a system.  Not prejudice, just a social judgment that some groups are more likely to succeed at some tasks than others.  Most people, for instance, do not expect women to reach the NBA, but I would not from that conclude they are prejudiced.

But now introduce a further assumption.  There are multiple layers of evaluation, and at each layer people, and institutions, wish to be seen as successful talent spotters, mentors, and coaches.  High schools wish to promote students who will get into good colleges.  Colleges wish to invest in students who will get into best grad schools, or get the best jobs.  Firms wish to hire workers who will rise to CEO, even if elsewhere.  And so on.  Let’s say there are ten levels to this “game.”

Each level will apply its own “statistical discrimination” tax, whether intentionally or not.  Say for instance there is (mild) statistical discrimination against women at the CEO level.  Firms that wish to hire and promote future CEOs will be less likely to seek out women to hire, including at lower levels.  This may or may not be conscious bias; for instance the firms may decide to look for certain personality traits that, for whatever reason, are harder to find in women.  They’ll simply be making decisions that give them plaudits as good talent spotters.

Colleges will then consider similar factors in their decisions.  And so will high schools.  And so on.  In equilibrium, all ten levels of the game will levy a partial “statistical discrimination tax,” with or without conscious bias in thee discriminatory direction.

Does this sound familiar?  It is a bit like the double/multiple marginalization dilemma in microeconomics.  The number of discrimination taxes multiplies, at each level.  Just like the medieval barons put too many tolls on the river.  All of a sudden the initially mild statistical discrimination isn’t so mild any more, due to it being applied at so many veto-relevant levels.  (As you will recall from the double marginalization problem, each supplier does not take into account the effect of his/her mark-up or tax on the gains from trade elsewhere in the system.)

So say the “Bayesian rational” level of statistical discrimination is a five percent discount.  You can get far more than that as the actual effective tax on the disadvantaged group, with everyone in the system behaving in a self-interested manner.

And of course these taxes will discourage effort from the disadvantaged groups, to the detriment of efficiency and also justice.

I am indebted to Anecdotal for a useful query related to this discussion.


That's actually an excellent way to think about it. I like that in the end you acknowledge the efficiency problem, which is all the lost talent along the way because of this multiple taxation system. Having identify the problem, what do you see as a potential solution?
I mean, everybody is behaving rationally, so asking them to remove the tax does not seem like the way to go.
One of the solutions I have thought about is more heavily tax individuals starting their own firms. If they manage to make it big, in spite of the weights against them, they can reshape who gets chosen for all levels on their own company, this lifting some of the tax on those people. Then those people can go on an duplicate it. I think of this as the Ava DuVernay or Oprah model.

It's a good model if everyone is trying to maximize the same objective function each step along the way. The reason why double marginalization occurs is that each successive monopolist wants to maximize profit. As pointed out by some below, this is not at all the case when it comes to the career ladder. What people look for in candidates for lower level jobs is not the same as what they look for in candidates for top level jobs. As people low on the totem pole gain experience and develop various skills, they will then be able to move up the career ladder. You go from junior data analyst with no underlings to senior data analyst with a few underlings. Your bosses see how you deal with that. But your skills for managing underlings was not why you were hired and were most likely completely non-existent before you got hired. You got hired because you passed some basic technical challenges they set you. People with better managerial skills might be left out in the cold, but your model is going to have to much more sophisticated to integrate statistical discrimination into it.

Hmm, and is the statistical discrimination caused by overt discrimination? For example, minorities get into university with a much lower test score, they are given extra help to shepherd them through the program, so they end up less qualified than a non-minority with the same educational qualification.

This creates a reason for statistical discrimination that would otherwise not exist. So attempts to remedy possible bias that involve a finger on the scale may make the problem worse?

This is also a good explanation of gun-control. One gun-control law may be OK (like being 18 to purchase), and even two laws may not violate the right to keep and bear arms. But the current leviathan of gun-control laws creates a significant undue burden to most people owning and carrying whatever firearms they please wherever they want.

Open carry in my state. Can own as many guns and have as much ammunition as i want. Not really sure what undue burden I am facing. Except for auto travel. The laws and rules and traveling with a gun in the car have always been hazy and after discussions with pro-gun lawyers still think they could be made more clear. Only time I actually drive under th speed limit is when I go to the range.


Well, one undue burden is that when carrying, anyone in law enforcement can arbitrarily and blow you away if you move (or even might move) and then claim they has a "reasonable" fear for their life.

sorry, should have been: can arbitrarily blow and claim they had

I'm not sure I'd really agree with this. A better application might be if you had national, state, local and even locally zoned gun control laws causing a host of confusing issues depending whether you are at home, at work or going between those points. For the most part national gun control laws are almost non-existent and states set their laws on the state level.

I could see some special cases where this might apply, though. For example a cross country truck driver who wants to tote his gun(s) with him as he goes through any of the lower 48 states may have a confusing time.

"carrying whatever firearms they please wherever they want."

That is the problem, isn't it?

"Does this sound familiar?"

Well, yes. It sounds like common sense dressed up in a lot of fancy words.

A drop of oil will flow through one kitchen strainer, but it won't flow through ten kitchen strainers put together.

If you go out in the cold wind wearing a thin sweater, you'll feel the wind. If you go out wearing ten thin sweaters, you won't.

Good analogies.

He is framing it as a "tax" to connect it to existing economic theory. You wouldn't get that if he discussed coffee filters or sweaters.

The framing of the question suggests a remedy that he hasn't discussed. It appears that removing the "tax" at each level is necessary. I dont know what he is implying. Using Steve's example of cinematographers, that might involve hiring dumb brutes to do the heavy lifting so aspiring cinematographers can focus on the work. The increase in costs could be absorbed by reducing entry level cinematography pay.

This concept already exists in discrimination law. It's called "barrier analysis." It is an affirmative effort by employers to seek neutral, non discriminatory facets of a job that nevertheless have an adverse impact on a protected class. Such factors must not be rationally related to job performance. For example, a swimming requirement might be necessary to be an Army Ranger notwithstanding the fact that that requirement has a disproportionate impact on Blacks who swim at lower rates than other races.

Someone clearly has some free time to ensure the comments are burnished to a high gloss.

What does your comment even mean? You must have a lot of time on your hands to be making unhelpful comments on the blog of someone for whom you have a longstanding grudge. You should seek therapy to deal with this mental illness and, above all, stay away from this blog. Move on with your life.

Did you ever consider that the evil Cowen is ruining your life by not blocking you? He has you trapped in a perpetual cycle of indulging your hurt feelings. You're reduced to the significance of a gnat.

Write your own blog, why don't you?

Of the first three comments, two (one from Lila and one from me) were scrubbed. Still not really sure what was wrong with Lila's scrubbed comment, to be honest.

'that the evil Cowen is ruining your life by not blocking you'

Prof. Cowen is not evil (nor is Prof. Tabarrok), and he has about zero to do with my life. I have never met or dealt with him in any capacity, to the best of my knowledge.

'You're reduced to the significance of a gnat.'

Oh please, I don't even have that significance. It seems as if you cannot imagine typing comments as a type of online hobby - for example, after spending the last couple of hours at a local lake, I wrote this comment while drinking an ice coffee and listening to WFMU, playing Aretha Franklin songs.

'Write your own blog, why don't you?'

You still haven''t figured out this web site is a PR exercise, have you? I did PR - and have no interest in anything but 'Hohn und Spott' regarding it.

Maybe they just banned Russian IPs.

Wouldn't Bayesians just factor this in as well...? ...Ah, you don't actually say people are Bayesians....

Anyway, analysis implies undervalued women, minorities, and for that matter, white males applying for e.g. HR positions. what biases/regulations/economic factors are stopping firms from picking up these gains?

They used to say that nobody ever got fired for buying IBM.

In the NFL, the coach least worried about getting fired, Bill Belichick, plays the whitest receiver corps.

And Aaron Hernandez

You see this in the NFL where all starting cornerbacks on all 32 teams have been black since Jason Sehorn retired after the 2003 season. It doesn't make much sense for a white or Samoan youth to specialize at cornerback since the hurdle of overcoming expectations so many times is so great.

On the other hand, does it make much difference? If football was 100% colorblind, would NFL cornerbacks instead of being 100% black be, what, 95% black?

More likely 100%. Look at who the top sprinters are, and in the sprints there aren’t ten layers of subjective evaluation: go to the track, do a timed sprint, and everybody can see how good you are.

It makes perfect sense for a white or Samoan to try out for cornerback IF they have the talent for it. Your last sentence explains why. You believe that Blacks make up 95% of the upper tail of the cornerback distribution. A white or Samoan in that upper tail might face some stereotypes or prejudice, but there are enough opportunities for him to prove himself on dozens of pro teams, Canadian football, European football, and before that in college.

A white or Samoan in that upper tail might face some stereotypes or prejudice

Stereotypes/prejudice in this context can be re-framed as a heuristic or rule of thumb. In NBA level basketball, the best men can beat all women players, therefore if you were recruiting for a new NBA team you make the assumption that you will save time by only looking at men (thereby cutting your search space by roughly 50% from the entire human population). Even bigger cuts in the search space can be made by other assumptions/prejudices/ biases that are less controversial...for example, you will probably also assume that it only makes sense to look at people who have actually played basketball!

I think the problem with Steve's analysis is that these rules of thumb heuristics save time and energy in the search....but if the position you are filling is very high stakes then why would you be using 'rules of thumb' to save time? What costs an NBA team more? Spending a bit more on recruiting to find the best possible player or saving on your recruiting costs but getting the 15th best player rather than the 2nd best?

I think you're onto something with the search cost argument. Search isn't free: maybe an effective way of testing this model would be to see if statistical discrimination rates are higher in higher search cost environments?

The other complication may be in skill sets which are harder to evaluate. If Manchester United hires a new striker who doesn't score goals, it's easy to see that he's not working out. But a portfolio manager whose performance rises and falls with the market? That's much trickier. Perhaps you'd see greater discrimination rates in sectors with less quantifiable contributions.

My guess is that white or Samoan athletes who have what it takes to make the NFL at cornerback tend to convert to safety or linebacker, or quarterback. I saw a white QB in a high school championship game outrun future all pro cornerback Richard Sherman on an 80 yard draw play. Now he's in the real estate development biz.

Whites with 100m dash talent tend to focus on the 200m or the 110 high hurdles. The French white guy who ran a 9.92 100m in 2011 gave up the 100m and won a 200m bronze in 2016 Olympics.

But all this is fairly marginal in the big picture.

> The French white guy who ran a 9.92 100m in 2011 gave up the 100m

What are you talking about? He has continued to run 100 m. He hasn't been as fast, so he hasn't won as much. But he did continue to run, for instance winning a silver in the 2014 European Championships.

Boonton / Ryan are right from a decision theory POV. As an quantitative analyst I'm amazed that more people don't understand the perfect rationality of most "prejudice" from this perspective.

Search/detailed information on subjects isn't free or immediate. It's entirely proper to use cheap heuristics winnow choices, even if they (occasionally) lose the best candidate or lead to the wrong decision that way. The proper extent of heuristic use is determined by the error rate on the heuristic vs a "proper" investigation, their relative costs, and the marginal value between selecting the nth and (n+1)th best candidates. Those young black men in hoodies walking towards you are probably harmless, but you can't check their criminal records right now and the costs of a false negative are high compared to a false positive, so why not cross the street to avoid them?

For entry levels positions or where limited information is available, it may be entirely reasonable to bin candidates based on race alone. But for higher lever positions or where more detailed information would be is available or sought out, the value of such racial heuristics will gradually fade away "Sure; X is white in a field where we know that the top percentile are 95% asian, but his stats are just as good, so bring him in for more tests... ".

I don't think this is a very good application of this idea.

With just 32 NFL teams hiring cornerbacks the goal isn't to find a better than average cornerback but the world's greatest cornerback. It doesn't matter if 95% of better than average ones are black, if the top 4 or 5 are white then your team should be 100% white cornerback!

Look at it another way, suppose you're betting on whose going to be the world's greatest chess player. Until the 70's, you'd bet on some Soviet because that's 'expectations' and on average that would be justified...until Bobby Fisher came along, then your expectation game would fail. If your job was to create a portfolio of better than average players, pulling from just Soviets might indeed work even during Bobby's era, but if your job was to build the world's greatest chess team, using the rule of thumb would be fatal.

Good effort, Boonton, but not quite right.

There's a given error rate even on the very best tests you can administer, and you run into decreasing returns. That is, you can't exhaustively test everyone forever and expect to improve your net outcome to global maxima; eventually the marginal testing costs will exceed the marginal expected gain in selection outcome. You should stop testing at this point. Even for the highest rank positions.

That is to say, from the decision theory POV, the value of the crude racial heuristic fades towards zero but never quite vanishes....its a Bayesian thing, ultimately...

Doesn't make much sense to me. If certain candidates for a position have been unfairly filtered several times before they apply, they will be consistently of much higher quality than the unfiltered candidates. So on that level the folks doing the hiring should catch on: "Just take the woman, they are consistently much better anyway." And of course that happens at every step along the way: Biases are adjusted based on the reality one encounters.

Sounds right. There should be a natural selection effect where the population that survives discrimination becomes progressively higher quality at each level. For example, subjecting bacteria to antibiotics can create a population of superbugs resistant to antibiotics. I think that in the 1980s black MLB baseball players as a group had higher batting averages than white players. One explanation was that black players had to be better than white counterparts to make it to the big leagues. I don't know what the situation is now.

If Tyler's example were true, we should see women CEOs strongly outperforming male CEOs, women VPs outperforming male VPs by slightly less, and so on down to lower levels. At lower levels, men might actually outperform women. The lower level performance differences would give rise to the discriminatory "social judgement" but, somehow, the reversal in performance at the higher levels would go unnoticed by the evaluators.

epistemic status: conjecture; feel vaguely uneasy about arguing for the status quo; afraid I'm just being instinctively contrarian

From a gatekeeper's perspective:
"Well, women are better at this point. But that is a known factor for the gatekeepers below me to a certain degree, if smaller degree. Thus the men most be really good as well, since they can keep up with these incredible women."
I think, this would be a potentially rational thought process negating that effect. How would a higher gatekeeper even reliably know, how the selection process at the lower levels have been applied? Or maybe a lower gatekeeper overdiscriminated (everybody acts on incomplete information after all)?
Or maybe between gate 4 and 5 the candidate made a first bad impression, because she was at that point hitting a skill plateau and performing lower than her average for a time. Maybe you need the Halo effect to even be able to overcome this.
Or maybe the first five gatekeepers need a longer time to let such a candidate pass, demanding higher standard of evidence. This would automatically make any stat-discriminated candidate older and thus look and even be worse at a higher levels, since other non-discriminated candidates would have been able to pass the early gates faster and have
more responsibility earlier.
It seems to me, that after a certain point the more gates there are, the less likely it is to create a meritocratic hierarchy (assuming all the gatekeepers even care about that goal). Though it is probably true,
that this setup will prevent disastrous candidates getting to the top, as well.
Having fewer gates or no gates however increases the variance at the top, which happens in the explorative part of
the economy. An exploitative company (mature business model and stable market position) like Volkswagen will never have somebody of Elon Musk's caliber at the top, but that is ok, since even a brilliant person at the top, can't do as much good, as somebody like Elizabeth Holmes can do damage.

I think, trying to match someone with the best, true potential to the exact matching position in any given hierarchy, might not be desirable, at all.
Sorting 'safe', highly conscientious, conformist and reasonably intelligent candidates at the top for established hierarchies (government bureaucracy, Fortune 500) and putting 'interesting', weird, unorthodox and charismatic people into startups might just be the best outcome, one can hope for.

tl;dr: The current system does discriminate 'unfairly', but a "MERITOCRACY UBER ALLES"-approach clashes with the optimal levels of safety/risk-taking in the exploitative/explorative parts of the economy and is thus overrated.

Good comments, BC & Phille. While reading them, the name Margaret Thatcher popped into my head. A lot more filtering 40-70 years ago.

No, because if the signal of skill is "years of experience," then the person filtered at the lower level will always look objectively worse at the higher level filters.

You're assuming that the higher level decision makers have the opportunity (and desire) to consider walk-on candidates. You're also assuming that those walk-ons will have adequate means to signal their superiority to those who passed through the filters.

I might be able to be the best CEO who ever lived, but my lack of management experience would never get me in the door for an interview. If (mild) discrimination at several rungs of the ladder kept me from rising to the penultimate rung, I'd have ZERO chance of attaining the top rung, not merely a small chance.

Or the lower level filters are passable, but slow a stat-discriminated person down so much, that once she's reached the penultimate rung, someone who was not being discriminated against competes against a person, whom she could have been outcompeted, if gatekeepers had not gotten in her way as much killing her momentum.
The 25 year old male phd is more impressive than the 27 year old phd, when deciding who becomes a professor, but a stat-discriminated 27 year old phd, could have been a 24 year old phd if not fot stat-discr.
Most elite performers are impressive throughout their lives. But they can stay constantly motivated, by rising through the ranks quickly. A stat-discriminated person might not have that advantage either.

I agree with Phille. Solve for the equilibrium, as Tyler would say.

The assumptions are completely unreasonable. Firms don't evaluate every worker for CEO-potential, firms only evaluate CEO candidates for CEO potential. Rest of jobs are selected according to their ability to do that job, and maybe some nebulous measure of "growth potential" specific to that job.

Without this assumption, the whole chain of reasoning breaks down. There's multiple metrics to look for at each layer: math score to be an engineer, English scores for writing jobs, etc. And for each metric, the statistical tax can fall on either gender, so when all the metrics are considered together, the statistical taxes can largely cancel each other out.

Also, there's no reason for each layer not to consider P(talent)|P(previous layer considered them to be a talent) since they have the degree/recommendation to back them up.

Sometimes there is one hurdle at the beginning of a career that can filter out an entire sex.

For example, no woman had ever been nominated for a Best Cinematography Oscar before Rachel Morrison in 2018. The main reason is because the entry level cinematography job requires carrying heavy lights up tall ladders. It's like being a roofer.

Few Oscar winners were still carrying heavy objects when they got nominated, but they all started that way.

Fascinating point.

Who or what is a CEO candidate? How is it determined? Isn't that the entire premise of the conversation?

Sorry, intended to respond to Len's comment above.

So it clearly isn't essential to Cinematography to carry heavy lights up and down. If we changed the way things worked, many cinematographer may resent the female upstarts who 'didn't pay their dues' but if the goal is to get great cinematography are we getting as much of it as possible by excluding half the population at the outset?

'The assumptions are completely unreasonable. Firms don't evaluate every worker for CEO-potential, firms only evaluate CEO candidates for CEO potential.'

Clearly, you have found the PC MR way to make this point - congratulations.

'Without this assumption, the whole chain of reasoning breaks down.'

Of course it does. However, you seem to have the better way to show the emperor not wearing clothes, even as everybody is expected to buy into the idea that Walmart evaluates all of its employees for their CEO potential.

A few college basketball teams specialize in recruiting the more often overlooked ethnicities, such as Gonzaga in remote Spokane (home of Rachel Dolezal), and then keeping them around while they physically mature, in contrast to Kentucky's one-and-done strategy.

St. Mary's specializes in recruiting Australian basketball players.

Kentucky's recruiting classes are all All-American top 25 players. They're good enough to reach the Final Four with a mostly Freshmen squad but not quite mature enough to win it all. This year they have a graduate fifth year senior on the team. Perhaps Kentucky can find one or two graduate transfers each year to complement their one-and-done All-Americans.

KU won in 2012. It depends on how good the freshmen are that year. I think Calipari is making a good bet.

Kentucky is UK, Kansas is KU.

KU *lost *in 2012. Rock Chalk :(

Agreed, and I'd rather have five one-and-done types than have a roster filled with projects. Makes my job a lot easier.

Mizzou's best quarterback over the past 60 years was "filtered out" of the bigger football schools because he was a couple of inches shorter than the ideal QB. They found innovative ways to move the linemen around so he could see downfield.

Thanks. Short quarterbacks are a good example because they are not politicized. Russell Wilson came up one yard short of winning back to back Super Bowls, but nobody was too interested in the discrimination he faced as a short quarterback at NC State (eventually transferring to Wisconsin and tearing up the Big Ten) because shorts aren't an organized pressure group.

Another one is lefthanded catchers -- there hasn't been one in the major league baseball since the 1980s, although nobody seems to be able to agree on why exactly. But lefthanders aren't a Victim Group in our society, so no organized group calls on us to question our prejudice against lefthanded catchers.

This comment would be more insightful if lefthanders had been enslaved for centuries. I'm lefthanded, No Justice No Peace!

Remember kids, just keep punching up!

Given that he stated in it how he liked the example because it avoided that baggage, why are you so desperate to add it?

Tyler, your example might have merit if all of these different organization up and down the chain coordinated their discriminatory practices and they each had monopoly-like power with few alternatives. But as long as there are many firms in the market, women can easily avoid the firms, colleges, etc., that discriminate.

Also, the double marginalization problem typically occurs when one firm is imposing some levy or tax on multiple downstream or upstream users, as in your toll problem. There is a common carrier monopoly in that specific example. You also see this occurring in patent licensing, but the patent owner price discriminates in a way that usually leads to optimal pricing.

I've come up with a toy problem which seems like it could work as a model for this situation. I haven't done the calculations yet to see what it says; I'm curious if anyone has feedback on the toy problem or wants to take a crack at the calculations.

(Question 5 below is the one that matches Tyler's scenario; 1-4 build up to it and seem valuable as comparisons.)

Imagine that there are 2 equally sized groups. In group A, talent is normally distributed with a mean of 1 and a stdev of 10. In group B, talent is normally distributed with a mean of -1 and a stdev of 10.

1. Suppose that we select the 1/64 of people with the highest talent levels.
i) What is the average talent level of the selected people?
ii) What fraction of the selected people come from group A?
iii) What is the average talent level of the selected people from group A?
iv) What is the average talent level of the selected people from group B?

2. Similar to 1, but we don't know people's true talent levels, and can instead only give them a test which has a normally distributed error with a stdev of 5. (e.g., If Alice has a talent of 7, then her test result is normally distributed with a mean of 7 and a stdev of 5.) We select the 1/64 of people with the highest test scores. Same questions i-iv.

3. Similar to 2, but instead of selecting the people with the highest test scores we do a Bayesian analysis which accounts for both their test score and which group they came from, and we select the 1/64 people with the highest expected talent level. Same questions i-iv.

4. Similar to 2, but instead of 1 test there is a sequence of 3 tests which each independently have a normal error with stdev of 5. We give everyone the first test and select the 1/4 of people with the highest test scores. Then we throw out the first test scores, give the 2nd test to the remaining 1/4 of people, and select the 1/4 of those people who score highest on the 2nd test, leaving us with 1/16 of the original group. Then we throw out the second test scores, give the 3rd test to the remaining people, and select the highest scoring 1/4 of them leaving us with 1/64 of the original group. Same questions i-iv.

5. Similar to 4, with 3 tests whose scores get thrown out after winnowing the group. But at each stage we do a Bayesian analysis (taking into account their group, their most recent test score, and the fact that they made it to the previous stage), and select the 1/4 of remaining people with the highest estimated talent level. Same questions i-iv.

6. (Optional) How do these results depend on the specific numbers chosen here. e.g., Does anything substantial change for i-iv if there are more than 3 stages in winnowing down to 1/64, or if the difference between groups is larger, or if the selected population is larger or smaller than 1/64, or if the test error is larger or smaller?

7. (Optional) Design a more effective selection procedure for the scenario in step 5, such that the average talent level of the selected people is as high as possible. With the constraint that you may only give 3 tests, and you must throw out the results of a test before administering the next test. What results to questions i-iv does it give?

I defined group 1 to have mean of 0, group2's mean=-1. Result of 1 million simulations. More will come at lunch time.
1a. The average highest score of 64 people: 22.97
1b. The percentage of the highest score from group 1: 0.559131%
1c. The highest talent from group 1 when it was sample max: 23.05094
1d. The highest talent from group 2 when it was sample max: 22.83663%

2a. Average talent of highest scorer 20.5, highest test score 25.8. decline makes sense as less likely to get "true" max.
2b. Percentage of highest test scorers from group 1 55.4%
2c. Talent from group1 when person had max test 20.7
2d. Talent from group2 when person had max test 21.8
Group1's talent declines more when selecting on the test as you're no longer getting the maximumly talented person and as group1 had previously placed the more people, and the quality of the people they do place decreases.

Add in a feedback effect. What if the talent levels of group A and B depend on their likelihood of being selected, where the selection perfectly mirrors the distribution. What happens to the talent levels of group A and B over time, and what subsequently happens to the selection process?

That seems to be an argument against selective and effective training or education. I.e. the best performers get in the best training programs and become even better?

i think it's an argument against allowing statistical discriminaiton to determine who gets selected for training and education. I.e. don't just say "well, we're going to give the boys better math education, because they are the ones who are most likely to excel at STEM careers." if you do THAT then you're really setting up the self-fulfilling prophesy by ensuring that the girls who might be good at math won't get the same chance to develop their potential. And you'll be giving extra math training to the boys who aren't actually better at math.

If you test the class for math skills and let the highest scoring into the program you have the same problem, don't you?

"don't just say "well, we're going to give the best scorers better math education, because they are the ones who are most likely to excel at STEM careers." if you do THAT then you're really setting up the self-fulfilling prophesy by ensuring that the lower scorers who might be good at math won't get the same chance to develop their potential. And you'll be giving extra math training to the high testers who aren't actually better at math."

Presumably test scores actually are accurate measures of ability. At least it's a hell of a lot more accurate than just using sex as a marker.

So then it's not "an argument against allowing statistical discrimination to determine who gets selected for training and education." So your principal is what? Ignore weak signals?

Maybe I'm being slow but I'm not following why this would have a multiplicative effect on that 5 percent rational discrimination. If you think an NBA team is going to discriminate 5 percent and you're a college program, are you going to discriminate >5%?

Michael Jordan was cut from his HS basketball team. That's why you never heard of him.

Michael Jordan was not cut from his high school basketball team. He was just passed over for promotion to the Varsity team when he was a sophomore.

Thanks for the correction.

Michael Jordan was the second choice in the NBA draft his year. The Portland Trail Blazers used their first pick to bring in the legendary Sam Bowie. So much for foolproof filters.

Actually no, he was third.

The Rockets took Hakeem Olajuwon first...no shame in that.

Bowie was taken second because Portland already had a hall of fame shooting guard named Clyde Drexler.

Background rates applicable to very broad groups (e.g. women) have Bayesian relevance to toy problems, but simply can't be used in real life cases that aren't nearly cookie cutter - like hiring a CEO. ("can't": we aren't smart enough and we don't have enough data to get it right even if we were smarter). This is one reason why blunt anti-discrimination laws are rational as a least-bad rule to live by.

My preferred example, but it's the same thing really... Let's suppose group X (men? blacks?) have been statistically proven to commit more crime - take that as an unquestioned given. Now you are on a jury - a criminal trial with a type-X defendent, "x". You will learn all sorts of things about "x" during the trial, not the least of which is that "x" is a plausible enough culprit for prosecutors to bring the case. There's no way in hell you can invoke the background "X-criminality rate" in your verdict and justify it Bayesianally - at this point "x" is very far from a random "X" and the knowledge you have is precisely the type that is likely to subsume, undercut, (or on the other hand reinforce), the relevance of the base rate. The assumptions you'd have to make to invoke the base rate would be embarrassingly absurd if you had to spell them out. Best to have some epistemic modesty, and ignore the base rate entirely. It's not as if you have an honest choice.

It is strange that Cowen is worried about "mild amount of statistical discrimination" at various levels but ignores heavy discrimination in favor of the "under-represented" that occurs at many levels.

You've completely misunderstood the setup.

At no time did he state or imply that there isn't "heavy" discrimination. At no time did he state or imply that he didn't "worry" about that.

He is saying that successive layers of mild discrimination (e.g. stereotypes) can act as if it is a complete bar at the highest levels. Choosing a CEO, for example, from among a set of senior managers could be completely discrimination-free in and of itself, but because some groups were successively denied the skills and experience to be chosen, the ultimate choices are discriminatory.

But this would seem to fail if the those at the top are aware of the overtaxation.

Suppose I make decisions at the 7th round of such a setup. I am actively competing with others for talent and I realize that the over correction. I take a gamble on the marginalized and find immense rewards by more accurately weighting my efforts.

The situation thus described can only sustain itself if all actors are unwilling to change. After all, each gatekeeper has incentive to try something different and get better talent.

Well, with the first gatekeeper there’s really no motivation to gamble on the marginalized, because a) there’s not yet a big gap and b) doing so would give up the profit of stereotyping (see the Bayesian analysis referenced in #3 of Dan’s post).

And as you get to the subsequent gatekeepers, a larger actual ability gap starts to form because the discrimination of prior gatekeepers prevented the marginalized group from gaining valuable experience.

It’s an excellent point I hadn’t thought of before Tyler’s post.


Suppose we have 100 level 1 gatekeepers each looking for the best people to fill 1,000 slots in a 10,000 person community.

Say they use a heurestic to cull 5,000 from competition and sacrifice 10 of the best performers. This cuts their search efforts in half. So they all find the same top 90 performers. Somebody is short 10 positions, so they start bidding (e.g. better scholarships, better amenities). Eventually the cost of bidding up the easily located talent overwhelms the advantages in search savings.

Further, at least one firm should do the math that it would be just as easy to come the bottom 5,000 for 10 elite applicants where you have no competition than to come the top 5,000 and compete.

And suppose we also look at the human capital argument. Subsequent gatekeepers face the same pressures. They can bid higher to go after the optimal candidates for combined talent and human capital ... or they can pay less in bidding for the suboptimals. E.g. recruit your coders from non-English speaking areas and hire English instructors, pay for your bright under capitalized employees to get their MA on company dime.

Statistical discrimination only reduces your search costs. It can result in Tyler's Just So Story if, and only if, there is no bidding for the top candidates or there is no way to substitute bidding for talent and human capital (like on the job training, company subsidized education).

We also would need a world with no innovation in candidate measuring. After all, at every level a more accurate test of the applicants would be worth a huge amount to any firm that implemented it so somebody has incentive to develop one.

It is a very odd world where the only cost to be minimized is the search cost and the only productivity inputs your firm cares about are all locked with past behavior and there is no improvement ever in testing sensitivity or specificity.

Taken to the economy as a whole the Just So Story must break down. Some firm must be an outlier and reap advantages from taking a different cost strategy (after all we saw highly profitable firms built on exploiting the wage discrepancies of Jim Crow and today we see firms preferentially hiring ex-cons in some capacities because they are more lucrative due to their reduced life options). In an efficient market, many actors are going to try many different things. Expecting all of them to always find the same sub-optimal equilibrium is not going to happen for long in a large market.

As it has always been, discrimination needs forces behind it other than economic optimizing ones. Those can be cultural. Those can be from the government. The one thing they will not be is strict profit maximization.

The setup of the model is so messed up, I can hardly fix it.

First, the entire question has an implicit assumption that merit based decisions, absent discrimination, are best. Fair enough.

The reason there are no women in the NBA is simply because there are no women at the exemplary end of the distribution. There might be some women with the ABILITY, but those women might not have the desire. Remember that when Jackie Robinson broke the color barrier in baseball, he was an EXCEPTIONAL athlete, not just an exceptional Black athlete. He and others were kept out of baseball purely because of racism, not lack of merit.

While merit in the sense of sports is easily observable, merit for being a CEO is not. We could talk vaguely about several decades of successful management experience, but then argue about what successful means and how many years is enough.

Suppose we observe no women as CEO of the largest firms. Further suppose that this is caused entirely by several layers of discrimination. The women have the skills, but they have been suppressed. Nevertheless, regardless of talent, they lack EXPERIENCE and a RECORD of achievement. If we ended gender discrimination today, it might take 20 to 30 years for women to acquire the resumes necessary to be CEOs. To do otherwise shortchanges the system. It says that character and ability means everything and experience and a proven track record mean nothing. In other words, the premise attacks the WAY that CEOs have been chosen in the past.

Could someone with far less experience successfully run a large corporation? Of course they can. But when making decisions about innate characteristics that are poorly observed and measured superficially, the probability of such choices turning out badly are really high.

Steve gives a good example above of how a neutral factor can act as a sieve to certain groups of people without there being any discrimination whatsoever. It is possible that carrying heavy equipment isnt a good screening criteria for higher level cinematography jobs . Does a woman who never actually played football qualify to be a coach or a sportscaster analyzing play? In some cases yes, in some no, but we are not making choices with complete information. Lower level experience signals quality. Removing the signal makes our choices less good, on average. The only way removing a signal can be good is if the signal is bad, as racial discrimination is.

But with baseball, I dont think the prevailing wisdom was that Blacks couldn't play baseball well. It was a secondary preference for simply not wanting them around.

Tyler fails to take into account situations where one discriminated group overcompensates -- as with Asians. He assumes that discrimination will lead to less effort on the part of minorities. But if that is true only for some minorites, then statistical discrimination may be more efficient, inasmuch as it motivates the high achieving group more than the low.

I like Willitts baseball reference. Few black Americans choose baseball: most of the black players in MLB are from the islands. Why? The path to a MLB career is not nearly as appealing, or available, to black Americans as compared to basketball and football. Youth baseball is a year-round endeavor, requiring investment of time and money by the parents, something most black families cannot afford. Moreover, college scholarships in baseball are only partial: each team is allowed only 11.7 scholarships on a roster of 35 to 40 players. Finally, even if a player is drafted by a MLB team, most likely he will play in the minors for at least two or three years if not longer. Is this a form of discrimination? No, it apples to everyone, but has more of an impact on black players than white players. My point is that "discrimination" can take many forms, some of which is based on circumstances not race, gender, etc.

Significant scholarships exist, so not a filter. Must play in minors affects everyone, not a filter. Kids growing up on islands play all year round and are mostly black. Filter in favor of potential black players.

First of all, black Americans are proportionately represented in MLB. So in order to argue that there is some discriminatory filter on them, you have to posit that they are genetically superior to other races at playing baseball.

I question whether baseball is more difficult than baseball or basketball. Baseball is often not year round. It's certainly no more year-round than basketball. It doesn't require expensive equipment. You may spend time in the minor leagues but you get paid and you can also get a very large signing bonus. In contrast football players have to work four years for free in college and the vast majority never get any money, the ones that do are typically in the NFL for a couple of years on min deals and then out to enjoy their injured brains.

Vox has argued that black representation in the MLB is evidence of racial discrimination. They've remained silent on the matters of the NBA and NFL.

Sounds racist

I think this is a case where I'm skeptical and want to see a formal model.

>Not prejudice, just a social judgment that some groups are more likely to succeed at some tasks than others.

That is literally the very definition of prejudice.

>Most people, for instance, do not expect women to reach the NBA, but I would not from that conclude they are prejudiced.

This attitude should get you fired, especially given who you work for, and who you voted for.

Replace "women" with any ethnic group and "the NBA" with grad school to see what I mean.

Or are you simply saying it's fine to dismiss the physical abilities of women? That is approved discrimination, in your book?

What is this, 1912 in your world? Is it still ok with you if they vote?

">Not prejudice, just a social judgment that some groups are more likely to succeed at some tasks than others.

That is literally the very definition of prejudice."

+1, but I get why Tyler is dissembling on this point. It's certainly prejudice, but it's debatable whether it's justifiable, unless you take the a priori presumption that prejudice is never justifiable. That would explain why Tyler is concerned to argue this isn't prejudice, because that has carried a certain moral opprobrium for about 50 years. Back in the 50s and 60s, many white Americans drew a distinction between being "prejudiced" against black people, which was generally viewed as okay and understandable (if not necessarily good, just a personal failing), and being racist, which was Bull Connor, George Wallace, the KKK, and white supremacists in the South. For most people today, that distinction has collapsed; in fact, it had probably collapsed for a solid 70-75% of the country by the time Jimmy Carter was President. Hence why it is necessary to argue that something which appears to by definition be prejudicial is not in fact prejudice.

That is not literally the definition of prejudice. The literal definition of prejudice is:
'preconceived opinion that is not based on reason or actual experience'
Making assumptions of likely behaviors or the range of somebody's abilities of any given group member from the statstical average of his group or your past experience with that group, given no other or only limited information, is reasonable, thus not a prejudice. Limiting yourself to ignore all your own past and second-hand experiences would not be reasonable, as a matter of fact, that would be close to the literal definition of 'naive', which is:
(of a person or action) showing a lack of experience, wisdom, or judgement

I think this piece was good, productive, and perhaps a gateway for some to think about the interaction of "prejudice" and "social" or "statistical" judgement.

(I'm pretty sure that people with pretty strong prejudices think they are just being statistical.)

That aside though, a hypothetical non-prejudiced and hard working Bayesian should recognize that an accumulation of both not-prejudice and prejudice in the population adds up to .. structural prejudice.

As I think this piece is saying, there are "statistical" choices you think you are making, and there are "statistical" choice made out of your vision, and filtering your input data.

Color me confused. Does that first (or second - or are we going back to kindergarten?) level have to operate quite the same as all the other levels?

I'm thinking of the custom, in the years I was aware of such things, that MIT would annually select one senior girl and boy from our "highly-ranked!" local high school for admission. It was in a mild, unspoken way the top academic prize. The competition for the boys was very tight, not so much for the girls - but though there was disappointment at not getting in, I never heard of any sour grapes at a less-purely-talented girl getting one of the slots. Excellent schools awaited the losing boy candidate or candidates, and in any case they had been socialized at public school and would not have challenged the importance of girls going to MIT.

If in a given year there wasn't a slam-dunk whiz kid girl who could win admission to MIT on academic merit alone, the chosen girl would have demonstrated extreme diligence at her schoolwork, and a sport. Always a sport, even though this was MIT we're talking about. I want to say there was probably never a sport involved with the prospectively-MIT-bound boys.

She didn't have to be as exemplary in terms of math board scores, compared to the MIT-candidate boys, and maybe not necessarily even sport, compared to other girls playing that sport, since in that case she probably would have gone to a college more known for its athletics.

So she was spotted a few points, big deal, but are you saying that what looks like advantage, will work against her at higher levels? Or maybe it's mistaken to speak of "advantage" when someone is so obviously hard-working? Or is the existence of one slot for a boy, merely a byproduct, to give the appearance of fairness, of the one slot for a girl, and it's girl talent that is being squandered at that level?

More generally, it seems to me that at lower levels any assumption about what girls and boys tend to be good at or interested in, results in a somewhat frantic input of effort into reversing it - e.g. "Coding Camp for Grrrls only!" - because it's not true that girls don't adore coding! And we think coding's the crucial thing! So we made a camp just for girls to dispel this widely-held belief ...

I think on some level society kind of realizes that sorting girls into traditionally women's occupations, (either homemaking or standing around looking pretty while doing not particularly challenging work) is sub-optimal from an economic standpoint. Since software development and STEM work is generally not physically strenuous, there is no reason why women shouldn't be able to do it. And not every STEM job demands theretical physics level math skills. Assembling and soldering a circuit board according to a schematic isn't any more difficult than embroidery, and pretty much involves the same detail-oriented physical skills and hand control.

No doubt! It's unfortunate, though, that the obtuse have invented some kind of connection between what you're describing, and going to MIT (for example).

I completely agree. No one should do the sorting except the person themselves. There's been a study going around with a graph showing the more free a country is the fewer girls go into STEM. No doubt the girls could do it, but it's also OK if they chose not to.

Right, but if that's due to social brainwashing, then it might be a good idea to do something to counter the social brainwashing. People often make choices because they think it is what is socially expected. They dont want to be "weird.

Right, but if that's due to social brainwashing,

Hazel, you have personal agency. You did it, you own it. Sorry your father never persuaded you of that.

I have no idea what you mean. I have a very enjoyable job in a STEM career, and that's partly because my father taught me 7th grade algebra when I was in grade 2.

"social brainwashing" Girls have been taught for years that they can do anything. Especially STEM. And schools have been altered to cater to any disparity in education, even have overcorrected. Now the only way is to force them against their will. Are you willing to go there?

I think on some level society kind of realizes that sorting girls into traditionally women's occupations, (either homemaking or standing around looking pretty while doing not particularly challenging work) is sub-optimal from an economic standpoint. Since software development and STEM work is generally not physically strenuous, there is no reason why women shouldn't be able to do it.

Hazel, the ratio of computer science degrees awarded to women to those awarded to men was 0.59 in 1984 and is 0.28 today. A mess of chaff-babble about 'traditional women's occupations' is not going to explain that phenomenon.

Hazel, the ratio of computer science degrees awarded to women to those awarded to men was 0.59 in 1984 and is 0.28 today.

The fact that preferences change over time doesn't really support the theory that the difference is due to innate ability. Some environmental change caused that. Probably some sort of cultural change.

Perhaps a legacy of when computer programming was a secretarial job done by women, as the profession turned into a STEM profession women may have become less interested

Keypunching was a secretarial job. Programming was not. See Ted Cruz mother Eleanor Darragh's account of her early years in the profession. She made it a point to not learn how to touch type because she didn't want to be asked by others to type for them.

Keep in mind that at the peak of female interest in computer science as a discipline, 63% of the degrees were awarded to men. It wasn't understood as a secretarial program.

Students of the industry have suggested that it was a more orderly and predictable occupation in the mainframe era and that women began bleeding away because it grew to unstructured.

The fact that preferences change over time doesn't really support the theory that the difference is due to innate ability.

That's not my theory. You're arguing with the ghost of one of Steve Sailer's votaries. Your theory is that revealed preferences are illegitimate because they don't match you nebulous idea of what they should be. Attributing a deficit of women to 'social brainwashing' when participation in an occupation by women has been declining (in the teeth of You-Go-Girl propaganda of a sort which has been the Official Idea for 40-odd years now) is just silly.

It's not just layering, it's also feedback.
Say you have a true statistic that women are less likely to make good engineers. Then you have a bunch of people using that statistic to make judgments about who to hire, who to promote. So you end up with fewer women engineers, which reinforces the statistic. So the statistic becomes a self-fulfilling prophesy, and even if there is a bit of an underlying truth to it, that feedback is going to amplify the effect. This isn't unreasonable - it's a form of self-organization, sorting into groups, all sorts of systems do this, where small differences get magnified into large ones. Another example would be white-black racial sorting, where mixed race people tend to get sorted into either "white" or "black" and constrained in why they can marry because people don't date across racial lines. Like with barack Obama - he's half white, but society looks at him and say "black" so he's probably going to marry a black woman and his kids will be 3/4 black instead of 1/4 black. Society constrains people's choices and sorts them into groups with more magnified differences then are statistically real at the individual level.

"So you end up with fewer women engineers ..." Why don't you end up with a company that spots the opportunity to hire those underrated women engineers and thrives as a result? The whole analysis offered by Mr Cowen is based on an ultra-conformist society. Maybe that applies to the US?

Maybe that applies to the US?\

No, madam snotnose, it applies to the professoriate. Time for you to go back to assessing other people's accents.

"Maybe that applies to the US?" Nope.

1. Obama met Michelle Robinson when just shy of 28. All of his known girlfriends prior to that were white or mixed white / oriental.

2. "Say you have a true statistic that women are less likely to make good engineers. " That's a statistic generated by female college students. About 80% of the baccalaureate degrees awarded in engineering are awarded to young men.

1. But he chose to marry the black woman, not any of the mixed white ir Asian women. Why do you think that is?
2. These are just examples but it's possible that women could choose not to go into engineering because they think they will be discriminated against. I actually think it's mostly cultural brainwashing in this case, but the feedback idea could be valid at the higher levels, like in graduate school if it's harder for a woman to find an advisor because professors are biased against women. Then the women decide not to carry on into grad school, so you get fewer women with PhDs in engineering. So then the professors assume that the women won't finish their phds so the bias is reinforced

1. He was raised basically white, as Barry. He change to Barack later and identified himself as black for political purposes. No white kid with his resume would have gotten past city council.

2. https://twitter.com/SteveStuWill/status/1028672463729643521

Women have free will. Check out the rest of this twitter account, he's pretty interesting.

1. But he chose to marry the black woman, not any of the mixed white ir Asian women. Why do you think that is?

For any and all the reasons in the world you might marry one person and not another.

2. These are just examples but it's possible that women could choose not to go into engineering because they think they will be discriminated against. I actually think it's mostly cultural brainwashing in this case, but the feedback idea could be valid at the higher levels, like in graduate school if it's harder for a woman to find an advisor because professors are biased against women. Then the women decide not to carry on into grad school, so you get fewer women with PhDs in engineering. So then the professors assume that the women won't finish their phds so the bias is reinforced

You don't need an 'adviser' to complete a baccalaureate degree. You fancy people are 'brainwashed' because their preferences are not what you'd have them be. There's nothing wrong with them; there's something wrong with you. Young women may indeed look around an engineering class and decide it would be an uncongenial environment for them. That's the sort of apercu that just about anyone might have in just about any sort of college classroom.

So you think that if someone chooses not to go into a particular field because they think it's going to be an uncongenial environment, that's not an issue worth addressing ? That it's ok for people to create exclusive, uncongenial environments for "outsiders"?

Medicine and law are typically rated as the most hostile to women, and I believe women are a majority in medicine and law? Meanwhile computer programming is one of the least hostile but women are a small minority.

So you think that if someone chooses not to go into a particular field because they think it's going to be an uncongenial environment, that's not an issue worth addressing ?

As a rule, no, so long as tortious workplace harassment is not the order of the day.

What suits people is quite variable. Make it more congenial for Miss X and you make it less congenial for someone else. See Megan McArdle's account of why she got out of programming. She's describing a fairly benign discomfort, but one sufficient to persuade her to follow a different line of work. Was it the business of her co-workers to think and behave in ways that made her feel less out of place, or was it her business to reconcile herself to that feeling or do something different with her life if she couldn't reconcile herself? She says the latter. Most people who aren't hopelessly self-centered do.

Was it the business of her co-workers to think and behave in ways that made her feel less out of place

I would say it is the business of management to create a workplace culture in which the best people can feel comfortable and perform at their best. If your workplace culture is such that good employees feel unwelcome and leave, that is going to hurt your bottom line.

Your boss is not your mother, Hazel. As McArdle makes clear, their bottom line was uninjured.

See Megan McArdle's account of why she got out of programming. She's describing a fairly benign discomfort, but one sufficient to persuade her to follow a different line of work.

Pitty you didn't tell us what the discomfort was. But let's say it's Star Trek. Programmers love Star Trek and talk Trek all day long and love the culture and environment because where else can you talk Trek all day at work?

The approach you are taking is that there's nothing wrong with people chatting about Trek all day so what? But if the purpose of hiring a bunch of programmers is to get great programs shouldn't the question be what did the business and society lose by pushing Megan out of programming? The programmers who already have the job might love turning it into a 24-7 Trek discussion group, but do we get the best programs out of this?

She writes in public venues. You can seek out her column on the subject yourself. Just reallocate some of the time you devote to being a gassy bore.

The essence of it was that she realized one day in the staff lounge that her orientation toward her job and her free time was quite different than that of her co-workers and she was just never going to be like them. They were all talking about their week-end hobbies (which involved basement tech projects) and asked about her week-end She tells them about this concert she went to and this problem with her boyfriend and gets embarrassed stares.

Again this is a perfectly valid legal defense of the firm against a discrimination claim. Just because the programmers hired before Megan didn't have boyfriends or get out much during the weekends doesn't mean she suffered gender discrimination, for which she has a cause of action.

But there is a cost here. The firm, and society overall, does not have access to whatever Megan might have done as a programmer. On top of that we have no idea what contributions might have come from non-weekend dwelling basement inclined programmers had the culture not been so monolithic. Yes maybe the programmers that were already there were already the best possible AND maybe these types are every skittish about their environment so will quit or not do as well if they can't bask in their culture. Maybe there's just no way to make programmers open to people like Megan without costing more talented programmers who need their eccentric culture.

Not that it is the same but the #MeToo movement is a rejection of this 'essential person' thesis....that top people like Weinstein or Spacey are so important their abusiveness must be tolerated least we lose their contributions. It's thesis stands that on its head and is working with an implicit assumption that the people we lose because of such people cost us as much or more than what such top producers contribute.

I eagerly await your ascension as god-emperor, when talk in the break room will be banned

Hey baby, we already have a God Emperor, King of MAGA. And he's my daddy!

Obama met Michelle Robinson when just shy of 28. All of his known girlfriends prior to that were white or mixed white / oriental.

This is why these conversations rarely go anywhere useful. Anyway statistics on n=1 rarely works well but another way of framing this question/observations is

"100% of girls he didn't marry were white or mixes race"... in the context of that totally true statement, the fact that he married a black woman is totally unsurprising.

Also do we know *all* of the women he dated? "Known girlfriends" sounds very suspect. Where is it documented that the only black woman he ever dated in his whole life was the one he married?

The example was hers, not mine, and even her sample of 1 she misused. Obama wasn't compelled by social circumstances to marry a black woman, and you can tell that by an inventory of his consequential girlfriends. (No one's taken an interest in Michelle's social circle ca. 1985).

Well he may have dated non-blacks and due to the social circumstances of race relations found those relationships unsuccessful. People marry people they can relate to. I imagine that it's easier for a black man to relate to black women because the particular experience of racial prejudice in America that black people have is unique. Asians don't experience the same thing.

You're projecting on blacks your own social fantasies and you're projecting on Obama someone else's biography, Hazel. He grew up among Honolulu's haolie haut bourgeoisie. At age 20, he knew nothing of black America other than reruns of Soul Train and chatting with the decidedly eccentric Frank Marshall Davis. When Gov. Blagojevic said "I'm blacker than Obama", he was speaking the truth.

I always thought of O as kind of a white ivory-tower elite even when the adoring media was going black-black-blackity black.

I think the marriage was 1) political convenience; you need a wife and kids for the photo ops and/or 2) Michelle smartly saw O's prospects for upward mobility and reeled him in

I have no idea what your first sentence even means. When you sample the set of hysterical hatred for Obama, do you see a lot of 'he's too white acting' in their rhetoric. From Dinesh D'Souza's "take the boy out of the ghetto" to Trump's birtherism, a good amount of shade people threw at Obama was literally about his shade.

I think the marriage was 1) political convenience; you need a wife and kids for the photo ops and/or 2) Michelle smartly saw O's prospects for upward mobility and reeled him in

1. Most people do not make inconvenience a goal when getting married. When you chat with, say, a doctor do they brag about how their spouse makes it so much harder to run their practice?

2. I suspect if Obama had married a white woman, you would be patting yourself on the back for 'smartly' observing that he had carefully targeted his wife to snag white and black voters.

3. Women, and men, are attracted to people who appear to be on a successful track. I think your observation is banal pretending to be smart. If you were discussing Ted Cruz's wife, or Jordan Peterson's, or Alex Jones for that matter, you could make the exact same statement and it would be so dull and uninteresting the massive MarginalRevolution server farm would struggle to host the comment without falling into sleep mode.

Jordan Peterson and his wife were sweethearts since age 10 or something like that. I have no idea what Alex Jones is up to and only know that Ted Cruz' wife is a successful corporate type.

My first sentence means what it says. I never thought of Obama in terms of all the First Black President stuff lavished on him. He's just a typical college-campus type liberal; skin color not crucial.

So "white ivory-tower elite" meant a white tower rather than the color of the elite?

Obama's always struck me as about halfway along the line segment which runs from 'deputy dean of students' to 'local TV newscaster'. Not elite. He seems to have an ordinary companionate marriage, though I think there is some surviving correspondence or oral accounts which indicates that his interest in black women ca 1990 was contrived. He was a law student when Mooch met him and a recent degree recipient when he married her. Pat Buchanan has offered the view that she was rather dissatisfied with public life and I think there are reliable accounts indicating she was hoping ca. 2000 that he would land the presidency of the Joyce Foundation and get out of elective office. The smart money says Mooch just wanted someone who would nestle in the professional-managerial bourgeoisie and wasn't ambitious beyond that. People who know Heidi Cruz have said 'she has five-year plans, ten-year plans'. At the same time, it's a matter of public record that her husband's public career is something she accedes to, not something she likes.

Yeah, it's really just an example. How often do you think mixed race "black" men have the chance to marry anyone of any race? 90% of the time people are going to think of them as "black" and their dating options will be limited to mostly black women. Ok Cupid actually published results on this.

Per Pew, about 1/4 of black men are marrying out Hazel, FWIW, per Pew.

Hazel, people marry endogomously because they actually like people of their own kind. For small population segments, endogamy is essential to the community's continuing.

Well roughly speaking if blacks make up about 12% of the population then 88% of black men would 'marry out' if everyone just married people at random. So if you saying 25% of black men 'marry out' is also like saying 75% of them are 'marrying in' when chance would say only 12% should.

Strange as it may seem to you, Boonton, people on the marriage market favor what is familiar to them.

What's interesting here is the recognition that there are career/reputational incentives for bias in all organizations.

Valiant effort. The propagation of uncertainty is not apparent to the casual observer.

"The number of discrimination taxes multiplies, at each level. Just like the medieval barons put too many tolls on the river. All of a sudden the initially mild statistical discrimination isn’t so mild any more, due to it being applied at so many veto-relevant levels."

Wouldn't you also expect diminishing effects the further you were from the "source" of the discrimination? If each toll was half the size of the preceding toll perhaps the effect wouldnt be that large at all

True but that's not how tolls worked or work. Each toll taker is going to be inclined to take a roughly equal toll.

If the guy who decides who gets the mail room job has a 1% bias in his hiring process, that's no reason to think the guy who decides who gets promoted out of the mail room will have a 0.5% bias and the guy deciding the next level up would have a 0.25% bias and so on. More likely each one is likely to have the same 1% bias. Maybe near the top where the stakes are very high and the process is under a lot of scrutiny will the investments be made to cut down on even that small element of bias in the name of making the best decision possible. But the point here is that won't do much good if the candidates who make it to the last gate already ran through dozens of tiny biases.

But maybe it is how hiring works.

The mail room hirer has nothing to go on, so relies on statistics. He has an X% bias. That error goes both ways.

At the next level up, that hirer has a pool of candidates that were filtered by the mail room guy. The candidate in the pool at this level from the statistically less fit class . . . contains more Shannon information by virtue of statistics just by being there. Any risk averse hirer at this level should favor that candidate, ceteris paribus.

So, the bias is diminished, simply by the numbers.

So it looks like Rand Paul is going to try to be a Trumpian president in a few years.


Good luck Bayesian-moralist-libertarians, as you try to find a piece of ground to stand upon.

So it looks like Rand Paul is going to try to be a Trumpian president in a few years.

No, it doesn't, but pretending it does is the hook you need in order to say something snotty.

"Kentucky Sen. Rand Paul said he can’t imagine supporting anyone other than President Trump in the 2020 election, but paused for a long time when asked if a Republican should challenge Trump in the primary."

I am not sure which is worse though, to be honest.

Wow, actually.


I understand that there are principled libertarians out there, who haven't bizarrely decided that Putin's kleptocracy is a model for our future.

Where are you going to go now?

This is obvious. At every level someone has to make a bet. Will the person I hire, mentor, advance, delegate to meet expectations?

It is impossible to know beforehand. All you can do is look at track record, your experience, recommendations from others. Then you make a decision, are wrong lots of times as people disappoint. The higher the stakes, the less likely you will bet on someone who you are less sure of.

The solution? There isn't any, just as there isn't any surefire way to win bets at a casino.

Don't call this injustice. The universe is conspiring to separate each one of us into various discrete molecules to spread around randomly. Living and being conscious for 70-80 years is an unbelievable fluke, not worth wasting any moment on how hard done by you are.

Comparing it to accumulated taxes is ridiculous. Taxes take away. Getting opportunities is additive. Someone taking a chance on you is a gift, a benefit to you. Not getting that takes nothing away from you.

Could go either direction. If you benefit from a small bias then it can turn into a huge benefit after many layers.

Suppose left handed people are better 52.5% of the time but people, being bad at math, mentally round that to 53%. Going through a single decision point means a small benefit to a left handed person and a small loss to a right handed one. Going through 20 layers, though, those 1/2 percents start adding up.

There are some resemblances here with what Robert K. Merton called the "Matthew Effect" from the New Testament: "For whosoever hath, to him shall be given, and he shall have abundance: but whosoever hath not, from him shall be taken away even that which he hath." Matthew 13:12 KJV

That would mean that there is value being left on the table. A smart person will pick up that value and get higher quality people per dollar and win in the long run.

Not if your customers are the ones who are biased, and they know who you are hiring.

'Bias' on the part of clients is not why your managers aren't allowing you client contact.

No. That would be because I'm a goof and a dork. But in general I would think MOST women would actually present rather well to clients. Though there does not seem to be a large overlap between the sort of women who wear dresses and high heels and those who excel in technical fields.

Possibly not. Let's say we are talking about a higher position like CEO of a Fortune 500 company. Let's say it takes passing thru 100 'gates' to get there. If a small bias leaves some better candidates behind at early gates, there is value left on the 'table' but that table is in an alternative universe. If you're the Board of Directors at final gate trying to find the best CEO possible, it doesn't help that person may have got left behind at gate 53. The best CEO needs to be the best intrinsically AND pass thru all 99 gates....just like an NFL quaterback needs inherent physical quality AND the training from playing football for years. It does nothing for an NFL team owner to know that Gus who runs the Enterprise dealership down the street would have been the world's greatest quarterback if only his high school didn't drop football after his Freshman year!

That’s an argument for removing all racial/ethnic information from the processes. Classical music instituted blind auditions.

Let the chips fall where they may.

It seems to me that a rational firm would use blind audition if *possible*. What should you do if it is not possible? (i.e. you may not be able to test the potential employee for certain traits, yet you may know that it is statistically more probably between men)

But also a lot has happened by the time someone gets skilled and committed enough for any level of audition

Reminds me of a Schelling paper in which he argued housing segregation as the compounding and therefore drastic effect in equilibrium when a few individuals have a minor preference for living amidst people of their own race.

That's self-correcting, though. A legitimate NBA prospect sticks out like a sore thumb at a minor school.

Recall that Scottie Pippen went to Arkansas Pine Bluff. The NBA found him just fine.

And Stephen Curry went to Davidson

But that assumes a constant tiered "acquisition" model. If there's one thing we've learned is that the cumulative costs/choice errors of tiered distribution expose it to abrupt disintermediation. The example in sports I suppose would be the East Germans in women's sports. Unfortunately the test is damaged by the abuse of steroids. But perhaps the test could still be evaluated using German women's Olympic sports right after the fall when the best East German women were chosen but denied their steroid? Or look at other non traditional women's sports when alternative sourcing strategies have been pursued.

What about late bloomers? People here are discussing various biases but how about age -- if you don't pass through certain filters and gateways by an appropriate age you are not going to be considered a good investment.

You might develop or discover a skill relatively late in life and maybe capitalize on it as an individual entrepreneur but larger institutions are not going to look at you.

At first glance this seems like a good argument for affirmative action. The standard free market response is not going to work here because where it counts is at the top of the chain. But at the top of the chain the damage has already been done and anything invested to squeeze small biases out of the system will only do a marginal amount of good.

Example...let's just say that the 'truth' is one group may just not do as well in a big corporate job as another group. To keep the alt-right commentators here placated, let's just say blacks do 10% less well as whites, all things being equal. What that means is for a given job, given background blacks will do 10% less well as white....so if we are talking entry level CPA positions where the candidates scored 79% on the CPA exam, the idealized hiring manager would essentially randomly weight his hiring so that 10 whites are hired for every 9 blacks. If the 'truth' was that there was no difference then the hiring manager would flip a coin with every candidate who scored 79% on the exam and we'd expect the ratio to be 10-10.

BUT here's the problem, this type of fine tuning is not really something humans are good at. If the hiring manager is a bit biased because he read too many Steve Sailor posts and operates at an assumption blacks will do 15% worse, his hiring will be slightly skewed. It will be sub-optimal but not in a way the free market will help much. An accounting firm has lots of accountants and a firm that is perfectly optimized in its hiring of rookies is not going to have a huge advantage over one that's almost optimized.

From rookie CPA to junior CPA, to senior, to manager, Sr. Manager and up the line say this slight bias infects hiring/promotions managers. Near the top, though, being sub-optimal makes a difference. When it comes time to hire a new CFO, for example, there's a lot at stake so the selection team is much more professional, much more rigorous. More likely to stamp out the biases made by lesser hiring managers. But at that point the damage is done. The selection of the CFO is going to be made from the pool of those with a lot of experience and success already booked.

One response may be an argument for a broad affirmative action plan. If blacks are given a 5% boost in the bottom 80% of positions (on the assumption the top positions will be more rigorous in their hiring) , then you've countered the bias. BUT I see two problems with this solution:

1. How to quantify the bias so precisely? How are we going to know the bias is the same for all levels and that it doesn't change over time?

2. Hiring managers may assume the 'boost' and offset it by biasing their decisions against blacks by 20% instead of 15%.

I would propose then that the target of the affirmative action should come in the form of "many roads lead to Rome" rather than one road. In other words, any one metric or basket of metrics used to make these decisions is likely to embed these small biases that are really impossible to smooth out. There should be affirmative action in providing for alternative metrics.

In other words, rather than hiring 100% based on the top test score, hire 75% based on higher score and 25% based on low scores. Or hire 90% based on who passed the CPA exam with the highest score, and 10% based on people who do private bookkeeping who never took an accounting class. For 90% count a past criminal record as a negative, but for 10% count it as a positive After all, Breaking Bad and Better Call Saul does demonstrate the population of outlaws often contains a few geniuses who can run circles around the dull masses who always play by the book.

Anyway it seems to me the optimal strategy both from a fairness perspective and an optimization perspective would be to purposefully have several tracks into an organization.

This would not be the same as 'class based affirmative action'...like giving extra 'points' for coming from a bad high school. It would have to be a totally different track. If you're the best in one track, by definition you couldn't also be competitive in the other.

OTOH, the tax falls on the institutions as well -- taxing institutions would tend fail to find the best candidate, because sometimes that candidate is an outlier. Understanding this is the case, they would consciously avoid imposing the tax even in the absence of legal frameworks intended to force them to (e.g. by making bias awareness a common theme in the training of such decision-makers).

And note the tax is cumulative on institutions as well -- every level of taxation leaves their available choices more impacted by prior bias.

This seems a reasonable explanation for some of the glaringly incompetent URMs in the corporate world and academia. Affirmative action (or virtual signaling) in college, then in the first job, and so on, until you have someone who can barely speak (never mind think) at a high level position in a major corporation or university. Happens way more often than could be explained by chance.

Comments for this post are closed