Unintended Consequences of Information Bans

Luke Froeb at Managerial Econ covers a number of cases in which information bans have led to unintended and negative consequences:

When I was at the Bureau of Economics at the FTC, we were asked by Congress whether using credit histories to price car insurance was discriminatory.  The resulting FACTA report found that:

  1. as a group, African-Americans and Hispanics tend to have lower scores than non-Hispanic whites and Asians.
  2. …scores effectively predict risk of claims within racial and ethnic groups.
  3. The Commission could not develop an alternative scoring model that would continue to predict risk effectively, yet decrease the differences in scores among racial and ethnic groups.

As a result, banning the use of credit scores would result in insurers finding other, less good and possibly discriminatory methods of distinguishing high from low risks, like selling insurance only in low risk areas.  Good drivers living in higher risk areas would be “pooled” with other drivers living in the high risk area, and would have to pay higher rates.

Previous studies (here and here) finds an analogous effect of preventing criminal background checks in employment, that doing so increases racial discrimination against African American men.

Adam Smith’s words are evergreen:

The man of system, on the contrary, is apt to be very wise in his own conceit…He seems to imagine that he can arrange the different members of a great society with as much ease as the hand arranges the different pieces upon a chess-board. He does not consider that the pieces upon the chess-board have no other principle of motion besides that which the hand impresses upon them; but that, in the great chess-board of human society, every single piece has a principle of motion of its own, altogether different from that which the legislature might choose to impress upon it. If those two principles coincide and act in the same direction, the game of human society will go on easily and harmoniously, and is very likely to be happy and successful. If they are opposite or different, the game will go on miserably, and the society must be at all times in the highest degree of disorder.


Bravo, Alex. Unfortunately this kind of clear thinking is all too rare. Russ Roberts' interview with Neil Monnery about his biography of John Cowperthwaite shows a man who consistently and effective thought through questions of public policy.

+1 to the chess reference by AlexT, which is usually TC's domain.

The very question of whether using credit histories is discriminatory belies the definition of discrimination.

Discrimination consists of adverse actions against people BECAUSE OF their protected classes, not merely in spite of them. That is, one must show that using credit reports is pretextual for the purpose discrimination and without any justifiable business purpose. The fact that minorities have worse credit reflects their economic and financial behavior and conditions which adversely affect future credit decisions. It has nothing to do with their race.

The article makes a different point which is interesting and informative, but we really never should have been asking the question. Too many people wrongly assume that every adverse outcome to minorities is the result of racism or discrimination.

It will be interesting to see if the Supreme Court takes up the rationale of disparate impact. If you think Roe v Wade would be contentious, this would be up there.

This is exactly right.
Disparate impact aren't automatically racist. They can be if the policy itself is race-dependent or the process that leads to it's adoption is racially biased, but if they arise spontaneously from some process that doesn't take race into account they wouldn't be.

"That is, one must show that using credit reports is pretextual for the purpose discrimination and without any justifiable business purpose."

My understanding is that "disparate impact" precedent does not require proof of intent. However, proving "business necessity" is a sufficient defense.

That is, in employment law at least, if a selection criterion can be shown to produce a significant adverse disparate impact on a protected class then the burden of proof is on the employer to show that the criterion is significantly related to the job requirements, BUT, showing an absence of intent to discriminate is not sufficient.

This can get even more complicated in other areas that the "disparate impact" analysis applies, such as voting rights. In some cases, even if you are choosing the criterion for business reasons, policy, or some non-animus related reason, mere awareness that the chosen criterion might or tends to produce a disparate impact on certain protected classes, what others call a "discriminatory outcome", is sufficient to find a civil rights violation for discrimination. With wall-to-wall media coverage on these issues, it can be hard for someone to claim that they are "unaware". Take, for instance, the finding that in SV tech companies, when they switched to gender-blind hiring policies, the result was (unexpectedly) that more men were hired than women. Since this finding has been well publicized, does that now mean that gender-blind hiring policies in tech would be prima facie discriminatory because of their known tendency to produce greater numbers of male hires? That seems like an absurd result.

Disparate impact theory made sense back in the 60s and 70s and even sort of into the 80s, when racists could easily smuggle pretextual justifications into hiring/admissions/etc policies, but the theory is starting to reach its limits in terms of viability due to cultural, economic, and technological changes that are producing massive contradictions within the idea itself.

One adverse outcome today that was from discrimination was redlining in the 50s to 70s in places like Palo Alto. If you are black, the amount of equity you have to gift to your kids or otherwise monetize is tiny compared to the neighborhoods they were banned from. This is no tiny disadvantage when you want to buy books or send a kid to college.

Some people think that there should be a lending policy that helps those who were specifically left behind in this manner so they have less relative disadvantage. (Not my best grammar but you get the idea)

I can understand why we want to formally ban explicitly race-aware discrimination even when it is a rational response to a bayesian prediction system. Unfortunately, nobody is ever willing to make this argument formally because they are afraid of acknowledging the truth that racial discrimination could be rational in this way. Rather they want to deny that group differences exist, but then that leads us in circles with disparate impact claims with policies based on actual merit like this one. If you want to advance a disparate impact philosophy you have to be prepared to argue that racial discrimination is such a problem that we should be prepared to throw out valid policies in the concern that they have a hidden racists agenda that cannot easily be seen from normal formal analysis. (Written tests for firemen for example)

It's all about trade-offs.

On one hand, the difference in credit scores is most certainly due to long-run cultural and institutional racism (and I hold that belief until someone comes up with a clear link between genes expressing melanin and their impact on other traits). I'd be a fool to believe that my good credit score is due to solely to some personal will deep within me. It's a combination of favorable extended family finances stretching multiple generations, direct financial education from parents at a young age, good schooling, geographical luck, genetic luck, economic luck, and perhaps a smidge of good decisions.

Not saying there is a better alternative, but you have to ask yourself...do you want to live in a society where your success/failure is so heavily predicated on what happens outside of an individual's control?

I offer no good solutions. There are none.

I think the key here is hidden in the phrasing "scores effectively predict risk of claims _within_ racial and ethnic groups"

That is, instead of making a naive comparison of a black person't credit score verses all other credit scores, they're comparing a black person's credit score to other black peoples' credit scores. So, the overall level of credit scores isn't what's driving the actuarial decision, it's where the credit score lies in relation to credit scores of people of the same race.

At least, that's the way I read it. And doing the comparison this way effectively removes race from the equation, and thus shouldn't be driven by all the unobservable variables that lead people of color to have lower credit scores on average than white people.

If that is how it is done...then yes, I would agree it removes much of the issues.

"and perhaps a smidge of good decisions." That smidge is probably in the high 90s percentage wise.

But those good decisions were driven by historical actions taken by others and genetic luck. What percentage are truly "mine"?

"the difference in credit scores is most certainly due to long-run cultural and institutional racism (and I hold that belief until someone comes up with a clear link between genes expressing melanin and their impact on other traits)"

That's incredibly ignorant. You attribute all socioeconomic differences between races to racism? I suppose that means people are racist against whites in favor of Asians? Same objection applies to your "melanin" comment since Asians have more of that.

What matters is whether credit scores accurately price claim risk, not what causes someone's credit score.

Whoa. I do not attribute all socioeconomic differences between races to racism. I do attribute all* (~99.99999%) of the socioeconomic difference between races to things that have nothing to do with their "race". It has to do with a soup of a million factors, of which recent historical racism is one component. But things like geography, weather, spread of technological innovation, etc...

"you have to ask yourself...do you want to live in a society where your success/failure is so heavily predicated on what happens outside of an individual's control?"

The question is actually whether your car insurance premium should accurately reflect your risk for getting in a car accident or whether other people should subsidize you or you should subsidize others for some reason.

I want car insurance premium to accurately reflect my risk, adjusting for the things that are within my control. Of course, that's unreasonable, but that should be the North Star goal.

File this under "obvious to actuaries". Even progressive actuaries can figure this one out.

Free markets have disparate impacts -> regulation. Regulation has disparate impacts -> socialism. Socialism has disparate impacts -> genocide (e.g. Holodomor). That's how you end disparate impacts. Sure, genocide has disparate impacts, but only if you don't go all the way.

Another final solution for the disparate impact problem: separating into ethno-states. Within each ethno-state, at least, there would be no disparate impact.

Scores (I assume credit scores) effectively predict risk of claims within racial and ethnic groups. Effectively predict? What does that mean? Are we talking correlation? I would guess that poor people usually live in bad, unsafe neighborhoods and, hence, suffer losses (from theft, vandalism, etc. of the auto) at higher rates than non-poor people who live in nice, safe neighborhoods. Likewise, poor people often have poor credit histories because they are the first to be let go in down cycles. Cancer (or diabetes, heart disease, etc.) "effectively predicts" risk of claims for health insurance. Poor health is not a choice, but neither is being poor. I suspect the assumption here is that a bad credit history reflects bad behavior - the bastard won't pay what he owes. When the founders met in Philadelphia they were not the revolutionaries they have been depicted, but the nation's elite (i.e., creditors) concerned with protecting creditor's rights, as debtor relief measures were being passed in the states. Of course, we make distinctions between wealthy dead-beat debtors and poor dead-beat debtors. How so? Why do you think Rick Scott (and other wealthy people at risk of judgments) moved to Florida? Sunshine? Florida has more dead-beat wealthy debtor protection laws than Florida has sunny days.

We call it "asset protection planning", an innocent sounding phrase. Just protecting widows and orphans. In America we expect much from poor people, but we have very low standards for rich people.

As long as the scores themselves do not automatically discount someone due to their racial background, I don't see the problem. If you eliminate the score, people are going to start using race itself as an indicator instead, which is what we don't want. We want people to use objective measures, not ones that are race-dependent.
If a black or Hispanic person has an actually good credit record, they shouldn't get discounted just because they are black or Hispanic.

That said what we should do is be careful to make sure that the formula used to create the score doesn't itself use race as an indicator of creditworthiness.

There's a difference between African Americans having lower credit scores because of objective factors like poverty that indicate higher risk, and African-Americans having lower credit scores because the credit agencies use race as a marker for credit worthiness. Whether or not that is an accurate predictor in aggregate, the latter is going to be unfair to African Americans who otherwise have good credit histories as individuals.

Agreed. Any move to make the score or credit determination from the individual to a group is clearly wrong. I don't care what the group outcomes are as long as they are fair. Individual scores make it possible for people to modify their behavior to get better scores. Judge me by my behaviour, not the color of my skin.

"Whether or not that is an accurate predictor in aggregate, the latter is going to be unfair to African Americans who otherwise have good credit histories as individuals."

How can it be unfair if it is an accurate predictor?

It's only accurate in the average, and it's possible to measure credit-worthiness directly, not by guessing based on unrelated factors like race. if it's possible to take a direct measure of X, then you shouldn't judge people based on Y, just because of a statistical correlation between X and Y that may or not hold for a particular individual.

Sometimes, there is this feeling of time passing. No insurance company used credit scores in the 80s, they used age/experience, model of car, driving record, etc - and were also allowed to share data in a way that was allowed with an explicit carve out to anti-monopoly laws.

These days, it seems as if the game is completely different - someone I know in the U.S. rear ended a car a couple of months ago. He reported the accident to his insurance company, and when he received his new insurance bill a couple of weeks ago, his rate was almost a hundred dollars less.

Simply bizarre, but then, time passing makes many things seem that way.

Someone must have poked Alex with a stick. Two posts in a row.

Comments for this post are closed