The Incarcerating Power of Stereotypical Beliefs

by Scott Page on August 14, 2019

Freeman testified at trial that Abrahams was trying to stop him and Willingham from fighting, and that she hit Willingham in the face. Willingham, Freeman testified, then raised his hand as if he was going to hit Abrahams, but then lowered it to his waist, which prompted Freeman to believe he had a gun. Freeman shot Willingham three times, according to police.

-Detroit Free Press June 20, 2019

America is very good at some things and very bad at others.  We produce a steady flow of scientific, cultural, and engineering successes– in the past five years we can lay claim to a drug that cures hepatitis C, Hamilton: The Musical, and self-driving cars.  And yet, we perform just this side of miserable at addressing core social and economic challenges.  We are among the world’s leaders in income inequality, obesity rates, drug overdose deaths, and per capita greenhouse gas emissions.  And, we rank first, that is worst, in the proportion of citizens in jail at about seven people per thousand.  Even more troubling, blacks are six times as likely as whites to be behind bars.

Why is this so?  And what is to be done?  In their exceptional book, Shadows of Doubt: Stereotypes, Crime, and the Pursuit of Justice, economists Brendan O’ Flaherty and Rajiv Sethi examine our troubled, racist criminal justice system with depth, maturity, pragmatism, and focus.

The book combines thoughtful, balanced analyses of recent events – the shootings of Trayvon Martin, Tamir Rice, and Philando Castile – with innovative social science thinking.  The authors methodological range extends from econometric considerations of causality and selection bias to philosophical considerations on the value of integration.  Glenn Loury describes the book as technically impressive and morally serious. Danielle Allen says it should be it required reading for policy makers and citizens alike.   For these reasons and more, this summer professors from Philly to Phoenix restructured their fall course syllabi on criminal justice, race, and social justice with this book as a centerpiece.  But this isn’t just a book for academics.  Anyone involved within the criminal justice system – from front line police officers, to judges, to politicians – would be enlightened by this book.

Any analysis of race and crime must cut through the complexity.  O’Flaherty and Sethi accomplish this by separating out the key crimes:  robbery, murder, larceny, etc…  This restriction enables them to make comprehensible the hairball of data relating race and crime.  That act alone creates an invaluable resource.  They then construct a theoretical framework that reveals the extent to which our behaviors can be traced to stereotypical beliefs.  Stereotypes prompted James Freeman to believe Deandr’e Willingham had a gun.

That belief had lethal consequences. James shot.  Three times. Deandr’e is now dead.  James will spend the next quarter to a half century in jail.

The Willingham shooting, like too many others, resulted from a stereotype driven behavior.  These behaviors impact rates of crime, whether people testify, who police stop and how judges sentence.  And, these actions all interconnect to produce rampant racial bias.  The murder rate is connected to the reluctance to testify which is connected to the fear of retribution much like the shin bone is connected to the ankle bone and the ankle bone connected to the heel bone.  A decision to shoot first depends on a belief that the other person will shoot, on an expected probability of being caught, and on a belief that witnesses will testify.

If stereotypes are the cause, why don’t we just eradicate them? Stereotypes arise, in part, because they must.  They belong to a broader category of cognitive attention biases which arise because we simply cannot pay attention to all of the particulars.  We take shortcuts. We bin people into categories.  We lump to live.  That lumping may be the result of rational calculations – it’s not worth our time to consider every particular (rational inattention bias). It may be that we lack the cognitive capacity (behavioral attention bias) or the time or experience to draw precise inferences (categorical cognition bias).   Regardless of the cause, we could not navigate the world without categorizing reality and therefore stereotyping.

When threatened, our lumping becomes even more crude.  Is a person approaching us on the street a threat to our safety?  It’s a question that we all have asked.  If we could take all of a person’s characteristics into account, we could (in theory) accurately predict if the person poses a threat, and act without bias.    But on the street, we have to act fast, we lack full information, and our emotions hinder our ability to think clearly. Herein lie the seeds of our racial incarceration disparities. In threatening situations, we lack the luxury of thinking through a hundred visions and revisions.   We fall back on stereotypes.  We classify people by their race.  We then construct beliefs based on racial stereotypes, and rely on those beliefs to guide our actions.   We believe he has a gun.  We shoot three times.

The Willingham case is just one example. O’Flaherty and Sethi provide an empirically grounded chapter by chapter account of the behavioral implications of a range of stereotypical beliefs, which they ground in data. Whites are more likely to hand over their wallets but carry less money.   Blacks are more likely to be stopped and searched by police.  Murderers of whites are more likely to be prosecuted.  Black witnesses are less likely to testify, and so on.

Each bias on its own adds to racial disparities along a dimension: arrests, shootings, incarceration rates, etc…  The biases also accumulate because each potential criminal interaction consists of a sequence of belief based actions.  In revealing how linked sequences of racial biased actions accumulate, the authors make an original and important contribution. Each stereotypical belief can ring mostly true, each action can be rational (given beliefs), yet by adding belief upon belief and action upon action, the results can be tragic.

Consider the following two scenarios.  In the first, a homeowner with a registered handgun hears a noise downstairs in the middle of the night.  He grabs his gun and encounters a trespasser.  The homeowner must decide whether or not to shoot.   That choice depends on the homeowner’s beliefs: will the trespasser flee?  Will he turn and shoot? The homeowner faces risks to shooting – the trespasser could be the neighbor’s son who left his iPhone earlier in the evening.

The Castle doctrine, the legal doctrine that a person’s home is his castle, gives the homeowner the legal right to defend it.  This doctrine makes the homeowner more dangerous to a genuine intruder.  Thus, an intruder who believes a homeowner is going to shoot will be more likely to shot first instead of fleeing.  That belief in turn puts the homeowner at greater risk.  A homeowner who believes that a trespasser is not his neighbor’s son, who believes the trespasser is likely to shoot, and who believes he will not be arrested if he shoots the trespasser first, will likely shoot.  It stands to (stereotypical) reason, that if the intruder sees the homeowner reach for anything (a light switch, flashlight, or cellphone), the intruder will fire away.

Using this logic, O’Flaherty and Sethi argue that the Castle doctrine, though intended to make people feel safe,  contributes to higher homicide rates as do stand-your-ground laws which lower the risk of prosecution for those who shoot when they could safely retreat.  The evidence supports them.

In the second scenario, imagine two people involved in a heated argument in a bar who are escorted out to the parking lot.  One of the participants has a gun but he is unaware if his fellow combatant is armed.  The gun holder faces the same choice as the homeowner: shoot or go home.   His action depends on two beliefs:  the odds the other person will shoot and the odds that shooting first will result in prosecution.

Racial stereotypes affect outcomes in this situation as well.  If the combatants are white, the gun holder will believe (correctly) that shooting first will almost surely result in prosecution if the other combatant does not have a gun. Thus, unless he has a strong belief that the other party is armed, he will not shoot.

If, on the other hand, the combatants are black, the gun holder might stereotypically believe, like James Freeman did, that his fellow combatant is armed.   Moreover, if the first person might believe that if he does shoot, he can escape prosecution. That belief would be logical if he believes that onlookers will fail to testify.   That belief can be accurate if people in the neighborhood distrust the criminal justice system.  This concatenation of stereotypical beliefs results in a black gun holder being more likely to shoot a black person than a white gun holder shooting a white person.   The beliefs prove self-reinforcing:  black gun holders who shoot bolster the belief that a black person is more likely to shoot.

What then is to be done? Changing behavior requires changing beliefs.  Ridding ourselves of the worst stereotypes will take time and meaningful interactions across racial groups. Reducing stereotypical thinking becomes easier as society becomes more integrated, as we see experience our common humanity. O’Flaherty and Sethi also advocate better police community relations – more meaningful engagement and less stop-and-frisk. Police behaviors and protocols that increase trust can change beliefs and improve the likelihood of solving homicides with black victims.  That in turn reduces the incentives to commit crimes.

They bravely suggest that people act in ways that guide the beliefs of others. They do not demand that everyone whistle Vivaldi, only that we proceed from positions of trust and kindness and find common ground.  There exists a lot of space between acting white and acting threatening.   We need, as a society, to explore and better understand that space.

They also sign on to some novel policies worth our attention such as allegation escrows, an idea initially proposed by Ian Ayres.  Allegation escrows allow people to provide eyewitness accounts anonymously with the assurance that they will only be called upon if other witnesses also come forward.

As a first step, O’Flaherty and Sethi advocate self-awareness of our stereotypical reasoning and how biases build sequentially from our thigh bones all the way to the bracelet encircling our ankle bones. That’s sound advice.  What a society believes of a person influences that person’s life course. Belief in the goodness of others, like the belief that that another person will shoot, is self-reinforcing.   And, therein lies the hope provided by this outstanding book.

Scott E Page

Stephen M Ross School of Business, University of Michigan

Santa Fe Institute



Phil 08.14.19 at 8:42 am

Consider the following two scenarios. In the first, a homeowner with a registered handgun

It’s interesting the things we think we can change and the things we think we can’t.


oldster 08.14.19 at 12:38 pm

typo here:

“Moreover, if the first person might believe that if he does shoot, he can escape prosecution.”

either a superfluous “if,” or some missing prose.


Lupita 08.14.19 at 2:38 pm

Another way to approach this problem is for the US Census Bureau to ditch the “What is your race?” question since races do not exist, that way, professors and journalists would not have the statistics to write about race, reach conclusions, and feed more stereotypes to the public. The race quackery begins with the state which legitimizes the notion of human races by asking people to state their race at every turn.

The states that ask their citizens to categorize themselves into non-existent races are mainly the anglophone ones: the US, UK, Canada, and New Zealand. Brazil and Colombia ask it too, but that is only because the US, through a UN agency, promoted this categorization worldwide, just like it promoted independent central banks through the IMF.

Any analysis of the effects of 19th century race theories on 21st century societies must consider that most countries do not have the statistics to analyze their societies and societal problems by “race”. The alternative is, of course, class analysis but that crashes into another great American prejudice, supported by universities and the media, which is anti-socialism.


Cervantes 08.14.19 at 3:45 pm

Err, I’m afraid we can’t lay claim to self-driving cars. They do not yet exist and I wouldn’t bet on it any time soon.


Mike Huben 08.14.19 at 7:26 pm


Race quackery long precedes modern states, and has been adopted by states from folk taxonomy Folk taxonomies arise independent of states in every culture. Like all such social constructs, they are non-existent, but are shared because people find them useful: for example for association or prejudices.

Class is another such non-existent social construct. It is yet another dimension which has a set of stereotypical beliefs

When confronting stereotypical beliefs, we have to recognize the social constructs in use. Switching to alternative constructs is not the solution, though recognizing that more than one set of constructs is used can be valuable.



Sam Tobin-Hochstadt 08.14.19 at 8:33 pm

Cervantes, fully self-driving cars are unlikely to be widely used in the US soon; that’s certainly true. But in the sense that people have built cars that drive themselves with no human intervention, they certainly exist and work and drive on some roads. It’s just that they have a bunch of limitations that make them not generally viable for replacing humans. It’s a similar situation to machine translation: it really exists and works, but isn’t nearly as good as humans especially in hard cases.


Catchling 08.14.19 at 10:18 pm

Money is a mere social construct. This does not mean that problems like inequality, poverty, taxation, or debt are solved by any given person or institution “deciding” to ignore it for being non-existent.


J-D 08.14.19 at 10:40 pm

Mike Huben

Like all such social constructs, they are non-existent …

Non-existent? Of course social constructs exist!

Some of the things that (some) people believe about (some) social constructs are not true, but that’s not equivalent to being non-existent. In the case of race, for example, many people believe that it is a biological category. It’s not a biological category, it’s a sociological category, but it makes no sense to say that biological categories really exist but sociological categories really don’t.

Money is a social construct, but just because it’s a social construct is no reason to say it’s non-existent. That would just be silly. So too with race. In both cases what’s important is to point out how people’s beliefs about them (or some of those beliefs) are false and harmful, not to pretend that the constructs don’t exist.

In my country (and in many others), there is a census question about country of birth. (Countries are also social constructs, but not non-existent!) Discrimination based on country of birth existed long before there were census questions like that. If it were possible to remove that kind of question from censuses, would not eliminate or reduce that kind of discrimination. Likewise, taking the question about race out of the US census would not eliminate or reduce racism in the US.


David J. Littleboy 08.15.19 at 1:36 am

Chicago O’Hare has had a self-driving train for decades. Yokohama also, although it had a glitch (causing a minor crash with a few minor injuries) the other day, and despite being one of the more uninspired trains in the Tokyo metropolitan area (only 5 cars instead of 10 or 15), that glitch caused massive disruption for an enormous numbers of commuters. Even a 5-car train is a truly beautiful thing compared to busses.

But the idea that self-driving cars are going to be anything other than limited-scenario demonstrations in our lifetimes is problematic. The current ideas in AI are basically things that were proposed in the 1950s and 60s and shown to be inadequate back then. For example, recent attempts at asking how image recognition (using regular arrays of locally connected computational elements*) actually works found that the systems recognize textures, not shapes. Which explains why they miss elephants in the room (really!) and are easily confused by noise. (This was essentially Minsky and Pappert’s criticism of early neural network models.)

Basically, current AI isn’t up to the task, and isn’t doing anything that might change that. AI that’s up to the task is several generations away.

Roger Schank was recently asked about current AI, and he said “There’s no such thing as AI”. While this is, of course, easy for someone having an all-but-thesis from Schank, to agree with, I really do think it’s correct. Minsky and Schank made serious attempts at asking what human intelligence is. Current AI doesn’t. (FWIW, I think human intelligence is amazing beyond words, and that Kurzweil and friends are missing the boat: the singularity has already arrived and we is it.)

Bottom line: don’t invest your life savings in AI.

*: The point of this snark is that “neural networks” have almost no similarities whatsoever with actual neurons. For some reason, computer nerds (myself included) have an excessive fondness for parallel models and think interesting behavior will emerge from them magically just by letting them run free.


Person_XYZ 08.15.19 at 6:50 am

“Is a person approaching us on the street a threat to our safety? It’s a question that we all have asked. If we could take all of a person’s characteristics into account, we could (in theory) accurately predict if the person poses a threat, and act without bias. But on the street, we have to act fast, we lack full information, and our emotions hinder our ability to think clearly. Herein lie the seeds of our racial incarceration disparities. In threatening situations, we lack the luxury of thinking through a hundred visions and revisions.”

This is a completely lucid description of the problems people have everyday. We don’t know if people have criminal or nefarious intentions so we have to fall back on intuition, common sense and ‘stereotypical beliefs’. One stereotypical belief is that ‘men are more likely to be violent criminals than women’. Is this belief wrong? Is it foolish to factor it into account when one is out and about late at night, lacking deep knowledge of strangers?


divelly 08.15.19 at 1:40 pm

I participated in a telephone survey once upon a time from the U of Alabama!
The interviewer had a lovely Southern Belle lilt.
At the end, I agreed to answer some demographic questions.
“Which race do you consider yourself? Are you…?”
I interrupted her and replied,”Human.”
There was a long pause before the next question.
“Which Religion…”
“Pagan.” I said.
A longer pause.
“Well Thank You Sir.”
And she hung up.


Lupita 08.15.19 at 3:57 pm

taking the question about race out of the US census would not eliminate or reduce racism in the US.

I think it could reduce it somewhat; at the very least, it is a step in the right direction. For example, in the US, people have official documents, such as birth certificates, that state the race of the person, making a person’s supposed race an official and legal designation and racialism a state ideology, not just a folk taxonomy.

In Mexico, for example, the state does not quantify “race”. Neither are there race caucuses in congress, talk about the essence of “brownness”, politicians who claim to represent all the people who use their same shade of makeup, nor political correctness. I do see a connection between state ideology and the cultural and political expressions within that state.

In Latin America, we use the concept of “pigmentocracy” to refer to the tendency of people having less pigmentation the higher up they are placed in the social hierarchy. Of course there are prejudice, bias, and unjust privilege in all societies, but what good does analyzing them through the lens of quack races do?

And then there is socialism. There is no left in the US. Is this chance or one effect of state-endorsed racialism?


J-D 08.15.19 at 10:30 pm


I think it could reduce it somewhat; at the very least, it is a step in the right direction.

You think so; I think not.

For example, in the US, people have official documents, such as birth certificates, that state the race of the person

No, they don’t. If your conclusions are based on this kind of falsehood, they are automatically suspect.

Of course there are prejudice, bias, and unjust privilege in all societies, but what good does analyzing them through the lens of quack races do?

Denying the facts always makes analysis harder. Racial categorisation is a fact: not a biological fact, but a sociological fact. If you want people to stop doing it, you must first acknowledge that they are doing it. If you want to explain why people should stop doing it, you have to explain the effects of what they’re doing, and you can’t do that if you try to pretend it’s not happening.

Here is the official US Census Bureau explanation of why they collect information about race:
Notice that they acknowledge explicitly that race is a social construct and not a biological category.
Here is the report of an official review which goes into more detail about why the information is collected, and why the government uses the particular format that it does:


john burke 08.15.19 at 10:52 pm

Whistle Vivaldi?


hix 08.16.19 at 8:16 am

The reflexive reassurance about greatness in other sectors is already a good explanation for why such problems exist.


Gareth Wilson 08.16.19 at 10:32 am

Whistling Vivaldi comes from a tactic a black college student used to make white people less anxious when he walked down the street in Chicago. It really did make them treat him with more respect, not cross the street to avoid him and so on. Presumably because they were less likely to think of him as a violent thug if he was whistling classical music.


john 08.16.19 at 11:18 am

@J-D. Fact: My US birth certificate clearly states: Race: Negro. It is an official document, accepted as natural citizenship documentation for my US passport.
Fact: State issued documents may adhere to different guidelines than federal guidelines, especially those published via White House in 2017.


J-D 08.17.19 at 10:27 am


Thank you for the information. My mistake. I apologise.

I reflected on the fact that birth certificates in the US are issued by States and I decided that I wasn’t going to check the requirements of fifty different States before commenting and so relied on more general information. I could have been more careful.

All that said:
the fact that your birth certificate states your race as Negro clearly shows that it’s not relying on the US census classification, which doesn’t use that category.

Would you be prepared to take a guess at how many of the States use racial classifications on their birth certificates?

Comments on this entry are closed.