At my Substack Cotillion, young and cherry-cheeked, full of vim and not yet scorned from polite society, I debuted with a newsletter about the merits of reforming how we police traffic. I argued that police are trained to control situations and look for threats, even in low leverage situations such as a traffic stop. But that traffic stops are trivial from a public safety perspective. And, these stops are a potentially toxic mismatch of police training and the needs of the civilians they intercept and poor use of a sworn officers’ time. My three then-subscribers were tight with their praise and I left the whole thing to marinate a while. But sauces need agitation, so let me stir a little. And have a sip from the ladle.
Recently, I’ve read several excellent pieces summarizing a wide array of proposed policing reforms. Last month, the Brookings Institution released an excellent report that I highly recommend, A Better Path Forward for Criminal Justice. The first chapter focuses on ‘sensible’ police reforms including the end of qualified immunity, officer wellness, and ending the ‘us versus them’ policing culture. The remainder of the report reimagines pre-trial and sentencing, prison reform, police in schools, fostering desistance, training and employment for the incarcerated, and prisoner reentry. The essays are fine scholarship, thoughtful, timely, and nuanced. Implementing all of these reforms would most assuredly make America a more just place and I offer nothing but support.
But most of these proposals, like most of the reform proposals that seem viable at the moment, are limited in an important way. At the big end of the funnel, where people intersect directly with the police, the proposed reforms are very indirect. Ending qualified immunity, increasing implicit bias training, de-escalation training, demilitarization, all of these seek to achieve our collective aims by having the police more or less do exactly what they do today, only better. The other reforms are indirect in the sense that the key interaction—the first domino—has already fallen by the time we reform pre-trial or prosecutors or corrections. Too many people who should not be in the system are already in the system, and too few of the most predatory people have been identified.
What I would like to propose here is a more direct set of reforms. In addition to the reforms proposed in the Brookings report and elsewhere, I would like to start with the tip of the spear, the police. I want to focus on the boots on the ground. I want to ask what police do and why they do it. And I want to propose a solution that calls for police to spend far more time investigating and solving serious crime and far less time—far, far, less time—engaging in policing with people who are not a part of those crimes.
I think most people would reflexively agree with this concept in theory but would vehemently oppose it when the details and implications were explained to them. Because to get to what I am proposing, sworn law enforcement officers would have to be much less available and accessible to the general public. And there, of course, is the rub.
But, I think I can show you why this is the right course of action, through a pretty simply logic problem. And that logic problem revolves around what kind of errors we are willing to make as a society. How we police is a tangible expression of our core values, and we should not be careless with our error. We will inevitably make some errors and they cannot be predicted. But what can be determined by our choices is in which direction those errors most often fall. Those errors, of course, are about who is not caught and goes unpunished and who is caught and punished.
My takeaway is that adversarial policing—where police initiate stops—generates far more and bigger errors than does consumer-driven policing through calls for service. And as result, we should do far less adversarial policing.
A thought experiment
Suppose you were given the current police budget of some city and the opportunity to reassign every officer to their best use. Suppose the outcomes you cared about were maximizing public safety and minimizing injustice of all types, not just the injustice of an unsolved crime or a wrongful conviction, but the injustice of racial disparities. And suppose you knew that the current system did a poor job of solving cases and preventing crime, and the system was statistically utterly racially unjust. And suppose you undertook this effort in the middle of the worst spike in violence in 50 years, spiraling off a foundation of crime that started at a far higher level than our peer nations.
Here’s the twist. Suppose you also saw the police as generally soft-hearted and hard-headed and that your goal was to maximize their chance of success.
Now. You have these facts, your police officers, and a given budget. What will you have them do?
My guess is that the first thing you would do would be to ask, ‘what do the police do today?’ And you would be told that the police divide their time between responding to calls for service and initiating stops of one sort or another.
There is ample room for improvement in how police respond to calls for service. I just now looked at the calls for service data from a mid-sized city. Ranked in order from most common to least common, the first code that referenced anything suggesting any kind of altercation was well down the list, barely in the top ten. Many of the rest were false alarms and wild goose chases. But there is clear data on what kind of calls are received, and data-driven choices can be made about how to allocate resources to respond to calls for service.
On the other hand, when you turn to the other bucket of police-public interactions and ask, ‘what kind of interactions with the public do the police initiate?’ you would receive a far less definitive response. Traffic stops? Yes. Requests for information? Yes. Suspicious persons? Yes. Drugs and prostitution stops? Yes. Stop, question and frisk? There are a lot of gaps in those answers and a lot of room for error. Some of these police-public interactions go into the data. Many do not.
You might ask other questions, like, how much of an officer’s time is spent investigating crimes that have been committed but are not yet solved, or, how much of an officer’s time is spent preventing crime? The answer to both questions would probably be some version of, well, that is what the officers are doing when they aren’t responding to calls for service.
The Supply and Demand for Policing
So, at this point, you would probably try and create some sort of taxonomy or a rubric to try and distinguish between calls for service and police-initiated activities. You would want to differentiate them so you could figure out where to concentrate your scarce resources. And in particular, you would want to know which of these activities was most likely to successfully reduce violence and other serious crimes, and which of these two activities was more likely to lead to errors of injustice and racial disparities.
A good place to start is by thinking about what the actual difference is between calls for service and police-initiated contacts. Conventionally, calls for police service from the public are labeled ‘consumer-driven’ and they are, at least between the caller and the responding police officer, consensual. They are a transaction like any other where the two parties implicitly agree on the objectives of the deal.
On the other hand, any action initiated by the police is, by nature, ‘adversarial’. In an economic context, these labels are almost tautological. When a consumer calls 911, they demand services, which the police supply. When the police initiate an action, the supply a service for which there is no demand. If you supply a service, whether policing or plumbing or advertising, for which there is no demand, it is by its nature, adversarial.
There is a long literature on this, with much more traction in British criminology than in America, on the differences between these two types of policing. Using data from the 1982 British Crime Survey, Southgate and Ekblom (1984), put it this way:
Now, I can certainly imagine that some will raise objections. In particular, that police generated calls for service are not actually adversarial because the public has an aggregate demand for policing even if the person to whom the police are supplying policing (aka, the alleged perpetrator) most assuredly does not demand their services. This is, in fact, the very definition of an externality, where the gains or losses to someone outside the immediate transaction are not accounted for in the costs and benefits of the transaction. The opens the door to a fascinating digression since many economists argue that externalities dominate any study of the criminal justice system. But let me just acknowledge that this rabbit hole is there and let’s move on.
Pick your poison: what kind of errors should police make?
Now we are to the main point. In both designing and reforming criminal justice policy, we spend far too little time thinking about the consequences of error. All our reforms are guaranteed to work. I wrote about this a while back, in perhaps excruciating detail, but I want to revisit the idea of different kinds of errors. A problem with the criminal justice system today that never gets discussed, is the problem of Type I and Type II errors.
Type I errors are where you reject a true null. Which in English means that you become convinced that something works that does not work. These are the programs that do get funded that don’t work.
Type II errors are where you accept a false null. Which my handy-dandy Stats-to-English translator means that you become convinced that something doesn’t work that does. These are the programs that don’t get funded that actually work.
Ok, simple enough[1]. But now, let’s make it a little more complex to match the reality of the world we live in. In the real world, it doesn’t just matter whether something works, it matters if the benefits of doing that thing outweigh the costs. When you apply an economic lens to this, you end up with this table:
What this table says is that if you make a Type I error in estimating your benefits, you actually get no benefits because the program doesn’t work. But, if you make a Type I error on the costs side, it’s a bonanza, because the costs are actually smaller than what you thought they would be! Putting it together, if you make a Type I error on both costs and benefits it’s a no harm, no foul kind of situation. The thing doesn’t work so there are no benefits, but who cares because it turns out it did not cost very much.
Type II errors’ work in the opposite way and are much more insidious. Dastardly. On the upside, when you make a Type II error in your benefits estimate, you expect the thing you invested in to produce fewer benefits than it actually does. More bonanza! But, and here’s the rub, you have also underestimated the costs. And the costs can easily be so big that they swallow up all the benefits. Want an example? Mass incarceration. While mass incarceration almost certainly contributed to the crime decline, in the process it absolutely destroyed too many communities to count by removing whole generations of young, black men from their home communities.
So, let’s take our two types of police activities and see where we want to make errors. I’ll define our two choices this way:
Adversarial policing (mainly of violations and infractions);
Consumer-driven policing/calls for service (more of a mix of calls, including more violence).
What we are going to try and figure out in these tables is what happens when each of these works better or worse than expected. So, first, what does it mean for adversarial policing or consumer-driven policing to ‘work’? In these examples, they ‘work’ if either 1) new infractions are deterred, or 2) the ‘offender’ is identified. In policing language, this would mean that the case was ‘cleared’. Type I errors in costs and benefits mean they were both less than anticipated and Type II errors mean both costs and benefits were greater than expected.
In Table 2, I am simply replacing the values in the boxes with a statement about what would happen if that kind of error occurred.
Type I errors in the costs and benefits of responding to calls for service is sort of a yawner. They show that costs are less than anticipated (probably because there is less consumer-driven policing than anticipated) but that what was done has little effect. Type II errors are more interesting. They suggest an underestimate of the benefits, perhaps substantially. But, the added costs are not obvious. There is certainly an opportunity cost if we are making Type II errors about policing, the police could be doing more important things. But the impact on justice is less clear—are too many officers responding to too many calls for service creating injustices in general, and more racial disparities in particular? Yes, but there are a number of possible solutions that are under study that may help to solve the problem.
In Table 3, Type I errors in adversarial policing have the same outcome as in consumer-driven policing—not much benefit and an opportunity cost. But Type II errors are a whopper. Most racial disparities in the criminal justice system seem to be caused by differences in who is stopped by the police, that is, who has an adversarial stop. Why would there be more racial disparities in stops from adversarial policing than from calls for service? Subjectivity. Adversarial policing creates several layers of subjectivity that do not exist in calls for service. The officer determines who to stop among many eligible candidates to stop. Whereas in the call for service the criminal activity is already defined, the officer in adversarial mode can choose which criminal activity or infraction to stop and which to ignore from among many. There’s more to it of course, but that is sufficient for now.
The Takeaway
The obvious takeaway is that the biggest possible policing error on the board is a Type II error where we underestimate the costs from adversarial policing. None of the other seven possible errors are even close. All of the Type I errors are bounded by budgets—the size of the error is limited to how much you spend. The Type II errors in benefits are equal—the potential errors are equal in consumer-driven policing or in adversarial policing if we underestimate how effective they really are.
The big question is around the damage that is done if we make a Type II error and we underestimate the costs. I have already given one reason why underestimating the costs of adversarial policing is much greater than if underestimating the costs of responding to 911 calls. Let me give you another. Let’s think about the logical consequences of making an error in how much adversarial policing should be supplied. And let’s think about it in the particular domain of infractions which are law violations that do not result in a criminal sanction but rather a fine. This category includes most physical and social disorder, things like loitering, jaywalking, speeding, tagging, and riding a bicycle on the sidewalk. Legally, these fall into the same category as building a patio without a permit or failing to cut your lawn or shovel your sidewalk.
If you commit a Type II error with adversarial policing, particularly disorder policing, you have underestimated the benefits. But how big could those benefits possibly be? The reality is that we make these errors because we are terrified that if we don’t stop 100 cars for speeding, we will miss the one with the kidnapped kid in the back seat. Or, if we don’t grab 100 turnstile jumpers we will miss the one with a felony arrest warrant. But really, are the odds of that needle in the haystack bigger than if we swapped out all those traffic stops and spent more time investigating homicides? Isn’t that potential benefit almost certainly bigger?
And, is it really worth imposing all of that carnage on people of color, all of the damage to the legitimacy of the law and law enforcement, all those lives forced onto a new and worse trajectory because of an arrest out of fear of letting one guilty man go free? And in doing so, all but guarantee that other guilty men, guilty of more serious crimes, do go free because the resources to investigate them are tied up in all those adversarial tops?
At the end of the day, the conclusion of this thought experiment is that if you were starting behind the veil of policing ignorance, and only knew the risks of being wrong, that you would almost certainly pour your resources into calls for service and find other ways to police disorder than with police. And then you would start working on making the outcomes of those calls for service better.
[1] If you are playing Don Rumsfeld Bingo, these are both unknown knowns, things you think you know that you are wrong about. Because Don Rumsfeld never made mistakes, he doesn’t talk about this problem very much, so you will have to look at this instead.