Hiring 50,000 Police Data Scientists

The most frequent comment I get on my essays is: can you not meander so much at the beginning and just say what the thing is you are getting at? While there is not so much fun in that, it’s fishing in a river with no bends, I suppose it is a fair point. So, here’s the point of the day:

March 17, 2024, will mark the 20^th anniversary of one of the most influential books ever written: Moneyball: The Art of Winning an Unfair Game. The origin story is about baseball, but it has now become a theory of everything. It is to neoliberalism what the Fountainhead is to libertarians.

OK, that’s a little tongue-in-cheek, but over the last two decades, we have been utterly besieged by Moneyball. There is Moneyball for Government, Moneyball Medicine, Moneyball Church, even Dental Moneyball.  I am sure someone, somewhere has written a meta treatise on Moneyball for Moneyball.

The principles of Moneyball are thus: decisions made with scarce resources should rely on data and evidence—not gut feelings and intuition, which are, in the Moneyball world, what you think before you think about something. Thinking clear-eyed about hard choices, using rigorous data and analytics is a better way. There is nothing new here, these are essentially the principles economist Gary Becker wrote about in the 1960s. Becker applied these principles to much of human behavior, writing about human capital, time, marriage, crime and punishment, even, lest you think the current Moneyball fetishism goes too far, Child Endowments and the Quantity and Quality of Children.

Just about every industry has been altered in some way by Moneyball concepts. Except for one.

Policing.

Why is policing immune? Because policing is a monopoly. The whole point of Moneyball is to be successful in a competitive landscape, while the whole point of a monopoly is to avoid being in a competitive landscape.

Nevertheless, policing would unquestionably benefit from the application of some Moneyball principles.

If you go way back to the beginning of Moneyball: The Movie, to the opening scenes of Moneyball, you might recall the very first thing Brad Pitt did. Way before he invented on-base percentage, WAR and sabermetrics. Way before he drove around the refinery, before he traded away half his team, before he changed baseball. Way before anything else, Brad Pitt did something even more fundamental.

The very first thing he did was hire Jonah Hill.

Before everything else, he hired a data scientist.

All of Moneyball starts with people who can work effectively with data.

Policing should start there too.

Better Decision-Making in Policing

When she was little, my daughter and I would play a game on the trampoline. While we bounced, one of us would propose a sentence, then we would take turns yelling it out with the emphasis on a different word in the sentence. “Bubbins is a very large cat” was a favorite—Bubbins being the name of our very large cat. BUBBINS is a very large cat. Bubbins is a VERY large cat. Try it for yourself, it’s pretty fun.

Such is the power and subtlety of language when closely examined. The central problem of our time of course is that we are all grappling with who has power, and in power there is no subtlety. Language is just another weapon. This is a shame because better answers almost always hide in subtlety, and when language is wielded as a cudgel, few better answers are likely to emerge.

I want to write a little about a very subtle shift in language describing how we understand policing and see if in doing so it starts down a path toward a better answer. I think it helps to clarify why we need data scientists.

In the summer of 2020, with the protests after George Floyd’s murder roiling the streets, I was stuck on a comma. This, I suppose is the difference between an advocate or a changemaker—people who don’t get stuck on commas—and researchers, who do. It sounds trivial, but as I look back on the almost four years that have passed, the existence or nonexistence of that comma seems more profound.

The statement in question is this one, which a lot of people wrote in 2020:

There are too many cops doing the wrong things.

With a comma, that sentence changes entirely:

There are too many cops, doing the wrong things.

At the time, I wrote the second sentence, with the comma. I added the comma thinking it softened the sentence into more of what I intended, which is an acknowledgment that maintaining policing at meaningful levels is bedrock insurance for the social contract. But we would all be better off if the right number of police spent their time doing the right things.

In retrospect, I realize that this comma does the opposite. When you write, “there are too many cops, doing the wrong things” the implied interpretation is “there are too many cops. And they are doing the wrong things.”  In that construction, there are two failures of the police: there are too many of them. And they are doing the wrong things. And it also implies that they are doing the wrong things precisely because there are too many of them. So, perhaps that is three strikes.

This is not what I meant. What I meant was that some police are doing things that put them in position to make terrible mistakes. And that we know what those things are: adversarial policing practices, like high speed car chases, presumptive stops based on race, dangerous holds and restraints, tactical failures that lead officers unnecessarily onto dangerous ground and more. And we should ask the police to stop doing these things and do better things instead. And we should give them tools to do so. There is a lot of fertile ground there, there are many mistakes. It’s a very hard job.

In the intervening years, some of this has happened. Not a lot, but some restrictions have been placed on the police and some of these practices have been reduced.

But restricting police is not a path to a more empowered social contract. Imagine you run a noble pension fund as a stockbroker, and your job is to take the savings of widows and orphans and turn it into wealth so that they can have better lives. Now imagine that some of your more overzealous stockbrokers engage in shady practices to achieve this goal. They trade on insider information. They short stocks and then gossip and sabotage the social capital of companies to run down their stock price. They front run trades, buying and selling ahead of other investors to benefit the widows and orphans. Appalled, you put restrictions on your traders to prevent these practices.

That’s all to the good, but you must understand and correct for the explicit tradeoff. You cannot expect this to benefit your widows and orphans. You are engaging in more ethical practices, and you should, but the cost is going to be some degradation in the monetary returns. Putting restrictions on your stockbrokers is going to make them do fewer problematic things, but it is not going to magically help them do more good things.

The key is to give new tools to your traders to facilitate their success, to help them do even better than they did before the restrictions.

Data Science and Policing

I’m not going to spend much time on motivating the idea that adding data scientists to law enforcement would help policing. I think it is self-evident. Recall again in Moneyball, once the data scientist was hired, what the first and fundamental insight that analyst brought to the table. It was this idea.

The most important thing a baseball team can do is to score runs. And minimize the runs scored against them.

That’s it. That’s what Moneyball brought to the table. The team who scores more runs, wins. But why that mattered was because baseball had lost sight of this truth. And realizing this was a huge comparative advantage.

The equally self-evident Moneyball observation about policing is that the goal of policing is to reduce crime, to prevent it, to stop it. That is a fundamentally different goal from catching bad guys. The two Venn diagrams overlap, but way less than you would think.

The same kinds of problems that occur in policing occur in other fields. For forever, teachers tried to explain that there was only so much they could do in the classroom to solve the problems kids brought to school with them. The key, they argued, was to integrate what happened inside the classroom into what was happening outside of it. Your local police officer faces the same problem.

Data scientists can help to solve this problem. What data scientists do, at the core, are two things: 1) put together data from multiple sources, and 2) used sophisticated tools to analyze that data. Each of the actors in the criminal justice system knows very little about what the other actors are doing. The police get little information from the courts about what happens after an arrest. Judges have no idea how well people fare after they’ve been sentenced. Corrections has enough trouble keeping track of people within the system that they don’t have the capacity to tell the other systems when someone is being released.

Beyond working with other systems, data scientists can link data within policing, from the various departments and task forces serving a single police force. They can link across geographies to neighboring departments. And they can link data within geographies—for instance, 41 policing agencies patrol the streets of the District of Columbia, and you can be sure they would benefit from better coordination.  

Data scientists can analyze those data using sophisticated statistical models to provide real-time, actionable tactical information. And, more importantly, they can provide strategic guidance. What should officers be doing, when and how? How many officers are needed for routine patrol and special investigations? How can scarce policing resources be used more efficiently?

How many data scientists does a law enforcement agency need? A lot would be the right answer. One per precinct, per shift would be my guess. One to work with the precinct commander on that day’s assignments, to be on call throughout the shift to aid decisions about deployment, to run analyses (across data systems) on what worked and what did not on past shifts, and to make projections about future deployment, tactics and organization. Data science in policing is a hands-on job.

The Police Labor Shortage

This is the right time to think hard about adding data scientists to the police force, en masse. Five years ago, the Police Executive Research Forum warned of a crisis in policing—many of the nation’s 19,000 local law enforcement agencies were having trouble recruiting new sworn officers. You need only look at the labor statistics data below to see why they were concerned. And that the problem is getting worse.

It is frustratingly difficult to get recent Bureau of Labor Statistics data on the number of people employed in law enforcement in 2020-2023 as the Bureau of Labor Statistics data series weirdly stops in 2019. But I’ve cobbled together some data and here’s the long-term trend. To get back to the average across the trend, the US would have to hire something like 50,000 patrol officers.

Source: Bureau of Labor Statistics, Bureau of Justice Statistics

But patrol is just one part of policing, let’s look at their supervisors. While there are 650,000-750,000 patrol officers in the US, the number of supervisors tends to be around 100,000. To get back to the average of this trend, the US would have to hire 10,000 or so additional supervisors.

Conclusion

I have not even gotten into what I see as the third leg of the police reform stool (1. Data science would improve policing, 2. Police force staffing is way below historical norms and 3. Police labor supply). It remains head-scratching to me that police departments fail to recognize the reality that policing is not a high-demand position in the US right now. People do not want to be cops in the same way they used to want to be cops. Higher pay might solve some of this, but policing seems relatively well-paid today compared to similar jobs with the same prior experience.

In other industries the response would be straightforward. Leaders would ask, can we make these positions more appealing by changing what police officers do? And can we do that in a way that serves our mission, and perhaps even benefits it?

The answer here is a resounding yes! There is a happy marriage just waiting to be embraced. Data science is among the hottest professional fields today, and policing would vastly benefit from more data scientists. Police forces do not need to create new positions for data scientists, they merely need to hire them into the open positions they already have.

If I had six officers and now I have four, and if I hire a fifth who is a data scientist who can help the remaining four be just as effective (or more effective) than the six I used to have, why wouldn’t you do that?

Coda

I am aware that citing Becker is problematic and framing the entire discussion around rational choice and self-interest does not appear to center issues of equity and justice which are at the core of the need for reform. But I would propose that data science can be a really effective mechanism for both justice and equity. There are a lot of important studies on disparities in the justice system. But, I imagine, it feels distant from the day-to-day work of policing.

It is impossible to imagine effective data science being incorporated into routine policing without it naturally lifting up issues of disparity. Without question, disparities undermine police practice. Policing, by definition, would be more effective if it was more just and equitable.

Musical Interlude

What to do when the world discovers something you have loved deeply and passionately for a long time. My teenager pointed out to me earlier this year that Luke Combs, a country star I was vaguely aware of, had covered Fast Car by Tracy Chapman, an artist I revere. These are good moments. My wife had one when the same teenager put a Maggie Rogers/Zach Bryan duet at the top of her playlist. Maggie Rogers holds a special place for me, well an ambivalently special place I guess, because she was the one who confirmed for me that I had arrived in middle age. We saw her at the Mann in Philly on a freezing November night, and as this lovely woman moved about the stage in her concert gear, the little there was of it, all I could think, all night, was how cold she must be.

But Fast Car is both special and unique. Like Tracy, I was in my early 20s when Fast Car debuted and shared enough with her characters that I could dream with them a little. My Fast Car was an ‘81 Ford Mustang, dirty brown with giant orange and yellow racing stripes, with four on the floor but also a grossly underpowered 2.3L engine. The dice over the mirror were a paean to what almost was. I also had no money, though I was not poor because I had a solid middle-class safety net to fall back on in desperate times.

But I felt Tracey’s lyrics. I wanted more than anything in the world, to go somewhere. And be someone.

Overwhelmingly, the craft of songwriting holds me. I heard Paul Simon describe his approach as starting with something terrifically simple and building. Start with an image that is universal.

He's a poor boy

Empty as a pocket

Empty as a pocket with nothing to lose

Or,

Early one morning, the sun was shining

I was laying in bed

Wondering if she'd changed it all

If her hair was still red

Or perhaps the greatest six-word intro of them all,

Screen door slams

Mary’s dress waves

Or, from Tracy,

You got a fast car

I want a ticket to anywhere

The Chapman-Combs duet is being celebrated as a moment where intersectionality meets homespun. I get that, but it grossly misses something better, something more raw and powerful. It misses grace. Grace is what makes Fast Car work, grace in the lyrics, grace in the melody. The grace of Tracy Chapman. And the grace of Luke Coombs. Grace is in terrible short supply these days.

Hank Green tells the story of the first performance of Fast Car.

“I hope everybody knows that Tracy Chapman’s “Fast Car” was first performed in public when Stevie Wonder’s team was having technical difficulties at Wembley Stadium [to celebrate Nelson Mandela’s 70^th birthday]. Chapman went back out after having played a previous short set to buy Stevie’s team some time with an unreleased song. The crowd of 74,000 had been frustrated by the delays but what happened once she started playing...”

---

I am well aware that Bruce Springsteen, noted Bruce Springsteen anthologist, has stated that the correct lyric is “Mary’s dress sways.” With all due respect and humility, I note that Mr. Springsteen is wrong.