Why Violence Spiked in 2020, in One Graph

The Thanksgiving Edition!

Welcome to the Thanksgiving Edition of External Processing in which I tell jokes and share perhaps the greatest Ted Lasso video yet produced. But first, something serious to chew on.

Why Violence Spiked during COVID and Why Crime is Falling

Over the last three months, I’ve essentially quit social media. I had a fair-sized community on Twitter, but the algorithm has changed, and I am unable to reach my community and they are unable to reach me. What does reach me, I don’t have much interest in. I have little interest in reading polemics and self-affirmations of a rigid ideology. It’s super boring. Tell me something I don’t know, I’d like to read that.

I have also mainly given up reading the criminology literature, at least with respect to the Great American Violence Explosion of 2020.

If you are reading this essay, you probably know the contours of the data. The key statistic was a 28% surge in homicide from 2019 to 2020. How unusual was that increase? It was the largest one-year change in homicide since modern record-keeping began in 1960. In fact, it’s three times the size of the next biggest increase.

Criminology is usually pretty immune to silly culture war stuff, but the criminology research on this spike in gun violence has been more ideological than any other research question I can recall. There are two camps and let me introduce them.

In one corner is ‘the police have been hamstrung by their critics and can no longer do their job’ crowd. The evidence they give is that violence spiked at the same time the George Floyd protests began. The causal mechanism they propose is that, son, we live in a world with walls and those walls have to be guarded by men with guns. Who’s going to do it? You?

In the other corner is the ‘there’s nothing to see here crowd’. The argument they make is that there really wasn’t a Great American Violence Explosion in 2020. The evidence they give is data from the 2020 National Crime Victimization Survey which showed:

• The violent victimization rate declined from 21.0 per 1,000 persons age 12 or older in 2019 to 16.4 per 1,000 in 2020.

• The number of violent crimes, excluding simple assault, fell from 2.0 million in 2019 to 1.6 million in 2020.

The problem with the protestor narrative is that argues that the police pulled back and that caused thousands of additional shootings. But robberies and burglaries went down. Do we think police deter shootings but not robbery? It’s head-scratching.

The problem with the nothing to see here argument is that it ignores the more than 15,000 additional homicide deaths and tens of thousands of additional shooting victims over the last almost four years.

While both points are defensible at some level, they only nibble at the real question.

Which is, why was there a massive spike in gun violence in the spring of 2020 and why has it taken years to abate?

Since self-affirmations of a rigid ideology are super boring, let’s put aside the debating points and just look at the data. Here’s the key graph:

Source: Council on Criminal Justice.

The figure shows six years of monthly homicide data from the FBI, 2018-2023. You can see the seasonal ebbs and flows in the data, with spikes in the summer and troughs in the winter. You can also see clearly that prior to the pandemic (March, 2020), homicides per month were a little under 1.5 per 100,000. In the three years after the pandemic, homicides per month are well over 1.5 per 100,000.

In fact, after the pandemic, homicides barely dipped below 1.5 per 100,000 in any month. In sum, the average across the three years post-pandemic was much higher than the three years prior to the pandemic.

What Happened?

I’ve written a lot about this graphic and my take is essentially that this violence outbreak was a function of the pandemic and tore the social fabric. In March of 2020, everything closed down which forced people to be home in neighborhoods where gun violence was already rampant, the accumulated trauma was deep, and retaliatory shootings (especially around drug markets) exploded. The availability of firearms skyrocketed at the same time, creating the means to accompany the motive.

It's a tidy explanation, and I have made this argument a lot. What’s missing though is a critical piece. Young men in high-violence neighborhoods with accumulated trauma were home and isolated. But, what were they isolated from? What interactions did they not have that could have prevented this extreme cycle of retaliation?

A big part of the answer is local government.

How Smaller Local Government Led to an Explosion of Violence

The reality of life in America in 2023 is that if you live in a neighborhood with high levels of gun violence, you have many pro-social interactions with local government and very few with the private sector.

Government teachers educate kids in the neighborhood. Government social workers connect people with services. Government police deter bad decisions and (hopefully) counsel young people to make better choices. And perhaps most importantly, government civil servants fund local community-based organizations that provide services—mental health, public health, workforce training, after-school programs, and many, many more. And those government civil servants work to make all those programs not just function but function better.

Government, baby, is where the rubber meets the road.

What happened in 2020? Local government shriveled, and the desiccated husk was not sufficient to meet the accelerating needs of people struggling under the weight of COVID.

The graphic above is the number of local government employees over the last 70+ years. You’ll note some important correlations with crime in America. In the late 1970s when crime and violence in America exploded, local governments were not expanding at the same rate as in past periods. In the first half of the 1980s, when crime was at—and stayed—at historic highs, local governments shrank.

And it’s important to consider these data in the context of US population growth. In 1950, there were 157 million Americans—today there are over 330 million. So even a period of stagnant local government growth masks a per capita decline.

But I’m sure you noticed the steep decline in 2020, so let’s zoom in on that.

In March of 2020, there were 14.7 million local government employees. By May (just TWO months later), there were 13.46 million. That’s almost a 10% decline. At a time when demand for the services provided by local government employees absolutely exploded.

It is also worth pointing out that as of October 2023, the number of local government employees has not yet returned to March 2020 levels.

Early returns on crime in 2023 suggest that the peak of violence has passed and that crime and victimization are returning to pre-pandemic levels. That also fits the data, on the nose.

In closing, let me just say that there are about 55,000 coal miners in America, and we seem to talk about them all the time. There are something like 700,000 local government employees who wear a badge, and much of the political discourse revolves around whether that is the right number. There are more than 13 million other local government employees and we never hear a peep about them.

But they matter. Enormously.

You are Rooting for the Wrong Team

We are definitely in a moment of existential angst. Pandemic angst. Climate angst. End of the American democracy angst. Third World War angst.

My first angst was Cold War angst. I grew up in a close-in Washington, DC suburb and one day when I was about ten, the Washington Post ran a front-page graphic of what would happen to the city and the suburbs if a 50 Mega-Ton Soviet nuclear missile detonated over the White House. In the first concentric ring, everyone was dead immediately. In the second concentric ring, you might survive for a minute, until the radiation killed you in a horrible way. In the third ring, there was some chance you might make it. Lucky people in Paw Paw, West Virginia would probably be ok. We lived in the second concentric ring.

As you may have read in the papers, this did not come to pass. But like a childhood vaccine, this exposure has had some staying power and has made me somewhat angst-resistant as an adult. But I do perk up a little when a new existential threat appears. So I’ve been chewing on the Rise of the Machines angst from the threat of artificial general intelligence (AGI).

The first thing to know about AGI is that this is not new, and we have been headed down this path for generations. The first hard thinking about all of this started almost 100 years ago. And since we’ve thought a little about it along the way, maybe we should apply some of that thinking to our present situation.

Listen, before there was generative artificial intelligence, there were machine learning algorithms and neural networks. And before there were neural networks and machine learning algorithms there was stepwise regression. And stepwise regression was a Bad Thing, unprincipled, or at least that’s how I was taught. It’s perhaps helpful to think about why it was (and still is) considered a Bad Thing.

The idea of stepwise regression is that if you want to know if more education causes higher incomes, you have to control for a bunch of stuff that seems to matter, because if you don’t then you will think education matters a lot when it’s really something else. Other things might include age, race, and gender, and maybe privilege, like parents' earnings, and home neighborhood characteristics, and maybe some subjective stuff, like height, eye color, and whether your face is symmetrical.

And all of that should be based in theory. Like, you can’t just throw all that stuff in a regression and see what sticks, you have to have some reason for including it that you can explain. But stepwise regression just says, naw, let me handle that, and takes in all of these competing explanations and then spits out only those predictors that matter in a statistical sense.

The problem, of course, is that you get a lot of wrong answers this way, because stepwise regression takes the data literally. Researchers, on the other hand, take their data seriously, but not literally, at least not when a literal interpretation goes against theory. You probably scoffed at that the word ‘theory’, maybe even rolled your eyes a little, and mumbled ‘implicit bias’.

But theory is a bit of a bulwark against implicit bias, especially on measures of inequality. Because inequality is a latent construct—like alcoholism or beauty—it cannot be directly described, it can only be described through its attributes. And if you only include a couple of those attributes, you are not measuring the latent construct at all, and what you’ve got is soup. And soup is what they make at the restaurant on Monday with whatever is left over from the weekend.

Here's why it matters. Race matters enormously in determining who is poor in America. And poverty is really hard to measure well. That is because ‘poverty’ is a latent construct that encompasses so much more about limited opportunities than just income or even wealth. Poverty also includes all of the social structures that divide people into poor and not poor. But if your data only includes one attribute of poverty—race—and race is just a proxy for all the social structures that divide people into poor and not poor, then you will come to a lot of wrong answers about race and poverty.

So, that’s why I am just a wee bit skittish on the whole general intelligence thing to begin with. Our data about important things are not very good and will never be very good because important things like poverty are hugely difficult to measure and always will be. And having a faster calculator doesn’t make the numbers you put into the calculator any better. It just makes the consequences of the calculations more consequential.

Anyway, that’s my prior belief as we consider the Great Open-AI Dust-up of 2023.

The Great Open-AI Dust-up of 2023

Note: every time I review a draft of this essay, the situation with Sam Altman has changed. But Sam Altman really doesn’t matter in this story. What he represents—commercialization of something potentially dangerous—matters much more.

Theoretically, in the Great Open AI Dust-up of 2023, the board of the non-profit Open AI, Inc. is focused on weighty matters of right and wrong in the development of artificial general intelligence, and the subsidiary, OpenAI Global LLC valued at $86 billion, is concerned with the business, man.

To bring you up to date, in the Great Open AI Dust-up, the nonprofit arm concerned with right and wrong voted out the leaders of the business. And in a metaphor for the future we all fear, the businessmen refused to be killed. And chaos ensued.

And you, dear reader, are almost certainly on the side of the business. As far as I can tell, most everyone is. The world is smitten with Sam Altman, loves ChatGPT for the funsies, and sees a bright technocratic future of Instagram sunsets on a rainy day.

I think you are rooting for the wrong side here.

I’ve written before in great depth about Type I and Type II errors and I learned my lesson when no one read it, so I won’t bore you with it again. But let’s just play out what happens if we get a false negative or a false positive on AGI.

If we get a false negative and think AGI is dangerous when it isn’t, capitalism being what it is, the wonders of AGI will be delayed for a time but burst forth soon enough.

If we get a false positive and think AGI is not dangerous when it is, well, here’s Bloomberg:

OpenAI is a very strange nonprofit! Its stated mission is “building safe and beneficial artificial general intelligence for the benefit of humanity,” but in the unavoidably sci-fi world of artificial intelligence developers, that mission has a bit of a flavor of “building artificial intelligence very very carefully and being ready to shut it down at any point if it looks likely to go rogue and kill all of humanity.”

In summary, the false positive is thinking AGI is not dangerous when it is in fact dangerous, and if we succumb to the false positive we will not shut down AGI when “it looks likely to go rogue and kill all of humanity.”

Call me a Luddite, it's super trendy to do that, but this benefit-cost analysis does not look all that tricky to me.

Go Team Non-Profit Board of Directors!

On the other hand.

A venture capitalist, a philanthropist, and an academic walk into a bar.

And the venture capitalist says, “hey, I’ve got a new technology that can change the world, make all of humanity better off, end inequality, and create magic Instagram sunsets on a rainy day.”

And the philanthropist says, “that’s great, I will fund you in order to change the world, make all of humanity better off, end inequality, and create magic Instagram sunsets on a rainy day.”

And the academic says, “ok, but does it work in theory?”

Happy Thanksgiving

You can’t write about the pandemic and not write about Ted Lasso. It was a gift.