Two Big Thoughts About the Steep Crime Drop

Without a doubt, the most striking feature of U.S. crime rates in 2023 was the sharp nationwide decline in homicide over the year. According to Jeff Asher’s assessment of approximately 175 cities across the country, murder rates fell by about 12%. That decline was not enough to erase the entirety of the homicide spike of 2020, when homicides rose by nearly 30%, to levels last seen in the early 1990s, but it remains one of the steepest declines in recent history. And it is a decline that returned homicide rates to around where they were in 2005. Which means the decline from 2022 to 2023 — just two years — is equal in size to the decline from the early 1990s to 2005, which happened over the course of a decade.

Trying to explain the causes of the 2023 decline is tricky, if only because we still don’t fully understand what caused the 2020 spike. (I strongly believe that, in the end, the most significant explanation will be COVID-19 and its manifold disruptions, far more than, say, any sort of “depolicing” response to the George Floyd protests, but that causal question still remains open.) There are two aspects of the decline, however, I want to highlight here: why the speed of the decline should surprise us, and one underappreciated explanation for both the 2020 spike and the 2023 drop.

First, it’s easy to think that a major cause of the 2023 decline is just some combination of the worst of COVID-19 passing along with a big dose of reversion to the mean. But recent work by sociologists such as Northwestern University’s Andrew Papachristos suggests that homicide spikes should be hard to reverse, even when the underlying causes have passed. These sociological studies have examined how lethal and nonlethal shootings get “transmitted” through social networks, and their findings indicate that any given shooting can lead to a chain of retaliations and counterretaliations over several years. One study of more than 11,000 gunshot events in a network of nearly 140,000 people in Chicago, for example, found that nearly two-thirds of all the shooting incidents could be tied to an earlier shooting being “transmitted” through the social network. On average, every shooting produced nearly three more shootings down the line; in one dramatic case, the researchers estimated that one initial shooting led to a cascade of 469 further shootings over several years.

These sorts of social transmissions imply that any sort of homicide or shooting spike — and 2020 included both — could lead to sustained increase in shootings for years to come. And the magnitude of the surge in 2020 led to a significant increase in the number of potential social-network pathways available to “transmit” more shootings. The fact that homicides increased even more in 2021 (albeit more slowly) only further amplified the number of pathways. I myself predicted in 2021 that these sort of social networks made any sort of homicide rate “sticky,” and that we should therefore expect declines to be slow even as COVID-19 waned — and I appear to have been wrong about that. Surely some ongoing transmissions explain why homicide rates still remain above the near-historic low in 2019, but the degree of the decline, and its rapidity, are noteworthy. Some cities, including oft-maligned places like St. Louis, even saw returns to historic lows in 2023.

This, of course, raises the serious question of “why?” What happened that caused an unexpectedly quick reversion to not-quite-but-surprisingly-close-to-pre-2020 homicide levels? I don’t have “the” answer myself, but the second thing I want to do in this piece is point to an important, and unappreciated, variable that criminologist John Roman has identified — and one that we should incorporate more rigorously into how we analyze crime trends not just in the post-COVID-19 era, but to those in previous times as well.

Roman’s claim is that a major driver of both the homicide spike and its decline were the trends in overall government employment — not the police, or at least not just the police, but all government employees. Roman notes that, thanks to the pandemic, the total number of government employees fell by about 6% in 2020. Given that there were around 22.8 million government employees at the end of 2019, that’s a decline of about 1.4 million people. Government employment began to recover by mid-2020, and by the end of 2022, it was only 3% below its 2019 levels; by the end of 2023, government employment had more than fully recovered, with the 2023 year-end government labor force actually larger than it was in 2019.

It’s important to appreciate that many of these government employees — the ones who lost their jobs right before homicides spiked and the ones who regained their jobs as homicides fell — are intimately involved in programs that reduce crime, even though we don’t think of them as “crime prevention” programs. All sorts of nonpolice programs, like after-school programs, drug treatment and mental health counseling, housing assistance, etc., have documented crime-reducing effects, and all suffered from significant cutbacks in the early days of the pandemic.

Now, it’s true that, unlike the police, crime reduction is not many of these employees’ primary jobs, but it is also important to note just how many more of them there are. FBI data, for example, reported that in October of 2019 there were around 715,000 total sworn police officers and 1 million total police employees nationwide. In other words, the reduction in total government employment was larger than the total number of people employed by all police departments. Moreover, that government employee decline was not driven by trends in police employment, since despite all the claims to the contrary, the ranks of police departments did not decline over 2020, but actually rose by about 1.5% (although national aggregates can hide important local variation). It’s true that sworn police employment fell slightly in 2021 (by about 3%, or 20,000 officers total), but it then rebounded in 2022 to levels not seen since 2012.

In other words, while reducing crime is not the main focus of, say, schoolteachers and health care providers, their sheer numbers suggest that their impact on crime in the aggregate could be quite significant. This, in turn, would imply that their sharp decline in employment in early 2020 could have played a major role in the broader destabilization of that year — and that the almost equally sharp increase in (re)staffing may have staved off the potential vicious cycle of retaliation noted above. Correlation is not causation, but this is a theory worth further testing.

If Roman is correct that we are undervaluing the impact of general government employment on pandemic-era crime trends, that would suggest that we are likely undervaluing its impact in other periods as well (a point akin to the one raised in Patrick Sharkey’s “Uneasy Peace,” that the statistically invisible work of local activists and public-private partnerships played a significant but overlooked role in driving down crime in the 1990s). In turn, this would mean that the insights we draw from trying to understand our current, chaotic post-pandemic present could help us reassess other eras as well.