Why has crime declined? Part Deux
Also, infusing your statistical jargon with meaning
External Processing is an owner-operated venture that is free to all and supported only by your goodwill. Please pass the word.
I am struck that crime continues to plummet in an almost frictionless way with no obvious mechanism for propulsion. It is human nature, I suppose, to dwell on failure and pass quickly by any success, but there is something to the present moment that deserves far more attention, even in the face of a barrage of distractions. Today, I am exploring what local leaders cite as their secret sauce, and there are some common threads worth tugging on. Separately, I spend a minute chewing on the wave of slightly offkey statistical jargon that is increasingly infusing our daily badinage and offer some thoughts on looking for meaning rather than significance when we evaluate the performance of the world around us. Finally, a tip of the hat to serendipity and why you should talk to strangers on planes.
What is the German word for a Breakthrough You Didn’t See Coming?
Crime continues to plummet in the US in 2025. I am happy to note this since I forecasted this earlier this year (at least a continued decline in the first half of the year and then a leveling off in the second half.) I was admittedly vague on what is causing these historic declines. So, what are the cities saying is the cause? To find out, I took the extremely scientific approach of putting ‘crime drop’ into Google News and then the names of cities in whatever order occurred to me. These are their stories.
San Francisco Mayor Daniel Lurie says it is their Real-Time Investigation Center (RTIC) where “teams of analysts monitor live surveillance feeds, license plate readers, and drone footage to guide officers on the ground in real-time.”
New York City Police Chief Jessica Tisch says it is, “precision policing strategies" that involve placing officers in areas experiencing the highest concentrations of crime — known as Violence Reduction Zones. NCJA reports that “major crime in these zones dropped 25% compared to the same period last year.”
Louisiana State Police credit the crime drop in New Orleans to Troop NOLA, a “specialized unit created this year to assist the understaffed New Orleans Police Department” that works in partnership with Project NOLA, “as more than 5,000 cameras installed throughout New Orleans. Some of the cameras use artificial intelligence to control their movements, and others have thermal imaging capabilities.”
In Philadelphia, the reporting shows that “homicide incidents this year to date, police data shows, are down about 34% over this same point in 2024, and shooting incidents overall are down 23% over the same time period.” District Attorney Larry Krasner says this is the “largest improvement in homicides in the city's entire history. OK? The city's entire recorded criminological history." He notes that jail populations are down significantly and credits “continued support of anti-violence initiatives” and not “locking everyone up.”
In Los Angeles, LAPD Chief of Police Jim McDonnell describes that “a 14% reduction in homicides and a 19% drop in shooting victims are significant strides in our efforts to reduce crime. These improvements are a direct result of strategic policing, targeted enforcement, and the invaluable collaboration with community organizations dedicated to violence prevention. “
In St. Paul, they report a focus on nonfatal shootings and street outreach is driving violence down.
By the way the German word is Überraschungserfolg, which is a surprise success. Or maybe, Glücksfall, which is a lucky outcome. That seems to be where we are. How do we turn this into something sustainable?
Thinking of Crime Declines as Productivity Gains
I think three ideas permeate all of the answers. One is that cities are focusing efforts on violence reduction, at least in the places where violence is most widespread. This hardly seems revolutionary, but historically, there is a lot of evidence that this was not always the case. In the very recent past, many cities were slow to acknowledge that the COVID violence spike was real, substantial, and enduring. And historically, there are real horror tales of cities virtually ignoring the places with the most violence; see, for instance, Jill Leovy’s gripping Ghettoside.
Another consistent theme is that of experimentation. It is clear that cities are trying new things. A broader question—can we learn anything from these demonstrations—is less clear because there is little information about whether cities are formerly testing these innovations. And, if I can take you shuddering and muttering back to your college economics courses, you might also note that these cities are experimenting with at least two of three of the factors of production that improve productivity: labor and technology. And the appetite for big, possibly iatrogenic capital investments like Cop City in Atlanta seems to be waning.
The third theme is about what is missing in these descriptions. I find few references to police preventing crime. This is good—the evidence that police prevent crime is pretty scarce. As I’ve written elsewhere, while it is very clear that police can deter crime (by increasing the perceived risk of getting caught committing a crime), it is much less clear that police prevent crime. Prevention, after all, is specifically about reducing risk conditions and strengthening assets. Police can do these things, of course, but they are very low on the modern policing priority list.
By contrast, community violence intervention (CVI) is explicitly about prevention and not about deterrence. Investments in CVI have grown substantially in recent years (increasing the capital and labor factors of production), and at least some cities partially credit CVI with their reductions.
Why am I harping on the very boring factors of production here? Well, something is happening, and productivity gains in public safety seem like a very reasonable place to look for an explanation. But the big buckets that lead to productivity gains: more labor, more capital, and more technology, seem to explain only a bit of the gains on the policing side. There is some evidence that technology is being fruitfully employed, and a little bit of labor investment. But the labor side is all about substitution, the reshuffling of police from one set of duties to new duties in violence-prone areas. Pull back the lens, and there is a huge police labor shortage that shows no signs of slowing.
The place where there are both big capital investments and big labor investments is on the CVI side. While few people seem interested in becoming police officers, there are compelling numbers of people willing to work on CVI. Some of this is street outreach workers—people with lived experience directly intervening after shootings or in anticipation of a shooting. More of it is on the social work side. And there is increasing capital investment in physical spaces—mediation of vacant lots, beautification of public spaces, and direct investments in facilities to support CVI.
All of those investments have strong support in economics theorizing about why crime is declining.
On the flip side, it is interesting to subject policing to the same factors of production analysis. I would argue that much of the recent investment in productivity-spurring factors in policing is on the tech side. Things like body cams, GPS, and, more recently, AI. I’ll have a lot more to say about this because the trade-off in safety versus civil liberties will likely be what we talk about over the next decade.
Recent investments in policing technology are likely to be largely tapped out in terms of productivity gains. There have been big investments in better vehicles, better weapons, and better uniforms. It is not clear whether this has improved policing as measured in crime declines, nor if it has fulfilled the even more direct effect of improving officer health and safety. But regardless, it is hard to imagine further widespread investments in bigger and faster cars, weapons, or uniforms.
As noted above, labor has, if anything, declined some over the last decade. Police report labor shortages, and it appears to be widespread. The solution of reducing requirements for eligibility, including lowering standards for admission to academies, may help, but the long-term implications are not positive.
On the capital side, many departments invested in remote policing stations and equipment to police recurring hotspots like clusters of bars and large events. One could imagine more of these investments, and bigger investments in the form of greater capacity for crime labs, for instance. But it is not clear that capital investments akin to server farms to pump out freshly minted Bitcoin or to fertilize digital fields for AI plowing are something that police want or need.
In Closing
My take has been, and remains, that when cities are considered individually, they’re special, but not unique. Each city needs solutions that are tailored to its problems. But the problems are likely to be pretty universal, and the trends pretty universal as well. Almost every place in America experienced a spike in violence during the COVID period, and almost every place has seen that tide recede, and often substantially.
And almost every place would benefit from adopting evidence-based prevention and intervention strategies. Again, these need to be tailored to the local context, but on average, there is no need for each city to develop its own evidence of what strategies are effective and which are not. The devil is in the details for sure, but most places would benefit from figuring out how to implement approaches we already know are generally effective rather than rolling out something new and unique.
That is also a long-winded way of saying that while we do not know specifically why crime is declining, we do see it everywhere, and that is an important starting point for the investigation. Yes, there is a tremendous amount of variation across cities, but it is variation around a broad trend, not a trend of its own.
One Good Thing
We are in Austin for a bit of this week. And last night, we ventured under the Commerce Street bridge to witness the emergence of the bats at sunset. All 1.5 million of them. In this jaded age, there aren’t many things left that the internet hasn’t ruined. More than a million tiny bats emerging at the gloaming turns out to be really hard to capture digitally. But just fantastic to see live. I give the bats two hearty thumbs up.
But he Drooped, the Ennui
OK, please stop saying this, from Axios: “While a focus group is not a statistically significant sample like a poll, the responses show how some voters are thinking and talking about current events.” There are no statistically significant samples. It doesn’t matter if you are looking at a poll, a focus group, or an interview with your mom. Statistically significant samples are not a thing. I’m not picking on Axios. I see this all over the place. My sports talk radio is full of talk of sample sizes. I hear people choking on the phrase “statistically significant” all over the place. It’s like it's been forced into regular conversation, and people feel obligated to say it.
Interviewer: Professor, what does it mean when we say “statistically significant”?
Professor (quietly): “No one knows.”
OK, yes, we do know exactly what we mean when we say statistically significant. The problem is that it doesn’t mean what we want it to mean. And what we want it to mean isn’t really the thing we actually want it to mean. What we want ‘statistically significant’ to be is an idiom where the phrase means “important change.” And we want it to be an idiom, because idiom is fun to say, and also has a definition which is not quite what we think it is. An idiom is a phrase whose meaning is not deduced from or necessarily dependent upon the specific words in the phrase. So if, as an idiom, “statistically significant” just means “important change,” then it does not matter that “statistically” and “significant” mean something else. But taking a specific phrase and making it mean something else, something less precise and less meaningful, is a missed opportunity. So let’s not do that.
I suppose this calls for the very boring business of explaining statistical significance in a way that does not cause readers to swipe away. But I shan’t. I acknowledge that idioms are gonna do whatever idioms are gonna do, but it would be outstanding if we could focus this particular idiom on the ‘important’ bit and not the ‘change’ bit. And by that, I mean, we should have less focus on whether we have sufficient volume of information about the thing we are investigating to detect a change, but rather whether the change we are trying to detect is meaningful.
What really matters in this statistical significance business is, how big is the effect size you are trying to detect? Another way to say that is, how a big a change do you think there is?
If 92% of the people were for something and now 84% are for it, so what? Almost everyone was for it, and now, almost everyone is still for it! If I said to you, I held a focus group in January and 11 of 12 people were for publicly funding a new stadium and I just did another set of interviews with some different people and now 10 of 12 people are for publicly funding a new stadium you’d say, huh, people really like to pay for new stadiums. If I interviewed 3,000 people and got the same results you’d have the same reaction. Even though the results of the poll are probably statistically significant and the results of the focus group are not (which is what the mangled Axios header was trying to say). The point is, essentially 1 in 12 people changed their minds in both cases, and that’s a small change. It’s not meaningful.
What’s vastly, infinitely, more interesting are big changes. I’ve met Rich Thau who runs the Engagious focus groups of swing voters that Axios is marbling about, and he’s an interesting guy, doing an interesting thing: asking people with apparently no fixed ideology how they are responding to the events of the day. But what I look for from his focus groups are big meaningful changes, like when his groups go from 11 for and 1 against, to 6 for and 6 against. That’s when I want to see some poll results with a larger sample to see if the hypothesis he has generated holds more generally. Because that’s the value add from these focus groups, uncovering nuggets for further exploration.
Also, I would point out that one thing tracking polls—whether it is the New York Times or some other pollster—never do is tell you whether the change from the prior poll to this poll is statistically significant. Because they never are. Public opinion doesn’t move like that over a day or a week. If a candidate goes from +1 last week to +2 this week, that’s almost certainly not a statistically significant change.
What matters is the trend. And to understand whether a trend shows a statistically significant change, what you care about most is how many polls have been conducted over how long a time frame, rather than how many people are polled. In that case, small changes can have huge import. If your candidate has a 51-49 edge, that’s a nail-biter election. But if the trend is moving their way, and they can move just four people from the other candidate to their camp by election day, now it is a 55-45 blowout. That’s where the meaning is, in the size and duration of the trend.
So, the point is—as it always is, in everything—to think first about meaning, then about statistics. How big a change does there have to be, in this specific context, for the change to be meaningful? And there is no statistical test for that.
Musical Interlude and the Unfettered Joy of Talking to Strangers
A couple of weeks ago, in an unusually gregarious moment, I struck up a conversation with my seatmate on a tight commuter jet on its way from here to there. He turned out to be the tour manager for The Fray and spent two hours patiently answering my questions about life on the road. Tip of the hat for an afternoon well spent. This is also a song that unites the generations, with deep nostalgia for me, and apparently a hype song for my teenager.