Why Does Crime Go Up When School Starts?

Subscribe here: Apple Podcasts | Spotify | YouTube | Overcast | Pocket Casts

Back-to-school is supposed to bring relief. Relief to parents who work and have to figure out child care in the summer. Relief to kids with bad home lives and those who rely on school for meals. And relief that kids will be supervised, no longer free to, say, shoplift or vandalize.

A widespread narrative that criminal behavior peaks in the summer months has long been thought to include criminal behavior by children. But new research indicates that children ages 10 to 17 are most likely to be involved in a reported crime right after they get back to school and right before they are let out for summer.

On today’s episode of Good on Paper, I talk with the economist Ezra Karger, who wrote a 2023 paper revealing that, unlike for adults, kids’ criminal activity doesn’t peak in the summer. Along with his co-author, Todd Jones, Karger reveals that when back-to-school time hits, kids are being arrested for behavior such as simple assault, drug crimes, and sexual assault—raising questions about whether school is creating the conditions for criminal behavior and victimization.

“So the conclusion we came to while digging into this paper is that taking a bunch of 10-to-17-year-olds and putting them in a large building—where they’re interacting with their friends, but also maybe people who aren’t their friends—that is leading them to be engaged in crime that is reported to law-enforcement agencies, and that is leading them to be arrested at higher rates,” Karger explains. “And we have a lot of analysis showing that this relationship is causal, that these patterns occur exactly when school is in session, that they don’t happen on weekends, that they don’t happen over the summer.”

The following is a transcript of the episode:

[Music]

Jerusalem Demsas: We’re used to thinking of the ways that school solves problems: It’s a place to learn. Schools provide lunch, access to teachers and staff who keep an eye out for signs of distress. And for kids with troubled home lives, school is an escape.

But over the past few years, there’s been a steady flow of new information complicating the uncomplicated idea that school is obviously good for kids. Alongside all these benefits are some pretty serious costs. For instance, there’s research showing suicides spike during the school year. And also, that some kids fared better during remote schooling.

This episode is about whether school can create problems, even as it solves others.

My name’s Jerusalem Demsas. I’m a staff writer here at The Atlantic. And this is Good on Paper, a policy show that questions what we really know about popular narratives.

[Music]

What time of year do you think kids commit the most crimes? I’ll give you a second to think. Summer? Winter?

Before I came across the research in today’s episode, I had a vague sense that kids, like adults, got up to the most trouble during the summer months. After all, they’re more likely to escape the supervision of adults when school is out. And whether it’s shoplifting at Sephora or tagging a highway underpass, it’s hard to find the time when you’re meant to be in school eight hours a day.

But I was wrong.

My guest today is Ezra Karger. He’s an economist at the Federal Reserve Bank of Chicago and, along with his co-author, economist Todd Jones, published new research that challenges a popular narrative about kids and crime.

It turns out that while for adults, summer is when criminal activity peaks, for kids 10 to 17, back to school—and right before school lets out for the summer—is when this age group is most likely to be involved in a reported crime.

Ezra helps me dig into the data, and we get into what exactly might be going on and the limitations of studies that look at reported crime.

All right. Ezra, welcome to the show.

Ezra Karger: Thank you for having me.

Demas: So we’re here to talk about a paper you recently put out, but before we get into that, I wanted to take a step back and ask you: What does it mean for crime to be seasonal?

Karger: Great question. I think when people think of the seasonality of crime, they focus on the fact that crime happens a lot during the summer and less during the winter. And there are a lot of opinions about why this might be happening: Maybe people are stuck inside when it’s cold out. Maybe people get really riled up when it’s hot out.

Criminologists are still debating why we see the seasonal patterns we see, but I think of the seasonality of crime in general as meaning: During the year, crime is higher at specific points than other times.

Demsas: There are a bunch of different theories. What is the theory that you find most compelling?

Karger: I find the temperature theory pretty compelling.

Demsas: Okay.

Karger: But on the other hand, you also sometimes see seasonal crime patterns in areas where there isn’t as much temperature variation. And so I don’t think that explains everything that’s going on, but I do think this high amount of crime during the summer is really interesting, and it’s probably deserving of additional research by economists and criminologists.

Demsas: And when we say, “the temperature,” are we literally saying, It makes people hotter, and hotter people are angrier?

Karger: That’s my lay interpretation of exactly what’s going on.

Demsas: (Laughs.) Okay. Yeah.

Karger: But I want to make sure I don’t step on criminologists’ toes. And I think they have many thousands of pages of papers about whether this might be what’s going on or not. But yes, I think that’s exactly what’s happening. I think people are outside more. People are getting more angry in the heat, and that just leads to more reported criminal offenses and more arrests.

Demsas: I know that—I mean, you just said right there—“more reported criminal offenses,” and I think that that’s going to be hanging over our entire conversation, but there’s a lot of concern with crime reporting, right? Because there are reasons why crime reporting might vary that has nothing to do with whether or not crimes themselves are increasing or decreasing.

There’s some research indicating that when there are prominent instances of police brutality, people may become more loathe to report to the police because they just say, Well, this is a corrupt institution. They might shoot someone if they come here. They may have fears around what that reporting looks like or if it’s valuable.

But also, you may not report crime, because the people committing it are people you care about. You don’t want them to go to jail. Maybe you don’t report your kid if they commit a crime—something like that. So how do you think about that when you’re doing this research? How do we control for those problems?

Karger: Yeah. There are many datasets that people use to track crime, and some involve relying on the law-enforcement agencies to tell us how many crimes there are and what type of crime is happening. And others rely on asking people if they’ve experienced specific crimes. And I think what’s really interesting in the paper that we’ll talk about soon is: We find the same patterns in both of those data sets.

And so when we think about whether you’re looking at the victimization side or the law-enforcement-agency reported side, I think if you’re finding patterns that match in both of those datasets, you can be more confident that we’re not just seeing something that relates to reporting bias.

And so there are several papers recently looking at how reporting of crimes varies over time and varies as a function of what else is going on politically or socially. And I think those papers are really important. And I think this is why having data on asking people whether they’ve experienced criminal behavior—either as the victims, as the offenders, as other members of this group that’s involved in crime—is really important.

Demsas: So before we get into your paper specifically, what do we know broadly about when kids commit crime? I had this general sense, alongside the kind of seasonality stuff you just talked about, that kids commit more crimes in the summer and also in those hours between when school lets out and their parents get home from work.

And I was trying to track down where this idea I had came from— I’m sure people were just saying this to me. It also seems like police officers will often talk about this. But there’s a 1996 paper that’s produced by the Department of Justice’s Office of Juvenile Justice and Delinquency Prevention that shows that on school days, juvenile violence peaks between 2 p.m. and 4 p.m., and they don’t see that similar pattern on nonschool days.

Looking closer, it appears that data is just from South Carolina. But it seems logical. What do you think about this?

Karger: Yeah. Coming into this question of when juveniles are involved in crime, I had exactly the same belief about what we would find. I thought there’d be more crime maybe after school, when school gets out, but before children are supervised. There’d maybe more crime during the summer when children might be hanging out more. They might be less supervised by their parents who are working. And I don’t think there was a consensus that I saw in the criminology literature and the economics literature about what patterns this would follow.

There are some papers in criminology looking at, you know, National Crime Victimization Survey or data on crime that said, Well, it looks like maybe we see different patterns of seasonality at a monthly level for children relative to adults when we think about when crime is happening and when crime is being reported.

But when I talked to people who hadn’t yet seen the results of our paper, they definitely thought, Yeah, when children are unsupervised, when they’re out and about during the summer—that’s when we’re going to see higher rates of crime involving children.

Demsas: Okay. Let’s turn to your paper. It’s called “School and Crime,” so we know exactly what we’re getting into. So you and your co-author confirm the sort of overall criminal activity peaking in the summer, but then you look closely at the 10-to-17 age group. What do you guys find?

Karger: Great. So we take this data from NIBRS, the National Incident-Based Reporting System, and we ask whether this pattern of seasonality involving crime is different for people of different ages. That was our first question coming in.

And so Todd Jones, my co-author, had this idea of just looking at the raw data and seeing if anything popped up. And we dug into the raw data, and what we saw is, for adults, there’s this clear summer peak of crime—reported crime, reported arrest rates—that’s higher in the summer for crime involving adults.

But for children, we really saw the opposite pattern. We saw that crime was at its lowest level—and rates of arrests were at their lowest level—over the summer. And then once September, October rolled around, we saw crime involving children peak as either victims or offenders. And then in the spring, right before school got out, we also saw really elevated levels of crime relative to the summer, right after school had ended.

And so that was really the entire paper—that one fact. And then the other 80 pages are trying to dig into all of the underlying data and figure out whether this is reflecting a causal relationship, whether we trust these results because of problems with the reporting bias and other things that might be going on with the data.

Demsas: Why would this be happening? Why is it that we don’t see that kind of difference? Obviously, it’s a difference with the 10-to-17 age group.

Karger: Yeah. So the conclusion we came to while digging into this paper is that taking a bunch of 10- to- 17-year-olds and putting them in a large building—where they’re interacting with their friends but also maybe people who aren’t their friends—that is leading them to be engaged in crime that is reported to law-enforcement agencies, and that is leading them to be arrested at higher rates.

And we have a lot of analysis showing that this relationship is causal, that these patterns occur exactly when school is in session, that they don’t happen on weekends, that they don’t happen over the summer.

We don’t have a lot of data, because of the source of our underlying data for this paper, on why this is happening. So one thing we’re pretty agnostic about in this paper is the mechanisms. We are not going to be able to say, Well, it’s because of how recess is structured, or, It’s because of how teachers interact with students and send them to the school police officer, if there is a school police officer in their school. We are not going to be able to say anything about that.

We are just pointing to this striking regularity in the data, which is that children are much more likely to be involved in crime during the school year, during the school day, and when they’re in school, specifically.

Demsas: And are these crimes happening at school, or do you know where these crimes are happening?

Karger: Yeah. One amazing thing about the NIBRS data is they have a measure of whether the crime is happening in school or not. And we do see that this pattern is almost entirely driven by crimes happening in school.

We also have data from New York City, which has very clear geographic markers, like longitude and latitude pairs, of where each crime is occurring. And in New York City, we also can see that the crime that we’re measuring here and the patterns we’re measuring here are really taking place within a very small radius around the school.

Demsas: What types of crimes are happening? What sorts of things are we mostly observing?

Karger: What I found pretty shocking about this paper is we find this relationship for most types of crime. So this is not just driven by drug crimes. It’s not just driven by assaults. It’s driven by a lot of crime types that we can measure in the data.

Maybe to point to specific types of crime where we see a very striking effect, we do see that drug crimes peak during the school year. We also see that simple assaults—when we say assault, I don’t want you to have the same idea, maybe, in your head about a typical assault involving a 30-year-old, right? This is an assault that’s reported to a law-enforcement agency, but if it involves a 10- or 11-year-old, it might be less of a severe assault than what we’re thinking about with adults. But we see that assaults—

Demsas: What do you mean? Just, like, shoving another kid would send you to jail?

Karger: That’s a good question. We are collecting data here from many law-enforcement agencies. If two kids shove each other, and it gets reported as an assault, that is in our data as an assault. If two kids shove each other, and then they get arrested, that’s reported as an arrest involving an assault. And I think that’s important because when you think about what goes on their record, that’s going to be an assault, whether it involved a knife or just shoving.

And so we’re relying on the law-enforcement agencies, as we do in society, to define what these crimes are, for the purposes of reporting, and to define what’s happening with arrests. We aren’t measuring convictions or incarcerations. That’s something that’s farther down in the carceral pipeline that we’re not going to look at in this paper. But I do want to take that into account, where we’re really relying on what people have experienced and what gets reported as our measures in this paper.

But we see this, really, across many types of crime. So weapons-related crime, even property damage shows a weaker but somewhat similar signal. We don’t see this pattern as much for theft. And so that might get to this idea that some types of crime don’t face the same seasonal patterns. Maybe it’s easier to steal things when you’re not in school.

Demsas: Yeah.

Karger: But we do see this pattern for sexual assault. There are some other crime types that are very serious, which, thankfully, we don’t have a lot of in our data. So we don’t measure these patterns clearly. Like, you know, murders or killings, we don’t see this pattern for. And it’s unclear whether that’s because we don’t have a lot of data on them because, thankfully, 10- to 17-year-olds aren’t engaged in a lot of those types of crimes. But also, it might still be in the data if we had more power.

Demsas: One thing I’ve heard from criminologists often is that in order to make sure you’re not having reporting problems, you want to track murder rates, mostly because it’s very unlikely you’re going to conceal a dead body. It’s harder to do that. People disappear. Their body gets discovered. You’re going to have a pretty accurate sense of murder, even if you may not have an accurate sense of property damage or things like that.

Are you concerned that, given that there’s not the spike in murder that you’re able to observe, that maybe it’s a reporting thing that’s happening rather than an actual shift in the crime rate?

Karger: Yeah. This was our main concern about this project. So Todd and I spent a lot of time working with the NIBRS data, and we were very confident that in the NIBRS data, we were seeing these patterns. And so the obvious question is reporting bias. Maybe what’s happening is that when kids are in school, more of the crime that is already occurring is being brought to the attention of law-enforcement agencies.

Before digging into how we can check whether this is true or not, I want to talk about whether that’s important. You might care a lot about whether crime is occurring, but you also might care about whether it’s leading to reported crimes and arrests because arrests are something that affect how kids interact with law-enforcement agencies, and they affect what’s going on your record, right? If that arrest is leading to other things, those dynamics can be really important for children. And so I want to start by saying, whether or not this is reporting bias, I think it’s really important.

But we did want to see if it was reporting bias. And so what we did is we dug into this National Crime Victimization Survey, and we asked whether people who are 10 to 17 years old were reporting being the victims of more crime in the months when school was in session, relative to the months when school was not in session.

Demsas: Okay.

Karger: And what we saw was that if you look at adults in the National Crime Victimization Survey, you see exactly the seasonal patterns we expected to see: Reported victimization rates peak during the summer. But if you look at children, at 10- to 17-year-olds, you see exactly the pattern we saw in NIBRS. You see reported crimes involving children, where we’re measuring this based on the victimization reports, peak during the school year.

And so that really convinced us that, while some of this effect might be driven by reporting bias, these patterns are not entirely reporting bias. There are clear increases in crimes that are occurring, according to the self-reported victimization surveys, during the school year, relative to the summer.

Demsas: Gotcha. I guess also, about victimization then, too, do you find that children in this age group are the ones who are the primary victims of these crimes? Are their victimization rates also increasing seasonally with the school year?

Or I guess it’s possible that they’re committing crimes against adults, right? You could be having that happen with teachers or support staff or other people in these schools. So what do you find about victimization rates?

Karger: Yeah. I was really interested in who was being reported as committing these crimes and who is reported being the victims of these crimes. And I think these crimes are, really, students affecting other students. There are also, let’s say, drug crimes where you don’t have a victim, per se. Those are considered victimless crimes, but you do see those also change.

But let’s focus on something like assaults. So we can measure in the NIBRS data whether these are occurring between friends, acquaintances, family members, romantic partners. The data is incredibly rich. And what we see is that the victims and offenders are often very close in age. They’re often exactly the same age. We see this pattern where kids who are exactly the same age are more likely to be in this sharp increase in crime—reported crime and arrests—relative to kids who are very different in age.

Like, 17-year-olds are not assaulting 10-year-olds. You usually have 13-year-olds assaulting 13-year-olds in this data or 14-year-olds assaulting 14-year-olds. And this is consistent with an idea about the mechanisms, where you’re interacting with students who are in your grade. You’re interacting with students who are in your elementary school or in your middle school or high school. You’re not interacting with kids who are very different from you in age, and you’re not, in this data, assaulting 40-year-olds who happened to be stopping by the school. This is really about kids who are in school together and in this large building with other people like themselves.

Demsas: One thing I wanted to go back to is what you said about how you expected to find this kind of seasonality. So why did you even look into the data, then, if you were not really expecting to find anything novel?

Karger: Yeah. There are a series of papers showing that the school year really matters for children’s outcomes. There’s a paper about cyberbullying showing that during COVID, searches for cyberbullying-related topics dropped. And we think that’s very correlated with actual experiences of cyberbullying.

There’s also a paper about suicides that I find very sad, which is that during the school year, suicide rates among children increase. And so one of the reasons that I came into this paper with a prior that I would see the common seasonal patterns, but I was also curious about how taking children and putting them in this system—which has them in a building, has them interacting in very specific ways—I was curious if that would have different effects on reported crime and arrest rates for different types of crime, different types of arrests. And I wasn’t a hundred percent confident that I would go in and see the high peak in the summer.

Demsas: One thing I thought about when I was reading this paper was: Is it that these crimes would not have happened if school was year-round, or is it like the kids are, like, saving up the crime that they would have committed over the summer and then just doing it all right when the school year happens, or the school year commences? Is it that there actually would be less crime if kids were just not in school at all, or it would all be normalized if kids were in school year-round, or is it actually creating more crime?

Karger: There’s this amazing paper by Brian Jacob and Lars Lefgren, which looks at teacher prep days. So these are days during the school year when school is randomly closed. And they use these days to look at effects of school being closed on property crime and violent crime. And then they also try to check for substitution effects. They look at exactly what you’re talking about. And they find very little evidence that crime is then substituted to days when those kids are not in school nearby.

Our analysis strategy is not as well set up for trying to measure substitution effects, because we’re really using the school year as this shock—this big thing that’s happening during the year—and it’s much harder to say what would happen in the absence of a school year or if school was all year-round. That’s a huge change to policy that we can’t really generalize to.

What I will say is: I don’t think this is substitution. At least, I don’t think most of this is substitution. That’s my personal belief based on the data we’re seeing. And the reason is: If you take what is going on with crime involving children during the summer or during after-school hours, and you try to predict how much crime would happen during the school year based on those numbers, you see predictions that are much lower than what we actually see.

And so I just don’t think that’s consistent with children who are, like, saving up crime during the summer and then waiting until they’re there over the school year. And I think it would be a strange model where in July someone decides not to assault someone so that they can assault someone in October.

Demsas: It could be that, you know, someone has really irritated you. And they’re online, for instance, and then you can’t actually see them. And then the first day of school, you jump them. Do you know what I mean?

Karger: Yes. Absolutely. That’s an interesting question, and I would say that my best evidence that that’s not happening is that the elevation of crime does not just happen for the first week or two. It really takes longer. And so if people were saving up those crimes, they’d have to be saving up those crimes and then spreading them out over the first few months of school, instead of just dealing with this problem for the first week. And that seems really strange.

Demsas: Just a bunch of devious teenagers trying to mess with your data.

Karger: Yeah, exactly.

[Music]

Demsas: All right. Time for a quick break. More with Ezra when we get back.

[Break]

Demsas: Just walk me through a little bit about how you actually were able to construct this paper, because schools obviously start at different times of year. You had to normalize all of that? That seems like a very difficult process.

Karger: Yeah. I would say we used three strategies to try to figure out whether this relationship was causal. The first—and this doesn’t work in many papers—is just looking at the raw data. And the raw data was striking enough to make it clear that the patterns we saw for the other 85 pages of the paper were going to hold up. What we saw is that if we looked at reported crime rates involving children, we had this incredible U shape, where they drop during the summer and then increase when the school year starts, and then they stay elevated, and then they drop again when the school year ends.

And we contrasted that with crime rates and arrest rates involving people who are slightly older, so 19- to 24-year-olds or 25- to 30-year-olds. And for those groups, we did not see a U shape. We saw the inverted U shape—the hump of crime peaking during the summer. But then there’s the question of how we can actually estimate these effects causally using methods that economists often use.

And what we did is: We scraped information from a website that gathered school calendars that had exactly when the school district was starting school and ending school. And then we did line up all the school districts. So we had this very complicated process that Todd designed where we linked the school districts to law-enforcement agencies. And it’s not obvious how to do that, because law-enforcement agencies and school districts don’t cover the same area. So we threw out places that didn’t really link up well. We relied on the school districts that did line up well.

And then we said, If we take all of the school years that we have—some of them are starting right after Labor Day; some of them are starting in August. My school in the Northeast always started in September after Labor Day. Our 3-year-old just started school in D.C. It turns out D.C. schools start in mid-August. And so using that variation, we can line up all the schools and see whether crime is peaking the week or two after school starts, relative to the week or two before. And we can, in a regressions-continuity design, try to control for various characteristics of the school district or the law-enforcement agency.

The results are striking enough that we don’t really even need to do that. When you line everything up, you see stark differences between the amount of reported crime and arrest rates, especially at the end of school—so in the week or two following the end of school—relative to the week or two before when you line up all these districts. And you also see those patterns at the beginning of the school year.

Demsas: Obviously, some schools, as you said, start earlier in the summer. Do you see any differences between schools that are starting school in August versus starting school later in September? If we’re talking about your hot thesis, you would expect there to be more crime in those places?

Karger: Yeah. We haven’t dug into that yet. And it’s something I’m really curious about. I want to dig into temperature, maybe try to control for that, compare schools that started earlier or later. One thing that we haven’t talked about yet that is very important is that, thankfully, crimes involving children are not that common. And that means that trying to estimate these effects and splitting the sample up into different pieces or looking at groups of schools that were starting in September or August—that actually isn’t that well powered, because we don’t have that many crimes involving juveniles in our data.

So that makes analysis hard in some sense, but it’s also, I think, a good fact to remember. So we’re talking about big changes in reported crimes and arrest rates involving children, but those are coming from very small baselines. I think about 2,000 arrests of 10- to 17-year-olds per 100,000 people is roughly the rate we were seeing in the data in 2019 nationally. And so that’s a lot of arrests, but when we start to split up the school districts into different subgroups, it becomes much harder to measure these effects, especially when we’re focusing on comparing one week to the week before.

Demsas: So there’s obviously a lot of concern—especially in certain school paradigms—with sending kids to the cops or even reporting kids to the police, in general. Were you able to look at other sorts of reports of student behavior, whether it’s suspensions or anything that indicates that this kid has behaved in a way that might technically qualify for a criminal offense, but they just don’t want to report that kid? Were you able to look at those?

Karger: Yeah. So we were not. I would love to have underlying data on suspensions or expulsions, behavior that might lead to arrests or reported crime that we could then use as a baseline. What happens there is more of a data problem where the schools are collecting that data during the school year, but we specifically want to compare that data to what’s happening during the summer.

And there are no institutions that are collecting reports of misbehavior about students over the summer. That’s maybe one of the main mechanisms driving the results. And so because of that, we don’t have baseline data on that type of question to compare things to. So there are measures of school-district-level suspensions by age. I have another paper looking at suspensions of kindergarten through second graders in North Carolina, where we have very detailed information during the school year of what behavior kids were engaging with that was causing them to be suspended—but that data doesn’t exist over the summer, and it certainly doesn’t exist nationally.

Law-enforcement agencies are this patchwork across the United States who independently report data to some collective bodies, but even the NIBRS data that we use is only covering, during the years we’re using it, about half of the U.S. population. So just getting all of this data on reported crime and arrests accumulated has taken various government agencies decades. And I think it would be great to have data on suspensions. We don’t have that yet, especially nationally.

Demsas: Yeah. I think one of the secret things that people who don’t work on studies don’t realize is how much of it is just attempting to get good datasets. And the best people in economics are the ones who are just lucky or have done the work or able to build the relationships to get that data as soon as possible.

So then, just moving a little bit further than your paper, there’s a lot of research indicating that year-round school would be really, really good for students, that there’s a huge learning loss that happens over the summers, that a lot of students don’t get access to free school lunch or counseling, or they’re locked at home if you’re in an abusive household. There are lots of reasons why it’s better, on average, for students to be in school year-round. If kids were in school year-round, do you think crime rates go up?

Karger: Yeah. That’s a good question. As a parent, I’m very in favor of year-round school because summer sounds like a terrible thing to try to address, logistically. But let’s think about whether it would affect crime rates. So one of the things we try to do in this paper is describe the facts we see in the data. We have no clear policy implications of this paper. This is not a paper that’s saying we should abolish all schools. This is not a paper that’s saying we should extend schools year-round.

There’s other literature looking at what would happen if you extended the school day or if you extended the school year or if you changed when school started and made it start later. I think looking at criminal activity reported or arrests in those papers would be really interesting. My personal belief is that you might, based on our results, see higher rates of crime involving children if you extended school year-round. But I want to be very clear that that doesn’t mean we shouldn’t do it, because, as I said, crime involving children is rare.

I think it’s an important outcome, but it’s not something that people should focus on and only base policies off of. One of the responses we’ve gotten to this paper is people saying, Well, this is evidence that schools are kind of problematic. Maybe we should abolish them or have all of our kids stay home and homeschool them. We are not saying that, right? I think that’s like saying, You have hospitals. You have antibiotic-resistant MRSA for a few people every year. Therefore we should abolish hospitals.

There are other benefits to school. There are other benefits to hospitals. Lots of institutions have many effects, some positive and some negative.

We’re trying to have a puzzle piece there that’s saying this is another negative outcome you should think about when you’re thinking about the benefits and costs of school. And we think policy makers should focus on schools as a key driver of crime involving children and arrests involving children, which we believe is really important. But I would be really wary about extending this to clear policy implications about year-round school.

Demsas: I also want to ask how this interacts with other research that shows that education reduces crime, in the aggregate. You cite in your paper a 2022 paper by Brian Bell and his co-authors. Can you talk about what they find and how you think that then would interact with the aggregate crime rates? Because even if it’s increasing in schools, perhaps, overall, people being in schools means that you have less crime overall.

Karger: Yes. So there’s a large literature looking at school-dropout laws, compulsory-schooling laws, when children are required to attend school, and how many years they’re required to attend school for. And what that literature finds some evidence of is that forcing people to stay in school for longer reduces crime rates, reduces crime rates for the cohorts who were forced to attend school for longer. And so it’s an interesting contrast to our paper. We’re finding, in aggregate—or we’re arguing that, in aggregate—schools are increasing crime involving children and arrests involving children.

And those papers are finding that forcing children to stay in school longer is reducing crime rates. And so this gets to a really important topic when you’re evaluating an economics paper or any applied work that uses causal inference. The papers looking at school-dropout laws are measuring a local average treatment effect on the students who were caused to stay in school by the law.

And so it’s quite possible that the students who stayed in school longer because of these laws that required that they stay in school longer are very positively affected by those laws. Maybe that’s 2 percent of the population, and maybe that 2 percent of the population is half as likely to be involved in crime as they were beforehand. That’s very different from saying that, in aggregate, 100 percent of the children in school are seeing a 30 or 40 percent increase in the crime involving children in school and the arrest rate involving children in school.

And so I think what often happens when you have these papers looking at the causal effects of policies is: They’ll find some effect, directionally, that people then latch on to, and people will then say, Well, based on the fact that this policy has this positive or negative effect on an outcome we care about, we should do it everywhere, or we should do it nowhere.

And economists who think about general equilibrium effects often push against this because they say, Well, those papers, while really important, are focusing on the effect on a specific population at a specific time. And before extending a policybased on the results across the population, you should think about what that will do to the entire population. And that’s not something that these papers can necessarily talk directly about. And so I don’t think that that literature conflicts with our findings here.

I think what it’s saying is really important. It’s saying that, conditional on the setup of school we have in the United States, forcing kids to stay in school longer is having this positive effect on the kids who respond to that policy. And what we’re saying is: The way that school works is it pulls children into a large building, and it causes them to interact in a way that leads to higher levels of reported crime and higher arrest rates. And those two things are both true.

Demsas: If this podcast can make more people think in general equilibrium, I will have done my job. So you’ve referenced a couple of these earlier in your comments, but we tend to think of being in school as being good for kids, but it’s clear there are really heterogeneous impacts, depending on what that kid’s position is, what’s going on with them.

You mentioned the suicide literature. But also, I think the thing that’s been most jarring for me in recent years is learning about the remote-schooling literature and how different it was for different kids. Former guest of the show Chris Campos and his co-authors have a recent paper. They’re building on work that shows decreases in bullying during the remote-school-learning era. And they find decreases in both online and in-person bullying during remote schooling and that parents who were reporting that bullying were more likely to be demanding their remote learning.

So how do you think policy makers should react to the diversity of experiences here? Because, on average, it’s clear, their research shows that being in school has significant benefits, right? The negative average effect of remote learning is clear on reading and in math, but for the kids who are being bullied, it’s actually positive. They’re now doing much better. But often with schools, you’re making kind of blanket policies here. So how should someone think about this? Because obviously if you’re able to target these sorts of solutions at the specific kids you’re most worried about, that could lead to better outcomes.

Karger: Definitely. So we do look at how our effects vary by male and female students or black and white students. And we find very similar percent increases in reported crime and arrest rates involving those children of different types. And I was surprised by that. I expected to see differences. We also tried to figure out if county-level characteristics of a place predicted whether we would see larger or smaller percent increases in crime and reported crime and arrest rates involving children.

And we didn’t really see much of a difference. We didn’t see that there were strong predictors, whether it was poverty rates or income rates or rurality versus urbanity in these places and whether that predicted the outcomes of interest. And so I don’t think our paper gives you a clear population to focus on when it comes to trying to target a specific group that’s maybe more at risk of these increases in reported crime rates, in percent terms.

But there is a question of percent and percentage points. And we do have groups that start off at different baseline levels of reported crime rates and baseline levels of arrest rates. And so you might think—and policy makers focus on this—that some groups are more likely to be involved in crime more generally, that some groups are more likely to be arrested. And you might want to focus policies about how students interact with the criminal-justice system on those students.

But I think your point about the remote-work literature gets back to the general equilibrium effect. I know some parents who were very happy that their kids were not in school, because their kids were not enjoying school. I also know a bunch of parents who got no work done for a year because they were trying to supervise 5- year-olds who were engaged in remote schooling. And so just like we have to deal with those general equilibrium effects when thinking about what policies should or shouldn’t be implemented in school, I think that same comment and type of question applies to the remote schooling-literature.

There’s this question of, Well, does this mean we should give everyone the option to go to school remotely? We don’t know. We just have this one huge shock of COVID, and we can see how it affects people’s experiences when they were in remote school. And I think Chris’s work or other people’s work says, on average, this is not great. For some students, it was good.

Demsas: Yeah. I think one of the more difficult parts of policy making is that you often do have to decide things in averages, and that means that often some people would be better off on a different policy. But you can’t actually make policy for 330 million people very differently, because whether someone else goes to school actually also impacts how your kid does in school. Like, if for some reason, which we’re seeing, you see higher-income parents pulling their kids out of school, that has general effects on the rest of the population. I know you’re resisting the urge to continue to tell us what to do on policy, but I’m going to keep asking you anyway.

I was curious about curfews because, somehow, I feel like this research kind of does interact with the research literature on curfews as well. Every summer, I hear of dozens of cities implementing curfews, and I was looking at D.C.’s, which I did not realize was extremely strict, but we have a year-round curfew. For children under 17, you’re not allowed to be outside at night. During the summer, that curfew is midnight. During the school year, it’s 11 p.m. And I found this paper by Jillian Carr and Jennifer Doleac, who is a listener of this show. So hi, Jennifer. They have a paper about juvenile curfews and urban gun violence, and they find that curfews are actually counterproductive.

They’re looking at the D.C. curfew, and, as I said, during the summer, it’s midnight; in the school year, it’s 11 p.m. So they look at that 11 p.m. hour when school gets back into session. They find that gunfire is increasing by 150 percent during that marginal hour.

So this kind of interacts with what you’re talking about here around how we think about keeping kids monitored or inside or somewhere at school or at home or wherever as being better for crime rates. But here it seems to be a substitution effect that is happening here. I don’t know what you think is going on in that space.

Karger: Yeah. So I think that’s really interesting. It gets back to exactly what we were talking about with compliers and the school-dropout laws. So a curfew can’t really affect the trends we’re showing in this paper, because a curfew won’t affect the children who are actually in school during the day, right?

So which type of children does a curfew affect? A curfew affects children who are thinking about whether to be outside between 11 p.m. to midnight, in the case that you’re talking about. So what we’re saying in our paper is that a significant fraction of the crime involving children happens during the school day, during the school year. That’s a separate policy question. I think it’s unrelated to the curfew question.

But if we try to think about curfews, there’s a series of papers, including the one you mentioned, showing that curfews can have this counterintuitive effect or maybe can have some positive effects—I’ve seen a few papers, I think, of each type. And what should we learn from that? Well, we can maybe learn whether we should implement more curfews on the margin or not.

But we also know from the results from our paper and other work that that won’t have a huge effect on total crime involving children. That’s a very small fraction of the crime involving children. And so when you talk about a 150 percent increase, the baseline level there is small, relative to the total number of reported crimes and arrests involving children.

Demsas: For me, my big takeaway from this is: First, the number of crimes that are involving children are so low that this should not be, potentially, the variable we’re most interested in when we’re thinking about whether or not kids should be in school at different times of the day.

And it also reminds me: I wrote an article recently about the impact of remote schooling on parents, and there was a paper that came out recently, and they found that there was an increased antidepressant use by mothers and increased alcohol use in school districts or in counties where schools were shut down longer.

And it makes me think about how much education policy is—we often talk about it explicitly as being for the kids, but often the unspoken thing is that it’s really also about everyone else. It’s also about parents. And then, of course, it’s generally the broader question of: What crimes are we mostly concerned about? Are we that concerned about the drug deals happening in schools? Are we that concerned about, maybe, a scuffle that happens in the hallway? Or when we think about crime involving juveniles, is it the property crime that we’re mostly concerned about? Are we upset about shoplifting that’s happening outside of it. And making that policy feels, often, not really focused on the individual well-being of those students.

But this has been a great conversation, Ezra. Thank you for coming on the show. I learned before the show that you were a super forecaster, which means that our last question will, I think, be really good for you. So what is something that you thought was good on paper but didn’t pan out in the end?

Karger: Yeah. I was trying to think about forecasting, but maybe I’ll mention something that relates to our conversation about policy and causal policy effects. So I’ve worked on several projects where we can’t really find the data we would need to estimate effects that I believe are really important. And so just to mention two of those, I started this project with a co-author about ambulance systems, emergency-medical systems, trying to figure out what the rollout of those across the United States did to health.

And it turns out that the law that helped implement some standardized national EMS systems also was the one that started to collect the data that you would need to evaluate whether those systems had any effect. And so we ended up in this very annoying place where there was no pre-policy data to compare to. So we couldn’t really measure whether these policies had any effect.

Similarly, I was working on this project trying to figure out if poison-control centers did anything. So, What effect does having access to poison-control centers do to rates of poisonings involving children? which are very common. Kids will drink bleach or other things by mistake because it’s in a little cabinet. And it turns out that when the standardization of poison-control systems happened, that also tended to be exactly when people started to collect data on how many times kids were reported as having been involved in various accidental poisonings.

And so as it relates to this paper, I would say we could only work on this project because of this incredible data that exists about reported crime and arrest rates that people have spent decades trying to put together. And there are all of these policies that seem really important that economists and social scientists can’t really dig into because we don’t have the outcome data we need to measure whether the policy had any effect. And so that’s something I think about a lot when I’m starting these projects, and I end up getting very frustrated that we don’t have, from the 1950s or ’60s, the data we would need to measure these effects that we care about.

Demsas: This is actually a very funny “good on paper” because it’s also, like, the forgotten bin of ideas that were attempted and poured heart and soul and tears into, and then at the end of the day, you can’t do anything with it.

Karger: It’s great to finally get to mention them.

Demsas: Yeah. Well, thank you so much for coming on the show.

Karger: Thank you for having me.

[Music]

Demsas: Good on Paper is produced by Jinae West. It was edited by Dave Shaw, fact-checked by Ena Alvarado, and engineered by Erica Huang. Our theme music is composed by Rob Smierciak. Claudine Ebeid is the executive producer of Atlantic audio, and Andrea Valdez is our managing editor.

And hey, if you like what you’re hearing, please leave us a rating and review on Apple Podcasts.

I’m Jerusalem Demsas, and we’ll see you next week.