Episode 32: Sarit Weisburd

 
Ep32_Weisburd_photo.jpg

Sarit Weisburd

Sarit Weisburd is an Assistant Professor of Economics at Tel Aviv University.

Date: July 21, 2020

Bonus segment on Professor Weisburd’s career path and life as a researcher.

A transcript of this episode is available here.

Episode Details:

In this episode, we discuss Prof. Weisburd's work on how police presence affects crime:

"Police Presence, Rapid Response Rates, and Crime Prevention" by Sarit Weisburd.

OTHER RESEARCH WE DISCUSS IN THIS EPISODE:

 

Transcript of this episode:

 

Jennifer [00:00:08] Hello and welcome to Probable Causation, a show about law, economics, and crime. I'm your host, Jennifer Doleac of Texas A&M University, where I'm an Economics Professor and the Director of the Justice Tech Lab.

 

Jennifer [00:00:19] My guest this week is Sarit Weisburd. Sarit is an Assistant Professor of Economics at Tel Aviv University. Sarit, welcome to the show.

 

Sarit [00:00:27] Hey, Jen, how are you?

 

Jennifer [00:00:28]  I'm great. Very happy to have you here. So today we're going to talk about your research on how police presence affects local crime rates. But before we dive into that, could you tell us about your research expertise and how you became interested in this topic?

 

Sarit [00:00:44] Sure, so my research expertise within the economics of crime, focuses primarily on policing and specifically in understanding the relationship between police allocation, police patrol, 911 calls and crime outcomes. And if I try and think about how I got interested in this, well, my father is actually a well known criminologist and he likes to joke that from the time I could talk, he would tell me about his research projects as I was always a captive audience. But the truth is that I really grew up surrounded by his research ideas and projects, and he spent a lot of time working with police officers. And so for other kids, your parent might take you to the office for a second to pick something up. For me, those stops often included law enforcement and that that was in a much more positive sense than what generally comes to mind.

 

Sarit [00:01:48] So I guess I could end the story there and say that's how I joined the economics of crime community. But the truth is that growing up, I didn't really consider academia at all. And I actually got interested in research when I started studying undergraduate economics at University of Maryland and was super fascinated by this idea of causal inference and specifically that, you know, correlations or relationships between different variables and data can be spurious. But we can use different techniques to uncover the true causal relationship. And I was really fascinated with that idea that there's kind of this mysterious relationship that we need to uncover. And I would say 15 years later, it's still the most interesting piece of my research projects. So when I started off, I really planned to do, you know, make my own way and do something totally different than the research I had grown up with at home. And I actually started looking at driving behavior and how financial incentives and police officer deterrence can impact the way people drive. But once I started looking at policing behavior, I was just fascinated. There's a wealth of data out there that really allows us to track what officers are doing in a way that you really can't do in any other type of work environment. And it just seemed like there was a lot to learn there and understand and it's such a fundamental part of our spending in our society as a whole. So that's how I circled back to this big question of policing.

 

Jennifer [00:03:38] All right. So let's talk about the purpose of policing. Obviously, the intent of police departments and police officers is to reduce crime, but there are several ways in which they might do that. So what are the various mechanisms through which we might expect that adding more police officers could affect crime?

 

Sarit [00:03:56] That's really a great question, and it's often referred to as one of the black boxes of policing and crime. So unfortunately, despite all the research, we have a bunch of ideas of the way we might think police affects crime, but we're not quite sure what's actually going on. There are two main categories, and the first is incapacitation. So if police are out on the streets arresting people and all the bad guys are in jail, then perhaps there would be less crime committed. And the other is deterrence. So deterrence goes back to Becker's model in the 60s, which is this idea that a potential criminal will only commit a crime if the expected benefit is higher than the expected cost. And we might think that when more officers are around people will perceive the cost of crime as higher and therefore won't commit the crime.

 

Sarit [00:05:01] Now, within these main categories, the black box is that we're not quite sure what police could do to create deterrence. Right. So is deterrence created by officers just walking down the street and being seen by people? Is it created by officers patrolling in cars and racing through neighborhoods? Is it a specific policing strategy like Broken Windows Policing or Neighborhoods Policing or Rapid Response Policing or Hotspots Policing? And those are, I think, kind of the big questions of our times that even if we think police impact crime, we really need to understand what they do to impact crime.

 

Jennifer [00:05:46] So what are the main challenges that researchers have to overcome in order to measure the causal effects of police on crime overall, but but also these various mechanisms that you're talking about? It seems like such a fundamental question. So why is it so difficult to answer?

 

Sarit [00:06:02] So let's say that really our goal is to measure the causal effect of police presence on crime. So how does sending another officer to your neighborhood impact crime outcomes in your neighborhood? And the first hurdle is the data itself, right? Usually we don't have data on the number of officers in your neighborhood. We have kind of this aggregate measure of the number of officers employed in a given city in a given year. And we know very little about what these officers were doing, where they were located, how many of them were even assigned to actual patrol versus behind the scenes work. And, you know, to put it more precisely, we could look at this data and see that more officers were employed in a given city in a given year. And this could mean that there were more officers patrolling all streets that year or specific streets that year or at specific times that year or more officers employed could mean actually no change in officer patrol. Right. Those officers might have been used for behind the scenes work.

 

Sarit [00:07:13] Similarly, when we talk about crime data, we also only have an aggregate measure of the number of crimes committed in a given city in a given year, and we don't really know where the incident occurred to know how it could correlate with where officers were located. And the other issue here that you always have to keep in mind when looking at research on policing and crime or crime in general, is that the data we have on crime is crime reports. Right. And we always have to be a little bit concerned about what is the difference between a crime occurring and a crime actually being reported to the police. And is that going to vary across locations or times of day or crime types? So those are kind of the data issues that we need to deal with in this field, which are huge. And I think one of the issues that despite the large amount of research in this field, you see very different outcomes because sometimes we're kind of dealing with the data that's available instead of the data that that we'd optimally like to have to answer the question.

 

Sarit [00:08:25] The other issue is, you know, estimating a causal effect. So whenever we're working with data, and especially in this era of big data and machine learning, there's this idea of "Let's just let the data talk. Let's not make a model. Let's just look at the real relationship." And that can often be very problematic because if there's a certain variable X that's very good at predicting the outcome of another variable Y, it really doesn't mean that X caused Y, right. It could mean precisely the opposite, that Y caused X or that there's some other variable that's related to the two of them. And policing and crime is a great example of this type of really complex relationship. Right. So my research focuses on the effect of police patrol on crime. And if I graph the relationship between these two variables. Right. So if I put a police presence, how many officers are in a given area at a given time on the X axis and I put crime on the Y axis, what you see is that as police presence goes up, crime goes up. If this was a causal estimate, it would suggest that as police patrol increases, crime gets worse. But the problem with this type of interpretation is that we know that crime impacts police patrol. Right. When a crime occurs, what do the police do? They rush to that area. So when there are more police, there are more crimes. So it's true that more police are predicting more crimes, but it's not because they cause the crime. It's because they're responding to a crime that already occurred.

 

Sarit [00:10:15] More than that, we know that officers try to prevent crimes by being in more crime prone areas during more crime prone times. So that means if there's something like an outdoor event or a block party, we'd expect to see both more police patrols and in any event, more crime incidents may occur. Again, this isn't because more police patrol cause more crime, but it's because this outside event resulted in both more police and more crime. And this hurdle is really a key factor in this policing and crime literature.

 

Jennifer [00:10:55] So your paper is titled "Police Presence, Rapid Response Rates, and Crime Prevention," and it's forthcoming in the Review of Economics and Statistics. So before this paper, what had we known about the effect of police on crime?

 

Sarit [00:11:08] The thing that often surprises people who aren't familiar with the economics of crime or criminology literature on policing is that generally up until the end of the 90s, the common belief—and this is both among researchers and the population—was that policing couldn't prevent crime. I mean, the role of police officers was to respond to crimes, to arrest people. But this idea that police could create deterrence was really not accepted at all. Now, why was that the case? So, first of all, as we just discussed, it's really complex in data to try and find a causal relationship between these two variables. Right. This isn't something that we're going to expect to come up. Once you make all this effort to get the right database, it's still not clear you're going to get anything causal out of the data.

 

Sarit [00:12:07] The other thing is that in the early 70s, there was this experiment called the Kansas City Policing Experiment, where police patrol was exogenously increased in specific beats. So it was thought of as kind of the optimal experiment and they failed to find any impact of increased preventative patrol on crime. So I think these factors combined together just really made people doubt that policing can have any impact on criminal behavior.

 

Sarit [00:12:42] But beginning in the late 90s, things really began to change. So Larry Sherman and David Weisburd in 1995 run this randomized trial in Minneapolis that doubled police presence in hot spots and resulted in a decrease in crime from 6 to 13 percent. So this shift in the literature wasn't saying that randomized control works, so it wasn't going against the results of the Kansas City police experiment, but it was basically saying policing can work, but we need to think about how we do policing. And Anthony Braga and coauthors follows up and there're actually a bunch of papers that replicate these types of results and hot spots. Braga's paper is looking at Jersey City and they're looking at problem oriented policing. Again, they're running it at hot spots, but they're saying, you know, if we really look at the characteristics of the neighborhood and think with the community about what needs to change, we can reduce crime. And they show the same in their paper.

 

Sarit [00:13:52] And then moving on to how this literature removed and closer to the type of analysis I run, Steve Levitt has this paper that kind of had a huge impact on the literature, even though afterwards there were some concerns in terms of the validity of the standard errors in that paper, whether these these effects were significant. But I think the main contribution of his paper in 1997 that was published in the AER was that he took this aggregate data on cities—which we had known about before but couldn't figure out how to get any causal estimates of police presence on crime because of the complex relationship that we have discussed—and he introduced this idea that we could instrument, you know, police hiring with whether or not there was a local election in that year. And that once he applies this instrument, he begins to see these deterrence effects. So now, as opposed to seeing that more police results in more crime, he's seeing that more police results in less crime. And Will Evans and Emily Owens follow up with a different instrument showing a very similar effect.

 

Sarit [00:15:08] And kind of the last group of papers that relates closely to this question that I'm asking works not with instrumental variables, but looking at natural experiments and primarily large increases in police presence surrounding terrorist attacks. So Di Tella and Schargrodsky in 2004 kind of started off this literature. And what they look at is after a terrorist attack on a main Jewish center in Buenos Aires, Argentina in July 1994, it was decided that police presence would be increased surrounding Jewish community centers or areas that could hypothetically be a focus of future attacks. And what they do in their papers, they look at what happened to burglaries and other types of crime surrounding those areas. Gould and Stecklov run a similar analysis in Israel following terrorist attacks and Draca and coauthors' "Panic on the Streets of London: Police, Crime, and the July 2005 Terror Attacks" also use a shift in police presence that's driven by a terrorist attack to try and look at what happens to crime in those surrounding areas.

 

Sarit [00:16:29] So that's kind of the the literature up to date when I'm starting to run my analysis. And the key kind of missing ingredient that I'm hoping to contribute here is to look at the specific location of police officers across an entire city and changes in police presence that aren't driven by an external event or an out of the ordinary event, hopefully like a terrorist attack. But just by the day to day mechanisms of how policing works in a large US city.

 

Jennifer [00:17:09] So in this paper, you use very cool data, which we will talk about from Dallas, Texas. Along with that police department's focus on rapid response to 911 calls as a natural experiment. So let's talk about the context in Dallas. How are police officers allocated across beats in that city? And why does that initial allocation differ from actual police presence over the course of a day?

 

Sarit [00:17:33] So beat level allocation in Dallas is first kind of determined by calls and crimes from previous years. So the data will be analyzed. And there will be this general set up of where the department thinks they need to have officers in order to respond to calls. But this can be adjusted weekly by division commanders based on current crime concerns and whatever information that they have available to them. But at the same time, DPD always aims to have a wide spread of police presence across Dallas. And this is driven by the rapid response philosophy that they follow, where it's very important to them that the police be able to respond quickly to any incident in any location within the city. So they're kind of trying to balance these two things of putting more officers where they need them more. But at the baseline, trying to have at least one officer available throughout the different areas that make up the city.

 

Sarit [00:18:41] Now, these allocation goals are really very different from what we observe in the data of actual police presence. First of all, keep in mind that police availability, just like any job availability, is dynamic. Right. So sometimes an officer will call in sick or he might take vacation. You might have an officer that would usually be allocated to a specific beat, but might need to be in the courthouse that day or involved in an arrest or in some other type of activity. And really, the most important factor that creates a difference between police allocation and police patrol is this rapid response philosophy, because dispatchers are constantly aiming to respond to a 911 call as quickly as possible. Therefore, a large fraction of these officers patrol time involves responding to calls, and these calls are usually outside of the beat they were allocated to patrol at the beginning of their shift.

 

Jennifer [00:19:46] OK, so tell us a bit about the natural experiment you're exploiting here, which you just alluded to slightly. So how do you use the response of officers to 911 calls to measure the causal effect of police presence?

 

Sarit [00:19:59] So recall that our discussion previously about a complication and measuring this relationship between police presence and crime is that police presence can be correlated with so many different things that are directly related to crime outcome. And for this paper, what I did is I tried to think about the ideal laboratory experiment. So if I could just go in and do whatever I wanted to figure out the causal effect of policing on crime, what I would do is randomly change police presence in a certain area at a certain time, having nothing to do with crime risks at that time. And that's what I'm trying to get at with the natural experiment. What I do is I'm exploiting the fact that sometimes incidents occur outside of a beat that will not be connected to crime risks within the given beat, but still require police attention. So, for example, there could be a call reporting a mental health incident outside of my beat. It shouldn't have anything to do with crime at my beat, but it could cause the officer that's currently patrolling at my beat to leave and respond to the call. This is the natural experiment. So it's this outside call that's unrelated to crime at my beat, but could be impacting the number of officers patrolling my beat.

 

Jennifer [00:21:30] So what data do you have to investigate all of this? Let's dig into those those details.

 

Sarit [00:21:37] The data used for this project was collected by the Dallas Police Department and it was shared with the police foundation as part of a project promoting the use of AVL data, Automated Vehicle Locator data, to improve policing. And this AVL data is is this really interesting data that Jen was alluding to, which I think is really important to think about and brings us into this world of big data. So the main databases I use for the analysis are the 911 call data. So this takes advantage of 684,584 calls that were reported to Dallas, Texas police in 2009. For the main analysis, I'm focusing on about 300,000 calls that report incidents of crime. And importantly, remember what was generally available is this aggregate measure of calls in a given city in a given year. The data I'm using is precise information on the type of crime reported and the precise location, latitude, longitude of each of these calls. So I know exactly what time they took place and the precise location of the call.

 

Sarit [00:22:53] The other data piece is this AVL data, the Automated Vehicle Locator data. This consists of roughly a 100 million pings of information. It tracks the location of 873 patrol vehicles that are active in 2009 by pinging the database roughly every 30 seconds with the precise latitude and longitude location of each car. So this is the big data that, you know, the pros of it were there was a lot of information and a lot to be learned from. The con is it took a while to figure out how to work with this type of data.

 

Sarit [00:23:34] And together, I use these two databases to create my database for analysis, which was analyzed at the beat level for each hour of 2009. And I could count the number of officers that were present in each of these beats, how much crime had occurred in each of these beats, and how many outside calls had occurred.

 

Jennifer [00:23:57] I think you're one of the first to actually make use of these data, right. I feel like there's one other paper maybe in the U.K., but this is the kind of data that police departments have been collecting, but as you said, it is really tough to work with. So I haven't seen anyone before actually figure out a way to make use of this very rich data that allow- basically allows you to map out where every police car is at any given time, which is really just amazing. So what outcome measures are you interested in here?

 

Sarit [00:24:28] The main outcome I'm looking at in the paper is crime, but we may think police presence will have different impacts on different types of crime. So I also separately consider public disturbances, burglaries, violent crimes, and theft.

 

Jennifer [00:24:43] All right, so let's get a bit more into the weeds of your empirical strategy. So you're going to use the number of outside calls, that is 911 calls from other nearby beats to construct two instrumental variables. So walk us through the intuition underlying these instruments.

 

Sarit [00:25:01] So recall that the goal of the instrument is to count the number of outside calls that officers patrolling this beat will answer outside of the beat in this hour. Intuitively, the most precise way to get this measure would be to simply count the number of officers who left the beat in the hour to respond to an outside call. Right. Like I could really use this detailed data I have on what police are doing throughout the day to get a precise count of exactly the measure we're talking about. However, this raises the question, well, why were these specific officers assigned to an outside call? Perhaps that's because there was a low risk of crime in their current beat. If this is the case, it would invalidate the exclusionary restriction or the assumption that responding to an outside call is not connected to the crime risks in your beat. So I couldn't use this really precise measure.

 

Sarit [00:26:05] What I do instead is I introduced these two instruments, as you said, that estimate the probability that officers from this beat will be assigned to an outside call in this hour, focusing only on factors that are unconnected with crime risks at this beat. I refer to the first as the outside calls ratio. The idea is that officers in this beat will be more likely to abandon the beat when more outside calls occur in that hour and when there are less officers available in the area surrounding the location of the outside call. The way I calculate this is by counting the number of relevant outside calls and dividing that by the number of officers patrolling the area surrounding the location of the outside calls. The other instrument I use is just outside calls. This instrument focuses only on the number of outside calls that occur in that hour, ignoring the availability of surrounding officers that could respond to these calls. While this instrument has a weaker first stage, it alleviates exclusion restriction concerns that police availability at the outside beats could be correlated with crime risks that this beat.

 

Jennifer [00:27:29] A valid instrument must be correlated with the treatment variable that's the first stage you just mentioned. So in this case, actual police presence in the beat of interest, that's a straightforward correlation that you can test. But an IV must also satisfy what we call the exclusion restriction, as you mentioned. So that is it can only be correlated with the outcome measure—local crime rates—through its effect on the treatment variable—actual police presence. So talk us through this a bit more. What might we be worried about in this case? And how do you convince yourself that your instruments satisfy this exclusion restriction?

 

Sarit [00:28:05] This is a good question, and it's already come up a bit in our discussion so far. So let me just walk you through the steps that I tried to do in order to make the exclusion restriction more convincing. First of all, I'm not using a count of the number of outside calls the officers were actually assigned to. This is what I just discussed before, because we're going to be concerned this could be directly related to crime risks at the beat. Second, I'm not using crime risks outside of the beat. So hypothetically, I could have defined outside calls as any call, whatever is being reported. But I specifically don't include crimes that are being reported outside of the beat. The reason for this is that you could imagine a string of burglaries occurring across Dallas so that an outside burglary could be directly related with the probability of a burglary at this beat. I therefore focus specifically only on calls that are unrelated to crime. So these are calls reporting incidents related to mental health, child abandonment, fire, animal attacks, dead people, suicides, abandoned property, fireworks, and drug houses.

 

Sarit [00:29:28] The third thing I do is—because one might be concerned about a general crime increase in the city could increase police presence at that- both this beat and outside beats—I use this second instrument. Because I really could have stopped my analysis with just the outside calls ratio. But the reason I use outside calls is because the exclusion restriction could be considered stronger for these calls, as I'm not taking police presence, even in outside beats, into consideration. And the last thing I do for both of these instruments is I run a falsification test where I show that in types of calls that we wouldn't expect a response to police presence. We can use exactly the same analysis and we find no effective police presence on these types of calls. So I don't find that police presence affects suicides or abandoned child reports or fire reports or drug house reports. Or we might think that all of these types of incidents are less likely to be affected by whether there's an officer present this hour or not.

 

Jennifer [00:30:40] OK, so intuitively, you're going to use your functions of outside calls—those instruments you just described that are unrelated to crime—to predict actual police presence in a particular beat. So this isolates what we would consider the good or possibly random variation in police presence so that you can use that as your treatment variable. You'll then estimate the effect of predicted police presence on local crime rates. So getting to the results, what do you find is the effect of increasing police presence in a particular beat on crime rates in that beat?

 

Sarit [00:31:14] I do find a significant deterrence effect specifically that if you increase police presence in that beat by 10 percent, you will see a 7 percent decrease in crime. So a 10 percent increase in police presence results in a 7 percent decrease in crime.

 

Jennifer [00:31:31] Do those effects vary across different types of crime?

 

Sarit [00:31:35] Importantly, while if I run the OLS analysis—so if I don't instrument for police presence, if I just kind of look at the broad correlations in the data—across all types of crimes, you see what looks like more police results in more crime. But when I apply the instrument, I find the strongest effect on violent crime and slightly smaller effects for public disturbances and burglaries. And the one type of crime that I don't find a significant effect is theft, although it does get rid of this significant positive effect that you see in terms of correlations.

 

Jennifer [00:32:16] So how did your effect sizes compare with those from previous studies on this topic?

 

Sarit [00:32:21] My effects are large. So generally, if you look at the randomized experiments, they tend to report the smallest effects. They look at very large changes in police presence. For example, the Sherman and Weisburd paper that I quoted at the beginning, they double police presence and find a decrease in crime of between 6 and 13 percent. Generally, the difference-in-differences papers—so these papers surrounding terrorism—tend to find that a 10 percent increase in police presence results in a 3 to 4 percent decrease in crime. The IV estimates tend to be larger and closer to my estimates, but generally with much larger confidence intervals—they tend to be less precisely measured than the analysis that I'm running.

 

Jennifer [00:33:19] Do you have a sense of why it makes sense for your estimates to be bigger than those other estimates? Do you have some intuition there?

 

Sarit [00:33:26] Yeah, so I I generally think what is going on here that could be different than what's been going on in these other papers is what we're looking at in this paper, is what happens when you take an officer away. I think taking an officer away from an activity that he or she was doing that they perceived as important is very different than assigning an officer to do something or to go to a certain location. So you actually had an officer that thought they were doing something important, and then in the middle of whatever they're doing, you know, they're rushed to a different area. And I think that that that's important and different than what we were looking at in these other papers. Right. So if we're looking at a paper on general police hiring and we're looking at a change in the number of officers that were hired, we don't quite know what those officers were doing. Right. And how they were being used and whether it was effective. When you're looking at a paper like mine where you have a patrol officer who's assigned to an area that he or she is assigned to every day, that has a general idea of what needs to be done and where they should be, and then suddenly they're not there anymore, that raises some concerns about what did we stop from happening when we rush them off to respond to an outside call?

 

Jennifer [00:34:54] That's really interesting. OK, so then you dig into the data that you have a bit more to try to understand more about the mechanisms driving your effects. So what are you able to look at and what do you find here?

 

Sarit [00:35:06] So the first thing I do is try and understand, similar to what we just discussed, what were these officers doing? So when we take an officer out of their beat to respond to an outside call, what's being left behind? Because hypothetically, you might say, well, you know, when they're not responding to a call, they're just sitting in their car in an empty parking lot and filling out some paperwork. So what does it really matter whether they're in their car or outside of the beat responding to something? And to do this, I look at arrests. So instead of my outcome variable being crime, I say, how does police presence impact arrests in your beat? And again, I instrument for police presence with these outside calls.

 

Sarit [00:35:55] And what I find suggests that when officers are in your beat, they're conducting arrests. They're actually active on the streets and people see them. And therefore, when they're being sent elsewhere, the community sees this and it could impact their choices. I have to say a caveat here, that this also raises the question we discussed before of what are police doing? Is this, you know, arresting people or creating deterrence? And and so part of the reason why it might be important that officers are present in your beat is if they're arresting people and that's also preventing crime, that's an important activity that they can be involved in.

 

Sarit [00:36:51] The other thing I look at is how police presence affects crime and whether this matters, how big the beat is that they're patrolling. And perhaps not surprisingly, what I find is the impact of each additional officer is larger in smaller beats. But what's interesting is that at the margin, the effect is the same. So if you're going to increase police presence by 10 percent at each of the beats, it's going to have pretty much the same effect across beat sizes. Why is that? Because usually the larger beats tend to have more officers. So a 10 percent increase is going to mean more officers at these larger beats.

 

Sarit [00:37:34] The last thing I try and look at is this question of deterrence versus displacement. So one of the things you may be asking yourself when you hear about this paper is when I say that an increase in police presence this hour reduces crime this hour in this beat, you may ask, well, what does that mean? Did you actually prevent a crime from occurring or did you just push this crime into a neighboring beat where there wasn't an officer present? And that would be called displacement. So to look closer at this issue, I aggregate my data up from the beat level to the sector level so each sector is roughly five or six beats. And I run the same analysis where we might expect a smaller effective police presence on crime in these larger sectors because a person could just move from committing the crime in this beat to committing the crime in another beat. And I actually continue to find a significant deterrence effect that's pretty similar in size. The only crime type that appears to perhaps show some evidence of displacement are public disturbances.

 

Jennifer [00:38:52] OK, so that is your paper, and as I said before, it's now forthcoming at REStat, but it's been around a little while. Are there any other studies that have come out more recently that add to our understanding of how police affect crime?

 

Sarit [00:39:05] So there's a working paper out by Jordi Blanes i Vidal and Giovanni Mastrobuoni, "Police Patrols and Crime." So, clearly very related to my research. And interestingly, they're also using AVL data, so very similar data to what I'm using. They're exploiting a natural experiment in Essex, in the U.K. And I find both the experiment really interesting and the results really interesting. So they are comparing treatment areas and control areas, where treatment areas are areas where a burglary has occurred, and in those areas the location receives an extra 10 minutes of police presence per day, which is a 33 percent increase in police presence, for one week. And they're able to look at whether this increase in police presence after the incident of the burglary impacts crime outcomes. And they find no significant effect of this change. Right. So after all these papers showing evidence of deterrence, their paper using great data and a really clean experiment shows no effect.

 

Sarit [00:40:21] And I think it goes back to, again, thinking about police patrol. So really, the difference between their paper and my paper is my identification is coming from what happens when you take officers away from an area where they wanted to be and their paper's looking at what happens when you bring officers into a new area. And I think it raises a lot of important questions about, well, what type of police patrol works and what type doesn't.

 

Jennifer [00:40:56] Yeah, and in particular, it seems- it highlights the the real potential value in in really knowing the community that you're working in. Right. So if someone's been patrolling an area for a while, they might have a relationship with the people who live there. They might have a general sense of who the kids are who are trouble and who who isn't. And- but if you're just being added because of local crime that's being committed, you won't know any of that stuff. Is that the general hypothesis that you have in mind for that?

 

Sarit [00:41:26] Exactly. I could not have put that together myself.

 

Jennifer [00:41:32] Perfect, awesome. So putting it all together, your findings in this paper, along with the other work in this area, what are the policy implications? What should policymakers and practitioners take away from all of this?

 

Sarit [00:41:42] In my view, we all need to think carefully about rapid response policing. You know, what are the real benefits of the policing strategy and what are the costs? I think there's a huge emphasis these days on how quickly officers respond to a large range of incidents. And I think the public is often unaware of the costs of this policy. To be clear, I think rapid response is very important for certain types of calls. And I think we need more research to understand which calls those are and think more about how to balance rapid response and deterrence focused policing in an optimal manner.

 

Jennifer [00:42:27] It also reminds me of- there's a paper by Justin McCrary and Aaron Chalfin that argues that most US cities are under policed. So if you look at what happens when you hire an additional police officer, on average, it reduces crime in a very cost effective way. And so the finding from that paper is that what we really should be doing is adding more police across the board in all these cities because we're very far from an efficient margin. And this, of course, abstracts from all of the questions that we've been talking about for the past 45 minutes or so about what exactly these police are doing. And people worry about overpolicing certain neighborhoods and all that. But on average, it looks like most US cities are under policed. So another takeaway for me from your paper was just maybe all of these needs to have more officers in them so that when a 911 call comes in, if we want police to get to the scene quickly, we don't have to pull officers from other beats. Is that is that another important implication here?

 

Sarit [00:43:24] I think you've raiseed- I think it's a great question. And I think we need to think about it and think about how to answer it. I think it raises other questions. Right. So there's other analysis in this field that looks at other countries like the UK, where generally there's a separation between police patrol and responding to calls. So you don't have the same officers that respond. Right. So that's in some way taking your idea to the extreme. And I ask myself if that's better. I think and this is not based on you know, I think you really need much more research to figure out if what I'm saying is true or not. But just my intuition is that in some ways, combining patrol with rapid response is optimal. Right. Because you're kind of using everyone all the time because you can imagine a day where you have very few calls and then you have all these officers that are just sitting in the department waiting for the call to come in who could actually be creating deterrence on the street. So intuitively, I think that the way Dallas does things—and most departments in the US—runs their department, where they combine the two is optimal. And thinking about how to allocate officers is really important here.

 

Sarit [00:44:52] So, yes, I would generally think if we want to reduce crime based on research, you should clearly hire more officers. But what to do with these officers, whether the right thing to do is just to put them in every beat so that you could both respond to calls and create deterrence or whether you should be putting them in specific hotspots location? That's still a big question that we really don't know the answer to. That's that's part of this black box that you know, yeah, so it looks like hiring police officers is a good idea, but what should we do with these extra police officers? I think the jury's still open on that.

 

Jennifer [00:45:33] So that alone is a big piece of the research frontier. What are the other big questions in this area that you and others will be thinking about going forward?

 

Sarit [00:45:43] Well, you touched on this before, but for me, the biggest concern regarding the friction between rapid response and deterrence policing is thinking about how this impacts neighborhood policing. Right. So if officers are often responding to calls in an outside beat in an area that they're unfamiliar with and the local population doesn't know them or trust them, how does this impact the outcome of the incident or the way the community feels about the police? And and for me, that's kind of the big question I'm interested in right now, which is, you know, what are the dynamics that go on in this constant friction between, you know, deterrence policing and rapid response policing and community policing? How to all of these factors play against each other and what is the outcome?

 

Jennifer [00:46:39] My guest today has been Sarit Weisburd from Tel Aviv University. Sarit, thanks so much for doing this.

 

Sarit [00:46:44] You're welcome. Thanks for having me.

 

Jennifer [00:46:48] We recorded this interview in early May before the death of George Floyd and the ensuing protests and policy conversations about police reform. A lot has happened between then and now. I wanted to give Sarit a chance to comment on how her research fits into this broader conversation about the role of police in society. Here's Sarit.

 

Sarit [00:47:05] While all my interactions with police have been positive. This doesn't discount the fact that, sadly for many people, especially Blacks in our community, this isn't the case. The current project deals with the average effect of police presence on crime, which I think is an important one. But it's only a first step in understanding the complex dynamics of what police are doing to reduce crime and the costs and benefits of these policies. I hope that my future research will be able to play a role in understanding this dynamic and improving both police effectiveness and their relationship with all of the communities they are charged with protecting.

 

Jennifer [00:47:48] You can find links to all the research we discussed today on our website, probablecausation.com. You can also subscribe to the show there or wherever you get your podcasts to make sure you don't miss a single episode. Big thanks to Emergent Ventures for supporting the show and thanks also to our Patreon subscribers. This show is listener supported, so if you enjoy the podcast, then please consider contributing via Patreon. You can find a link on our website. Our sound engineer is John Keur with production assistance from Elizabeth Pancotti. Our music is by Werner and our logo was designed by Carrie Throckmorton. Thanks for listening, and I'll talk to you in two weeks.

Jennifer DoleacInstrumental-Variables, Policing