Episode 51: Amanda Agan & Anna Harvey

 
Untitled design (2).jpg

Amanda Agan & Anna Harvey

Amanda Agan is an Assistant Professor of Economics at Rutgers University.

Anna Harvey is a Professor of Politics and Director of the Public Safety Lab at New York University.

Date: June 8, 2021

A transcript of this episode is available here.

Episode Details:

In this episode, we discuss Prof.s Agan, Doleac, and Harvey's work on the effects of misdemeanor prosecution:

"Misdemeanor Prosecution" by Amanda Agan, Jennifer Doleac, and Anna Harvey.

 

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 [00:00:18] I have two guests this week. Amanda Agan is an Assistant Professor of Economics at Rutgers University. Hello, Amanda.

 

Amanda [00:00:25] Hi there.

 

Jennifer [00:00:26] And Anna Harvey is a Professor of Politics and Director of the Public Safety Lab at New York University. Hello, Anna.

 

Anna [00:00:32] Hey, guys.

 

Jennifer [00:00:33] So we have two guests because the three of us recently put out a new paper on how prosecuting defendants for low-level offenses affect those defendants' future criminal justice involvement. And we thought it would be fun to get all three of us together to talk through that paper for the show. So both Amanda and Anna have been on Probable Causation before. I'll post links to their earlier episodes in the show notes. But for those who missed those conversations, Amanda, could you tell us about your research expertise and how you became interested in this topic?

 

Amanda [00:01:01] So most of my research focuses on criminal legal system and estimating impacts of various policies either on recidivism or employment for people with records. And after the Ban the Box research, which we spoke about previously, I have been looking more into the impact of criminal records themselves. And this question that we're studying here fits squarely into that realm. How does one come to have a criminal charge in the first place? And combined with the current wave of policies to decline prosecution for various misdemeanors that are starting to be implemented across the country, you know, when Anna got in touch about embarking on research on this topic, I was immediately game.

 

Jennifer [00:01:40] And your turn, Anna, what's your research expertise and how did you become interested in this topic?

 

Anna [00:01:45] Well, so I'm a political scientist by training, and I started working on criminal justice questions about four years ago. But this question I think I really got interested in because of the policy conversation. So I started talking to district attorneys' offices in 2018. Larry Krasner had just been elected in Philadelphia and Kim Fox had just been elected a couple of years earlier in Chicago. And there was starting to be this conversation about the volume of misdemeanor prosecutions that we pursue in this country. And these prosecutors were offering alternatives to misdemeanor prosecution. And I thought, well, gee, that's a really important policy question. What do we know about that? And I couldn't find anything in the literature about what we do about that. And so I started trying to find a district attorney's office that would be willing to share data with us.

 

Jennifer [00:02:33] And I am, of course, also an economist. And my own research focuses heavily on recidivism and what we can do to reduce the likelihood that people cycle back through the criminal justice system. So this question of how to reduce recidivism for people accused of misdemeanor offenses was of great interest to me for that reason. So our paper is titled "Misdemeanor Prosecution." There is a lot of research out there on felonies, murder, rape, robbery, assault, big crimes like that. But as Anna just said, we know much less about these lower level offenses, things like trespassing, shoplifting, and minor drug possession. Amanda, why should we care about misdemeanors?

 

Amanda [00:03:08] So misdemeanor charges make up a vast majority of charges in courts across the United States. More than three quarters of charges are made up of these low-level misdemeanor crimes, representing nearly 13 million cases each year. And so a vast majority of people with criminal records are going to have misdemeanor charges or convictions in their criminal history. And as you mentioned, these misdemeanor crimes are lower level crimes than felonies and can encompass anything from low value theft and minor violence to selling cigarettes without tax stamps, operating a vehicle with an expired license, or driving with an object hanging from your rearview mirror.

 

Jennifer [00:03:48] And so in this paper, we focus on the effect of not prosecuting someone in Suffolk County, Massachusetts, where Boston is located. I'll emphasize again, our treatment here is going to be non-prosecution because the default in most of the country is to prosecute most of these cases. So, Anna, two questions for you. Why Suffolk County? And how do we define non-prosecution?

 

Anna [00:04:08] Well, you know, in 2018, when I first started thinking about this would be a really interesting question to investigate, I went to my local district attorney, the Manhattan District Attorney's Office. Cy Vance was elected in 2010. And, you know, people don't really think about Cy Vance as being one of the reformer district attorneys, but he did quietly, more under the radar, start rolling back prosecutions for minor offenses. And, you know, I talked to the office and they were really intrigued by this question of what effects did that policy change have. But they were also concerned. I mean, they were worried. Well, what if the results of the study came back and showed that, in fact, those policies were endangering public safety? So I was connected to Rachel Rollins. Rachel was elected in 2018. She took office in January 2019. We're connected by a nonprofit justice collaborative. And I went up to Boston a couple of weeks after Rachel's inauguration, expecting to get the same answer that I'd gotten in Manhattan, especially after I said, you know, Rachel, I can't guarantee you that the results of the study will show that you're- you know, non-prosecution is the right strategy. But she had a completely different response. She said, I need to know the truth. I need to know if the policies that we're implementing are hurting or helping public safety. And if they're hurting, public safety will change. She signed a data use agreement that day. So so that's why we're here in Suffolk County-.

 

Jennifer [00:05:32] Amazing.

 

Anna [00:05:32]  Right? Because Rachel agreed to work with us. So non-prosecution, the way the prosecution of misdemeanors works in Suffolk County, which includes Boston, is that when new criminal complaints are issued by a law enforcement agency, they come to the court with jurisdiction for the location where the offense allegedly occurred. And a prosecutor, an assistant district attorney with the Suffolk County District Attorney's Office, sees all those criminal complaints for the first time on the day of arraignment. And an ADA can either dismiss a case, all charges prior to arraignment and the defendant is free to go or the ADA can proceed with arraignment, which means the defendant will enter a plea before a judge. So we define non-prosecution to include any case that is closed on the day of arraignment without a conviction for a defendant, that does not proceed any further. And we define prosecution to include any case that proceeds past the day of arraignment, irrespective of whether that case eventually ends up with a conviction or not.

 

Jennifer [00:06:36] So we're going to be considering the effect of that prosecution decision for those on the margin, that is, those for whom different prosecutors might disagree about whether to move forward with a case. So this leads to the question of why that decision might matter. And we see several possible channels that we talk about in the paper. So the first is that prosecuting someone and potentially convicting and punishing them can have what we call a "specific deterrent effect." That is the experience of the prosecution and punishment might make the defendant want to avoid going through it again, and this could deter them from committing crime in the future. It's also possible that prosecution could provide an opportunity for the court to intervene and mandate services or programs that could be helpful, but that the defendant might not otherwise engage with on their own—things like drug treatment, for instance. So those are reasons that prosecuting marginal cases could have beneficial effects and reduce future criminal justice contact.

 

Jennifer [00:07:27] On the other hand, we might worry that prosecuting someone for a low-level offense leaves a mark of a criminal record that is visible to other law enforcement agencies and to potential employers. We know that employers discriminate against people with felony convictions and there's some evidence that they don't care as much about misdemeanor convictions. So it's not clear how much this mark matters, but it could matter. And it's not just convictions that show up on someone's record. So even if the case is eventually dismissed, the record of arrests and prosecution is visible. And that could also push employers and landlords and other courts to treat this person differently. And finally, dealing with a criminal case is stressful and can be disruptive. You have to take full days off of work to go to court hearings, which could mean lost income and maybe even a lost job. And of course, if you're convicted and punished for the offense, that could mean probation or fines that create additional hardship. So those are the reasons we might expect the prosecution decision to be harmful.

 

Jennifer [00:08:17] And of course, it might not matter at all. Maybe nobody cares about a misdemeanor arrest on your record and all the cases on the margin of being prosecuted are simply dismissed in a week if they're not dismissed today. So figuring out whether and how much all this stuff matters on net is the empirical question that we tackle in this paper. So, Amanda, before this paper, what did we know about the effects of misdemeanor prosecution?

 

Amanda [00:08:38] About the causal effects of misdemeanor prosecution for defendant outcomes? Very little, as we have mentioned before. There have been some great recent books and papers bringing attention to the size, scope, and qualitative consequences of misdemeanor prosecution, including a great book by Sasha Natapoff and a recent Law Review article by Megan Stevenson and Sandy Mayson that documented statistics about misdemeanor cases in major cities across the United States and was the source of our statistic that nearly three quarters of these court cases are misdemeanors. In terms of estimating causal effects of alternatives to prosecution, there was a paper by Mike Mueller-Smith and Kevin Schnepel that studied deferred adjudication for felony charges in Texas. And now deferred adjudication is not exactly like non-prosecution. It's a plea that an individual defendant takes. They then get a probationary period. And if they successfully complete that probationary period, then the case is then dismissed. And they found using variation from two policy changes, that defendants who received deferred adjudication were less likely to recidivate and more likely to be employed than those who were prosecuted in the more normal way. But for misdemeanors, which again make up a large majority of criminal cases in the United States, we had little evidence about what the impact of prosecution or non-prosecution on defendant outcomes like recidivism was going to be.

 

Jennifer [00:10:04] And so why don't we know more than we do? Right. So this is a tough question to answer, mainly because prosecution isn't random. Prosecutors choose which cases to move forward with based partly on information that we can see in the data, things like the current charges and the defendant's criminal history, but probably also a bunch of stuff that we can't see—their demeanor in court, the histories of substance use or mental health issues. So if we simply compared people who are prosecuted with those who are not, even if we controlled for everything we have, this would not tell us the causal effect of prosecution, because we'd be worried that those people are different in some way that we can't account for. Those differences might be what caused the different prosecution decisions initially and could also explain any differences in future criminal behavior down the road. So that's the identification challenge in this setting, and it's a big one.

 

Jennifer [00:10:50] The other challenge is data access. So as Anna's already alluded to, getting data from prosecutors' offices in general has historically been really difficult, and getting data on misdemeanors has been especially hard. The dataset we got from Suffolk County is really unique and we would not have been able to do the study without it. So, Anna, what information do we have in this amazing dataset and what outcome measures are we going to focus on here?

 

Anna [00:11:13] Well, so we're using the internal data from the Suffolk County District Attorney's Office, which is really, as you said Jen, it's really important because the office keeps a record of every criminal complaint in the county that comes to them, irrespective of what the final outcome is and irrespective of when that final outcome happened. So for these nonviolent misdemeanor charges that are our focus of interest in this paper, if that complaint comes into the Suffolk County District Attorney's Office and, you know, even prior to arraignment, an assistant district attorney decides not to pursue the case and the case is closed, the office still has a record of that. And that's really important because in Massachusetts, if your complaint is dismissed prior to arraignment, there's no record of that case in the statewide criminal justice information system. So we wouldn't be able to see any of these cases that are dismissed prior to arraignment if we were working with, say, the statewide official criminal records. So we're using internal data from the Suffolk County District Attorney's Office and we are following these defendants over time. And we're seeing whether after the initial arraignment on which we focus, whether they come back within windows of one, two or three years. We're looking at whether they have new criminal complaints within those periods of time. We're looking at the number of new criminal complaints they have. We look at whether those new complaints are misdemeanors or felonies. We look at the kind of charges in those new complaints, whether they're violent crimes or property or motor vehicle violations. And then finally, we look at whether those new complaints are themselves prosecuted by the District Attorney's Office and whether those prosecutions result in new criminal records for the defendants.

 

Amanda [00:13:05] Can I add one thing to that that I think is really great about our data. In addition to having the information about the complaints that don't end up being prosecuted, is we also have that information on which ADAs handled kind of the initial part of the case versus the later part of the case, which, as you're going to see soon, becomes very important for our identification strategy. And I know that when I've worked with court data in other jurisdictions, if we were working with court data instead, that information is also often given to you as the last prosecutor who worked on the case. Right. And so if you have a case that eventually goes on to be convicted, you're going to know the- you might know the name of the prosecutor who ended up convicting the case, but you won't necessarily know the name of the prosecutor who made that initial decision to move forward with the case in the first place, which is another great thing about having this internal data from the district attorney's office.

 

Jennifer [00:13:58] Great. I will also clarify, I don't think we've defined what a complaint is yet. People should think of it as just an arrest. It's slightly more complicated than that, but it is basically all arrests and summonses. So, Amanda, tell us about the defendants in our sample. What offenses are they charged with? Do they typically have a criminal record? What do they look like in general?

 

Amanda [00:14:16] So the defendants in our sample that we're focusing on, were all charged with nonviolent misdemeanor offenses. And the largest category of crimes within this were motor vehicle related crimes, including license and insurance violations and charges for driving under the influence. About half the defendants in our sample had had a previous complaint in Suffolk County before, and more than half had had a previous record in the state of Massachusetts' DCJIS record system, which we had mentioned a little bit before. For the defendants for whom we could measure demographic characteristics, the average age is around 34 and about a quarter are less than 23. And nearly 46 percent of the defendants within the administrative data that we have were Black; whereas, in Suffolk County itself, only approximately 24 percent of the residents are Black.

 

Jennifer [00:15:06] OK. So it turns out that in Suffolk County, cases are assigned to arraigning ADAs, or assistant district attorneys, in a way that is essentially random. I'll walk through that in a moment. But first, let's imagine the ideal experiment we'd like to run here. We'd randomly assign any new case in Suffolk County to be prosecuted or not. That would, of course, be an extremely unethical experiment, but it would allow us to measure the causal effect of prosecution because we could be more confident that there aren't some unobservable factors that drove the decision to prosecute a case or not. We'd simply flipped a coin. So how can we approximate that experiment in the real world?

 

Jennifer [00:15:42] So it turns out that when people have discretion, as prosecutors do, they often make very different decisions than someone else would have made. In this context, prosecutors vary quite a bit in the likelihood that they will prosecute the case in front of them. This means that instead of randomly assigning cases to be prosecuted or not, we could randomly assign them to different prosecutors and get effectively the same result. Some people are lucky and get a lenient prosecutor and so are not prosecuted, while other people are unlucky and get a harsh prosecutor and they are prosecuted as a result. Then, when we compare two people who look the same in every way, but one was prosecuted and the other wasn't, we can plausibly argue that this difference is because of their different prosecutors and not because of some underlying difference between the cases that we can't see.

 

Jennifer [00:16:24] Now in Suffolk County, cases aren't exactly randomized to prosecutors, but the process has the same result. Cases are assigned to courtrooms on a given day, typically the next week day after the arrest. And the ADA who happens to be assigned to that arraignment room on that day, handles all those cases. They have a huge stack of paper files to work through, and they quickly decide whether each case is worth the court's time to move forward with or not. And sometimes the ADA assigned to that room has another meeting and needs to step out for a bit and someone else takes their place. And in general, there are just so many of these misdemeanor cases and everything in the arraignment hearing moves so fast and ADAs schedules are so unpredictable that it's virtually impossible to game this system to try to get a more lenient ADA to make the prosecution decision. You get what you get. And so after controlling for things like the court and the month of the sample and the day of the week, it's essentially random whether a particular case is handled by a lenient ADA or a harsh one. And that gives us the natural experiment that allows us to measure the causal effect of prosecution. So, Amanda, walk folks through how we actually do this in practice. How do we calculate a prosecutor's leniency and what's the effect of that leniency measure on the likelihood of prosecution?

 

Amanda [00:17:30] In practice, we basically estimate the proportion of cases each ADA does not prosecute as a measure of their leniency. Now, it's a bit more technical than that. There are some statistical problems you run into if you use exactly that. And as Jen just mentioned, we do need to take into account that there are different courthouses within Suffolk County and different time periods within which these ADAs work. But it is basically close to that. And remember that because the cases are essentially randomly assigned, as Jen just described, with enough cases, the characteristics of these cases are going to look very similar across all of these ADAs. And this measure of leniency, this proportion of cases that an ADA does not prosecute varies considerably across prosecutors. At the low end, we see prosecutors who only declined to prosecute something like 4 percent of their cases. And at the high end, we see prosecutors who are declining to prosecute nearly 48 percent of their cases. And this measure of leniency of the ADA who makes the decision on your case is highly predictive of whether your case is going to be declined. A defendant assigned to an ADA who is 10 percentage points more lenient is 5.5 percentage points more likely to not be prosecuted. And that represents a 25 percent increase over the mean non-prosecution rate we see in this particular sample, which is about 21 percent.

 

Jennifer [00:19:04] And so then we're going to use that leniency measure basically as an instrumental variable for whether you're prosecuted—or whether you're not prosecuted, I guess I should say—to measure the causal effects on future criminal justice contact. And I'll also mention that we we check whether information about the defendant and the case predict the leniency of their prosecutor. And it doesn't. It looks totally random, which is what we'd expect based on our conversations with folks in the DA's office and the process as we understood it, as I described before. But it's, of course, always nice to see it confirmed in the data. So, Amanda, tell us the headline results from our IV strategy. What is the effect of non-prosecution on the likelihood and number of future criminal complaints?

 

Amanda [00:19:42] So for the marginal defendants in our sample, having their case not prosecuted significantly reduces their probability of a criminal complaint within two years. And the effect is large. Relative to similar prosecuted defendants, our estimates imply that non-prosecution reduces this probability by 58 percent. Or, put the other direction, prosecuting a marginal defendant significantly increases their probability of ending up with another criminal complaint within two years.

 

Jennifer [00:20:14] So one hypothesis for why this matters so much is that non-prosecution can avoid pulling people into the criminal justice system and down a different path than the one they were on. So if that's the case, we would expect to see bigger effects for first time offenders than for people who had already been prosecuted or convicted before. And indeed, this is what we find. When we look at results separately by first time and repeat offenders, we find that our results are much larger for those first time defendants. We can also look at the types of new offenses that are avoided since our non-prosecuted defendants commit less crime in the future. And we can see reductions in future violent offenses, future motor vehicle offenses, and future theft and fraud offenses. So a big range of stuff. Anna, let's talk more about mechanisms. What are the effects on the outcomes in the initial case that might explain the big effects we're seeing on future criminal justice contact down the road?

 

Anna [00:21:03] Well, as we talked about earlier, in Massachusetts, there's no criminal record of your charges if those charges are dismissed prior to arraignment. And we, in fact, find that cases that are not prosecuted by our definition are 56 percent less likely to have criminal records charges in the statewide criminal records system. So it's like the charges never even happened if they don't show up in the statewide criminal record system. And given what we know, including from your and Amanda's work, given what we know about the harmful effects of criminal records, this is probably pretty important. It's harder to get a job if you have a criminal record. And we think prosecutors probably treat you differently if you have a prior criminal record. Having a criminal record of charges is kind of like having a lifetime punishment simply for having been charged in the first place, even if those charges were later dismissed. We also find that 26 percent of the prosecuted cases—26 percent of the sample actually—winds up with misdemeanor convictions. And so then you have not just a record of charges, but now you have a criminal record of a conviction, which is probably also harmful. And then finally, cases that are prosecuted remain open for about six months. So it takes about six months to get to a disposition in one of these nonviolent misdemeanor cases. Again, even if that disposition is not a conviction, which it is in most cases. So during the six months, you have to continue to come to court, to get there on time, to not forget a hearing, you've got to miss work if you need to. And so all of those things too may make- kind of increase the bar for defendants in terms of trying to get into key legal employment.

 

Jennifer [00:22:43] I think that result in particular was super surprising to me, because I think going in, you know, you think these marginal cases, you happen to get assigned to this prosecutor rather than that one, you know, how much of a difference does it really make? It's not dismissed today. It'll just get probably dismissed in a couple of weeks because these are all just you know, you could go either way with these cases and they're not that serious. But then, yeah, you're stuck for six months dealing with this and they're- a quarter of them wind up with a conviction. So this initial prosecution decision actually is a big deal. One somewhat surprising result was that our IV estimates are much bigger than what we would have gotten with just a naive OLS regression, where we were really worried about selection bias. So, Amanda, one, why is this surprising? Say a bit more about that. And two, what seems to be driving this result?

 

Amanda [00:23:29] Sure. So as we were describing before, part of the reason that we wanted to use this kind of instrumental variables design was exactly because prosecutors choose which cases to prosecute and non-prosecute. Right. And they're going to make that decision based on some observable and observable characteristics that might be correlated with eventual recidivism. And when we're first thinking about this problem, we might assume or expect that prosecutors are choosing, quote unquote, "better defendants" to not prosecute—defendants that might be less likely to recidivate in the future. We would think of this as sort of positive selection, choosing those better defendants. And if that were the case, then we would actually find that once we purged our estimates of this sort of positive selection, that our instrumental variables estimates would be smaller in magnitude than those OLS estimates. But actually, as you say, our IV estimates are larger in magnitude than our OLS estimates, which would imply negative selection that the prosecutors are choosing on average, quote unquote, "worse defendants" at least when it comes to recidivism risk.

 

Amanda [00:24:39] Now, one characteristic that we can actually see that rationalizes this negative selection is age. We can see in our data that prosecutors are less likely to prosecute younger defendants, perhaps because they think of them as less culpable or more deserving of a second chance. And generally, as well as in our data, younger defendants are more likely to go on to commit further crimes. And so age has this sort of negative selection property that would rationalize the fact that our IV estimates are larger in magnitude than our OLS estimates. And so there's clearly a role for this sort of negative selection here. And it seems plausible that there are other unobservable, at least to us as analysts, characteristics that function in a similar way.

 

Jennifer [00:25:26] So we do, of course, a whole bunch of robustness checks to convince ourselves and our readers that the effects we're measuring here are the causal effects of misdemeanor prosecution. So, Anna, what is your favorite robustness check and why?

 

Anna [00:25:38] It's a funny thing to have a favorite over. Well, you know, the data that we have is great. And we're so grateful to District Attorney Rollins for having shared the data with us. But it's not perfect, and—you know, as with any set of administrative data—and one of the things that we were a little bit concerned about was the missingness of some of the information in the administrative data. You know, Amanda talked about how important it was that we could get information about that initial arraigning ADA assigned to a case. We don't have that information for every case in our sample. In fact, we really only have that information for about a third of the sample of all the cases that we could possibly use. So that's where we're leveraging our main estimates. But we do see- we know that ADAs were assigned to arraignment courtrooms by week. And we can see kind of the partial- it's like we can see the almost like the ghost schedule based on the information about ADAs assignment that we do see. And so what we do is we fill in that schedule based on the observed sets of ADAs that are working in different courtrooms on different days, during different weeks. And we have a set of progressively more lenient imputation schemes where we fill in more and more of the daily and weekly schedules. And what's interesting is our estimates are really remarkably stable as we increase the sample. We get up to about 76 percent of all cases through these imputation schemes. And the estimates stay really constant across the sample set. That that was reassuring that we saw that.

 

Anna [00:27:12] The other kind of imputation we do is for demographics or missing demographics, particularly race, for more of the observations than we'd like. And we're also- we're a little bit worried that the missingness of race is correlated with both non-prosecution and with re-arrest. So we rerun everything, including the information that we do have on demographics. We also predict race and ethnicity across our sample, using the information that we have on name and birth date. And then once again, we find that the results are really reassuringly stable once we include information on race or predicted race.

 

Jennifer [00:27:49] So I'll tell you what my favorite check. My favorite check is a placebo check. So a key assumption of our instrumental variable design is the exclusion restriction. So in this case, we're assuming that ADA leniency affects the defendants' outcomes only through the prosecution decision. So in Suffolk County, the ADA who handles the arraignment and makes the prosecution decision is not the same ADA who handles the case going forward, that's very helpful to us. So after arraignment, cases are distributed to other ADAs based on their workload. So that system means that our exclusion restriction should hold, but it's impossible to test it directly. However, we can do a placebo test. It's something that should turn up a null result if everything is working the way we think it is. So if the arraigning ADA only affects their trajectory and outcomes of a case through the decision to prosecute or not, then when we restrict our attention to only the prosecuted cases, the leniency of that arraigning ADA should not matter. And we can check this. We look at the effect of our ADA leniency measure on case outcomes like time to disposition of the case and likelihood of conviction within the set of cases that were prosecuted. And when we do that, we find no effect. This placebo check produced a null result just as we hoped it would. So this gives us more confidence that our understanding of how this process works is correct.

 

Jennifer [00:29:06] OK, so our results so far tell us what happens if someone on the margin of prosecution is assigned to a more lenient or more harsh prosecutor. But a very hot policy conversation right now is whether DAs' offices should dramatically scale back the prosecution for all defendants charged with certain offenses. And this could go beyond the range of marginal defendants that were able to say something about using the design that we are describing so far. But luckily, there was a recent policy change in Suffolk County and we're able to extend our analysis to begin to answer the question of what these policy effects might be. So, Anna, tell us about that policy change and what we find when we consider its effects.

 

Anna [00:29:44] Right. So our main results are for the period from 2004 to 2018. So they run up until just before Rachel Rollins is inaugurated. And I do think it's worth saying first that we also look at marginal treatment effects for that sample that is- consists of cases that were brought to the District Attorney's Office before Rachel was inaugurated. And that means that we're looking at how the treatment effect of non-prosecution varies across the distribution of ADA leniency. And what we find is that the treatment effect is growing in absolute value as ADAs become more lenient across that distribution. And growing in absolute value, but the direction is negative. Right. So we're seeing larger negative treatment effects as ADAs become more lenient. And that means that they are non-prosecuting a larger fraction of cases. And that's suggestive to us that a policy that expanded a presumption of non-prosecution would probably have effects similar to, if not larger than, the effects that we observe in the study.

 

Anna [00:30:59] So we look at the inauguration of Rachel in 2018. Rachel had campaigned on this platform of establishing a presumption of non-prosecution for 15 categories of nonviolent misdemeanor offenses. And we first- we do a couple of things. The first is that we use an instrumental variable design to predict non-prosecution as a function of the timing of Rachel's inauguration. And then we look at the effects of that predicted non prosecution on defendants' rearrest rates where we are only able to look at one year of rearrest data. We also use an instrumental variable difference-in-differences design, which includes a control group of nonviolent felonies. So now we're estimating the same effects, but relative to a control group of nonviolent felonies. And in both cases, we find that Rachel's inauguration, in fact, increased non-prosecution rates for the cases in which we're interested in—these nonviolent misdemeanor cases—and increased non-prosecution rates by about 15 to 20 percent. And that predicted increase in non-prosecution had an effect that was very similar to the effect that we saw for the pre-Rachel cases. Namely, there was a pretty large reduction in the probability of rearrests within one year after arraignment.

 

Amanda [00:32:16] Can I put on my interviewer hat?

 

Jennifer [00:32:18] Yeah.

 

Amanda [00:32:18] Now Jen, we also use that same policy experiment to look at what happened to reported crime after Rachel was inaugurated. What did we find there?

 

Jennifer [00:32:27] Yeah, so one of the big concerns about these types of policies that folks raise—including police officers in a lot of these communities and just concerned community members—when they hear about these kinds of policies where you're going to implement a presumption of non-prosecution for all of these offenses that we still might care about, right, we don't want people doing. If you say we're not going to prosecute or punish you for these things anymore then maybe just everyone starts committing more of these crimes and maybe it just increases crime rates. And so we're able to, again, look at the inauguration of DA Rollins to look at what happens to reported crime in Boston in particular. And we see no effect. There's certainly no evidence that reported crime is increasing as a result of this policy change. Depending on some categories, if anything, they go down, but there's definitely no increase in crime. Now, that could change over time. Right. And we only have about a year follow up. And I think, you know, we're used to all policies having trade-offs. So I think we would not rule out the possibility that there might be a reduction in what we call "general deterrence," that there're- you know, people might respond more broadly to these types of policy changes. But I do think one of the big factors to keep in mind here is that, you know, the policy that was implemented wasn't a blanket, "we're never going to prosecute this anymore." It was a presumption of non-prosecution. And so the DAs could still—or the ADAs—could still go in and prosecute a particular case if it really looked like the defendant was just flouting the law and just letting them go wasn't working. So I think that helps explain why we're not seeing really detrimental effects on crime rates, but we are interested to see how that evolves over time and in different places.

 

Jennifer [00:34:02] So what do you all see as the policy implications of these results and the other work in this area all combined? What should policymakers take away from this?

 

Amanda [00:34:12] As a tried and true academic, I'm always going to say, of course, we would like to see more research. But this and the other research that we've seen is really encouraging evidence in support of what many prosecutors have been doing already with increasing leniency and declining to prosecute some low-level cases. These policies here in Suffolk County are not decreasing public safety and in fact, if anything, increasing it and allowing individuals a second chance to avoid the mark of a criminal record. I think it also shows the power of research partnerships and data. And I hope one of the policy implications beyond the direct results that we found is that more and more offices across the countries should open their doors to researchers, share their data, to try to rigorously understand the impacts of the policies they have been implementing so we can get a better understanding of what works and perhaps of what doesn't. And what are the implications for each of these policy experiments, for defendants, the community and the office themselves.

 

Jennifer [00:35:10] I love that. Give us all your data.

 

Amanda [00:35:13] And particularly us.

 

Jennifer [00:35:15] Anna, what what do you think? Anything to add there on the policy implications?

 

Anna [00:35:20] Yeah no, I think that's exactly right. There's so much that we know, but there's also just a much larger set of things that we don't know. And if we want to make our policies more effective and more equitable, there's just- we just- there's no other way than to study their effects. And I hope that policymakers can take heart from Rachel's example of being so intellectually honest about wanting to know whether her policies were having a beneficial effect or not. And I hope that that's something we can see more of.

 

Jennifer [00:35:49] Yeah, I think all three of us are big Rachel Rollins fan girls now. I think that's the biggest result of this study. I will add and just sort of double down on what Amanda said about second chances. I think our results paired with the Mueller-Smith and Schnepel results on deferred adjudications in Texas and how they had such big benefits- you know, basically saying if you go through this probationary period and don't do anything wrong, we're going to wipe this felony charge off your record. And they also found the biggest effects for first time defendants. Right? So there- it was- again, it really seemed to be if you don't have any felony record before, then giving you the opportunity to avoid your first record had a really big benefit. And again, we're finding here- totally different context, totally different level of offense, no probationary period, but again, we're finding that the benefits are coming mostly from those first time defendants. And so this all to me really says that we should be erring toward giving more people second chances. And, you know, Amanda and I spent a lot of time thinking about how do we help people on the back end once people already have a criminal record? How do we help them reintegrate into society more successfully and all of that and it's really hard. It's really, really hard to reduce recidivism once people are already in the system just for a variety of reasons. And so helping people avoid getting pulled into the system to begin with just seems like a really- a much better place to focus our attention. And I think that is highlighted by the results that we're finding.

 

Jennifer [00:37:17] Alright. So next question. What's the research frontier? What are the next big questions in this area that folks like us will be thinking about going forward? Anna, you want to start this time?

 

Anna [00:37:27] Well, you know, I think one of the interesting things that I don't think any of us have really talked about that much before is that, you know, Rachel campaigned on this platform of introducing this presumption of non-prosecution for this defined set of offenses. And you would have thought the sky was falling from the reaction from law enforcement agencies in particular. And it turns out there was a pretty small increase in non-prosecution, even for the offenses that were on her list. It's something like a 15 percent increase in non-prosecution rates. These assistant district attorneys, they're in arraignment courtrooms generally by themselves, maybe with another junior ADA. There's not that much supervision or monitoring. The office doesn't have the real time capacity to monitor or to see what the ADAs are doing in these courtrooms. And so I do think one interesting next question is, how do you think about incentivizing arraigning district attorneys to make choices that are better for defendants and better for communities? What's the information that ADAs need to have to enable them to make those better choices? And one of the things that the three of us have been talking about is developing an interface for offices where ADAs and their supervisors can visualize in real time what their non-prosecution rates look like and what the outcomes for their defendants look like downstream. And I think that we'd all like to test whether having access to an interface like that could help to nudge ADAs in the direction of making more productive prosecution decisions.

 

Amanda [00:39:03] So I have several things written here, but I'll go with the first one, which is a question of where is the line? Where should we continue to expand leniency. Maybe, where shouldn't we? Our particular paper focused on nonviolent misdemeanor crimes. Most offices across the country that are implementing policies of presumptive non-prosecution are for the most part, focusing on nonviolent misdemeanor crimes. But nonviolent misdemeanor crimes are not the only misdemeanor crimes, or not the only crimes that are being charged and impacting defendant and potentially impacting their recidivism. What happens if we expanded this to some violent crimes, to domestic violence, to weapons charges? Is this something that's going to have similar effects or detrimental effects? Where- should we start thinking about felonies? What about nonviolent felonies? What would happen if we did something similar there? Because the extreme of this would be to say, let's not charge any crimes. That's presumably not quite where we're ready to go yet. And so trying to figure out where that line is, where offices should be focusing their time, money, and effort, and where offices should not be focusing their time, money, and effort and think about expanding leniency, I think it's going to be a really important question. And part of that will start with doing the same sort of research across the country in different offices that have chosen to non-prosecute different types of crimes. That will start to get us there. And perhaps there will be offices that will begin to experiment with expanded leniency, even outside of this kind of initially prescribed set of case types that will give us some more evidence on where exactly we should draw this line on expanding leniency.

 

Jennifer [00:40:45] Yeah, that was sort of going to be my my research frontier, too. Yeah. If you have others, feel free to pitch them. You know, we know in general, I mean, the US is just so punitive relative to a lot of our peer countries. And in general, I think we have a lot of evidence that reducing incarceration is probably the right direction to be moving just because there are diminishing marginal returns to incarceration. The first person you lock up is probably going to increase public safety. But by the time you get to locking up the really minor offenders that we lock up in the US, you know, the net benefit could be negative. Right. We could actually- there could be bigger costs than benefits. And so, you know, once we get down to these nonviolent misdemeanors, these are the most minor offenses. And we're not locking people up, but we are punishing them and there are costs involved in going through this whole process. And if most of them aren't going to go on to reoffend, you know, absent any intervention, then what's the point? And we could be doing more harm than good.

 

Jennifer [00:41:39] And so, as one of you said earlier, you know, this is the policy frontier and a lot of places are having very heated battles about whether to scale back the prosecution of nonviolent misdemeanors. So our results are certainly relevant to that policy frontier. But when we have been having conversations with other DAs' offices about maybe we could use other data to look at effects in other places, what we keep hearing over and over again is, yeah, that's all well and good, but that's the easy stuff. The hard stuff is what do we do with domestic violence cases and what do we do with gun cases? Because there, you know, our communities are going to give us hell if we don't prosecute these people. But it's not clear to us that there are really big benefits to prosecuting them. I mean, I think especially domestic violence is so tricky, right? Because it feels like it's so difficult to get people to report domestic violence. You could- that's the kind of crime where I could actually almost imagine it maybe going the other way, given the current context. But- so I'm just very curious what the effect of the prosecution is for those kinds of crimes. But, yeah, the punch line here is we should be erring more toward leniency. Then- that's all still on the margin. That's relative to where we are right now. But like, how far should we go? And I think we're just really far from knowing the answer to that question.

 

Jennifer [00:42:52] Amanda, what were your other other research frontiers?

 

Amanda [00:42:55] On a more micro level, trying to get direct measures of the employment mechanism in our particular context, in the Mueller-Smith and Schnepel paper, they had employment outcomes for the deferred adjudication cases for the felony deferred adjudication cases in their data. I would be fascinated to see- we posit that one of the potential mechanisms that this is coming through employment, that employers can see even a non-conviction, misdemeanor charge and that that might be having impacts on their employment outcomes. You know, our own work, focused on Ban the Box policies that had kind of a similar flavor. In the audit study that I ran, everybody had a felony conviction. You know, you alluded to the audit study, by by Chris Uggen and coauthors that had misdemeanor convictions. Right. And I'd like to know a little bit more about how these charges that didn't lead to a conviction are potentially impacting employment outcomes. And then I have some other research hopefully coming down the line that maybe one day you and I could talk about, once again, that might allude to that a little bit. But understanding some of these direct mechanisms, I think is going to be important moving forward, both from a research perspective, but also from a policy perspective.

 

Jennifer [00:44:08] Is there anything else that we haven't talked about that we should we should add in?

 

Anna [00:44:12] You know, one thing that I know we've talked about and we probably should mention is I think we just- our agenda for future topics is so large is that we didn't in this paper look at racial disparities in charging and in outcomes. But I know that that's something that we all are interested in and and are interested in finding ways to work with really good data on defendant race and some of the new, I think, econometric techniques for estimating racial bias in these kinds of leniency designs. I think that that's something also that we'd probably all like to dig into.

 

Amanda [00:44:44] Absolutely. That was the third one on my list, but I figured I should stop talking.

 

Jennifer [00:44:47] Great minds, great minds think alike. Alright, fabulous. Well, my guests for today's episode have been Amanda Agan from Rutgers University and Anna Harvey from New York University. Amanda and Anna, thank you so much for doing this.

 

Amanda [00:45:01] It was a pleasure.

 

Anna [00:45:01] Thanks for having us.

 

Jennifer [00:45:03] 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 Haley Grieshaber. 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 DoleacCourts, Pre-Trial, Re-Entry, Instrumental-Variables