Episode Transcript
[00:00:00] Speaker A: Imagine a justice system built on rigorous evidence, not gut instincts or educated guesses about what works and what doesn't.
More people could access the civil justice they deserve.
The criminal justice system could be smaller, more effective and more humane.
The Access to Justice Lab here at Harvard Law School is producing that needed evidence. And this podcast is about the challenge of transforming law into an evidence based field.
I'm your host, Jim Griner, and this is Proof Over Precedent.
Welcome everyone to another episode of Proof Over Precedent. I have with me today. I'm Jim Griner, your host, and I have with me today two of my favorite people. I've got Sonya Starr and J.J. prescott to talk about a paper that they published a few years ago and work that they're continuing to do on expungements in Michigan, expungements being one class of record clearing, criminal justice record clearing. And so Sonia, thanks so much for interviewing today. Can you tell us a little bit about yourself and jj? Get ready for the same set of questions. Sonia, lead us off.
[00:01:10] Speaker B: Yeah. So I teach criminal and constitutional law at the University of Chicago. I taught at Michigan for many years, which is when JJ and I started working on these issues in Michigan.
And, and I do kind of a mixture of quantitative empirical work and other and kind of more traditional legal scholarship.
[00:01:34] Speaker A: Super.
And what's a fun fact about you?
[00:01:39] Speaker B: Well, oh, let's see. I'm supposed to come up with a fun fact.
[00:01:42] Speaker A: Yes, you are.
[00:01:43] Speaker B: So I thought you were going to remind people of my or to tell people about my karaoke prowess.
Let's discuss that decade ago. I do love taking students to go do karaoke where I tend to rock 90s classics as well as Disney theme songs like Let It Go.
[00:02:06] Speaker A: And your karaoke prowess is formidable. So for anyone listening prowess may be.
[00:02:12] Speaker B: An exaggeration, actually, but enthusiasm is well founded.
[00:02:16] Speaker C: She also loves to score baseball games.
[00:02:19] Speaker A: I did not know that. Terrific. All right, terrific. Jj, can you give us a little bit of a background on you?
[00:02:25] Speaker C: Sure. I'm a law professor at the University of Michigan Law School. I'm also an economist by training.
I teach criminal law and courses related to economic analysis of law. I do a lot of empirical work also on criminal justice. I also write pretty regularly on litigation dynamics and also employment law, most recently on non competes and their impact on employment outcomes.
A fun fact about me is that this is my second year in a row of being a bluebird monitor at my local state park, Mayberry State park, where I once a week have to check in on a bunch of nests and record the status of both bluebirds and other types of birds. And fortunately, I've never found the kinds of birds that I'm supposed to get rid of. So that makes me very happy.
[00:03:16] Speaker B: I would like to register an objection that JJ's fun fact is less embarrassing than mine.
[00:03:20] Speaker A: I don't know.
Jj, do you give the birds names or. Or.
[00:03:24] Speaker C: I do not. I do not. You know, it only takes two or three weeks before they go from egg to flying out of the nest. But I will say I do. I do have to battle every time I go up there, both ticks and poison ivy. So I think the, the risks of karaoke might be much less.
[00:03:42] Speaker A: It sounds like hazardous duty. And of course, the children, they leave too soon.
But anyway, let's go on. We're going to talk about your article in the Harvard law review in June 2020. And of course, we'll include the link, the posting Expungement of criminal convictions and empirical study by both of you. Can you tell us.
Let's just actually suppose we're riding up in an elevator. We're going up 20 floors and we've got two minutes.
So tell us what is the just what's elevator story? What's the elevator lesson from this paper? And then we'll get into how it came about and some more details. What's the elevator summary of this story?
[00:04:24] Speaker B: We studied expungement in Michigan, which is basically in Michigan. What we mean by that is sealing of records. They're not eliminated for all purposes, but it makes it so that you can say no if asked on an application about a record, and it won't show up in a background check.
And we studied both the effects of expungement on people who received it, which were very positive. We found increases in wages and employment levels. And although we couldn't study this causally, we found quite low crime levels, in fact lower than the general population for people who received expungements. However, the bad news of our study was that under Michigan's old regime, in which you had to jump through a bunch of hoops to achieve an expungement, very few people who were legally eligible got them. So we looked at a pool of eligible people, tracked them for after they became legally eligible, and found that within five years, only 6.5% of those eligible received it. So we think that makes an argument for making expungement easier to receive and. Or for automating it, as Michigan has ended up doing for a large class of people with records.
[00:05:45] Speaker A: Super. And just to define some terms Here the Access to Justice Lab is doing some work on expungements as well. We've had a long running study, a randomized study in Kansas in the area that we'll be discussing on proof of repression as well. There's expungement, which people typically mean a petition based court process where you have to sort of do it one by one or you know, one, one set of criminal records at a time and it has to be initiated by the person who is subject to the criminal records. And then there is a broader type statute that's more or less automatic. Sonia, you mentioned that people sometimes refer to that as clean slate. What is. Or clean slate legislation. What does that do?
[00:06:28] Speaker B: Yeah, so the lingo can be a little confusing because clean slate is just the lingo adopted by a movement to make it easier to get expungement or record clearing or it's formerly called set asides in Michigan. So these things go by different names. But clean slate is the movement language for increasing access and particularly for making it automatic so that an algorithm can basically review the state's own records, figure out who has a record that has met the waiting period and meets other qualifications for being exposed expunged, and then they can just mark it as sealed within the state's records automatically so that a person doesn't have to fill out an application for it.
[00:07:23] Speaker A: And the Access to Justice Labs also doing. Done some work on that area. It recently came out we're going to feature that on proof over precedent about among other things, difficulties with those algorithms when they are attempted, that when they are implemented.
It turns out that sometimes the algorithm is very difficult to write because the statute is very complicated or sometimes the state of the data is such that you can't tell whether an offense is expungeable or whether a record is expungeable. And so we talk about that in the future. Just wanted to flag it.
So we've got. But let's. And so again let's just follow the. As you said, we'll say expungement is petition based and clean slate is some attempt to broaden access beyond just the petition, perhaps making the process automatic where the person subject to the record doesn't have to do anything in order to have it. Have it cleared.
[00:08:14] Speaker B: Yeah, yeah. The actual relief obtained is the same. Is the same basically just rent in Michigan anyway, the new clean slate law, it actually broadens access to petition based expungement too. But then for a substantial class of people, hundreds of thousands of people, it makes the, the expungement or the set aside in Michigan occur automatically.
[00:08:42] Speaker A: Yeah, terrific. Jj, tell us how this got started. You know, where did the glimmer of the idea begin? And where did you go from there?
[00:08:51] Speaker C: Sure. I mean, in this case, Sonia and I were fortunate to have a conversation with Miriam Aukerman, who is now a lawyer for the ACLU in Detroit, but at the time was working at Western Michigan Legal Aid and. And spending a lot of time on expungements. And hadn't, I suppose, she sort of brought our attention to the access issue, in part just by notifying us that a lot of people, even though There was a $50 application fee, a lot of people spent many months just trying to get that much money together.
And she encouraged us to look into this topic because. Because in her view, record relief, it was used far too sparingly in Michigan. And she thought for a lot of the people that she had interacted with, it had been, you know, it had led to really great results in their. In their personal life. And doing some more quantitative, combined with qualitative work to really understand what we're seeing in the data that was a little more than just her anecdotal. Her anecdotal stories would be really helpful in guiding future reform. And so that's how it started.
The project was, I think, for both me and Sonia, one of the longest projects we've ever worked on, in part because of the difficulty of accessing data. And I'm sure, Jim, that on this podcast you talk regularly about both data quality, access to data, if not every single. Every single episode. And we had the same set of issues in order to. To do what we wanted to do. We not only had to access records that had been cleared, which sort of by definition are designed not to be accessed, but we had to link those records to other state data that would give us some insight into potential outcomes, in particular future criminal records, as well as employment outcomes like wages and employment status. And the time it took us both to figure out who we needed to work with to set up a data sharing agreement between different agencies of the state of Michigan to then work with them and their data people, as well as their outside consultants who often handle the data.
Just, it was a real struggle, and it was a real struggle that oftentimes put us off the project for big chunks of time while we were trying to come up with different ways of approaching it. And honestly, our story is one of serendipity.
We kind of hit a barrier. Barrier that we were coming up with, you know, two Clever by half solutions.
And then there was a personnel change that led them to being much more reasonable about what, what they would allow us to, to do to find the kinds of data we needed. And, and, and we were off and running.
[00:11:52] Speaker A: And so tell me, who's the them? In your sentence you said led them to be much more reasonable. Who was the them?
[00:11:58] Speaker C: Yeah, I mean, there was just a change in the director of, of.
Of the Criminal History Division in the Michigan State Police at the time.
And you know, there's a lot of. There, There are regulations and statutes that govern how the state of Michigan works with and shares data, and people differ on how they interpret those rules. And the person we had been working with for a while, who was great in lots of ways, but required a process that was very difficult for us to, to, to use. Namely, we, we had to come up of records to generate a list of potential people we might want to use that we would then provide to the Michigan State Police. So we spent a lot of time working with the Michigan courts to figure out how to do that. And it was just a long process that had many strange turns and curves. And fortunately we wound up being able to do it much more cleanly and better in the long run. Although also as a part of that, we had some data issues along the way, which we talk about and acknowledge in the paper.
[00:13:06] Speaker B: I want to add, of course, that we're very grateful for the help that we ended up getting both from the state police and from.
We got unemployment insurance agency data to measure our economic outcomes. Then we had a different state agency working to integrate all the data for us. Partly I was just coordinating a lot of moving parts against the background that makes this kind of research difficult in every state, I think. I don't want to blame it on Michigan bureaucrats. Right.
[00:13:36] Speaker C: In fact, I'd even go the other way.
[00:13:37] Speaker B: The fundamental difficulty involved in expungement research is that expunged records are secret. Right. That's the point.
And so there are in fact state rules governing access to them. And so for us to be able to access them, they needed to be de identified and, or there might be other ways to access them, but it would have been under different kinds of security agreements than the ones that we had at the time.
What our agreement was for was de identified data. But then to identify the records, to figure out which records you need when you can't look at the records yet involves some creative workarounds. And so that's. In some ways there were some kind of fundamental research challenges and we ended up finding relatively easier, although still somewhat complicated, ways to resolve them. And eventually we got great data, which we're very happy about.
[00:14:43] Speaker C: And Sonia's right. The amount of time and effort and generosity of spirit of the people we worked with, they were great.
And everybody was just trying to follow the rules and deal with communication across multiple agencies and academics who were also at that time both working at a state agency, in a sense, at the University of Michigan, which just made it all. All the more complicated in some ways.
[00:15:07] Speaker A: Super. One of the things we try to do at Proof Over Precedent is talk just a little bit. We don't want to get too far, but it's just a little bit about some of the complications of running field research, of doing field research.
And so how. I mean, so far, I can tell, you needed the Michigan State Police, you needed the court system, you needed the department of the equivalent of the Department of Labor, the state Department of Labor, the unemployment agency that's going to run the unemployment insurance program. And the unemployment insurance program is the source of the wage data. Because in order to tell how much people have to pay into the unemployment insurance system, you need to know how much people are earning.
And so. And so that's. They have. And so. And then you. And then, Sonny, I believe you mentioned that there was a. That there was another state agency that matched the records in order to. Because they could do it sort of behind the shroud of officialdom and then de. Identify it and give it. And give it to y'. All. So that's four agency. That's four separate entities of government that you have to sort of, you know, each of these cats you have to herd. Am I missing any?
[00:16:17] Speaker B: That was all. Although if we had wanted to do research on other outcomes, like if we thought about maybe use of other social services of various sorts, we thought about maybe integrating corrections data, which would have allowed it. We actually focused our main analyses on people who were never incarcerated or served very short carceral terms so that we didn't have to worry about the fact that we otherwise would have had great ambiguity in when they became eligible for set asides, because it would depend on their release date from prison.
[00:16:59] Speaker A: Set asides, again, are expungement under Michigan law. Right. Go ahead.
[00:17:03] Speaker B: We had to, in some ways, in order to not prevent the project from sprawling totally out of control, with needing too many different data sources, we had to constrain the questions that we asked.
[00:17:20] Speaker A: Yep. And one that we've tried to take a look at the lab just to preview is housing via both survey data and administrative data.
You know, so HMIS homeless management information system data and then, and then some survey data. So that'll be coming up later.
Okay, so you've matched all this, these, you know, these various data sets, or with the assistance the state agency has matched all the various data sets de identified. You've got a data set and what are you going to do with it? So let's walk forward in terms of the research questions. Basically, from what I can tell, you're asking, first of all, what is the uptake rate of this potential remedy, which is expungement, criminal justice record clearing, petition based, criminal justice record carrying. Second is recidivism. Third is employment. And fourth.
Well, those are the big three. Excuse me. Those are the big three.
So how are you. Can you give me a little bit more detail on the research questions? Sonia, why don't we go back to you for this one? What are some of the. How are you defining a little bit more sharply these research questions?
[00:18:27] Speaker B: Okay, so for the more complicated one, even though the statistics themselves were sort of simple descriptive statistics, the more complicated one from a data perspective was analyzing the uptake rate of, of set asides, that is the rate at which legally eligible people get set asides. Because that's where we had to identify a pool of people who were in fact legally eligible by coding, by sifting through a larger pool of people with criminal records and coding, you know, what are the things that they've been convicted of? How long has it been since the last convictions? We had some data limitations where we could only really code up to a certain year.
Sometime in 2012 was when we lost a date field that we needed in the data that we.
[00:19:31] Speaker A: This is the quirks that just happened. Right?
[00:19:34] Speaker B: So basically, because we actually ended up with a full sample of everybody who had ever gotten an expungement in Michigan. But the, but our cohort sample, the larger eligible cohort sample that we had to identify, we chose like a five year range of dates to work with and worked on accurately coding so we could pull that sample accurately.
[00:20:06] Speaker A: Let me stop you one second. Let me stop you one second. Just ask a follow up question. You're doing all this in roughly what years? Roughly sort of 2017, 2018 timeframe frame. Is that roughly correct?
[00:20:17] Speaker B: Yeah, I mean, it took us, as JJ said, you know, we started working on this project like negotiating with the state for the data in 2010.
No, 22,009 was when we made our first data request, actually.
So. And it's papers published in 2020. So when JJ says it's the longest project either of us has worked on, that's why. But, you know, but some of it was the grant applications and the, you know, we, we had a lot of steps to go through, But by the time we really have the data in hand and it's in a point where we can start analyzing it. Yeah, it's about that time, about 20. So about 20 at that point goes back. I mean, our last actual date in the outcome data is I think, 2014 or something like that. So that's when the data range ends. And at that time, Michigan actually had a super stringent set of rules for getting set asides. Actually, in some ways it was pretty broad. It was not substantially limited in terms of what kinds of crimes qualified. It excluded sex offenses, as is routine because of sex offender registry laws. That's routine in every state.
It excluded class A felonies, which were those that you can get a life sentence for.
But basically, substantively, there were few other exclusions in, in the law. But it did have one very strict exclusion or limit, which was that you could only ever have been convicted of one crime in Michigan.
So you could only get one crime expunged, and you couldn't have any other crimes on your record, no matter whether they were very old crimes or more recent crimes. And then you had to go five years since that conviction for it to become eligible. And so one question.
[00:22:13] Speaker A: You could get a conviction, A conviction could be expunged. Is that right? Because not all states allow that. Some of them only allow arrests that don't result in charges or charges that don't result in convictions. And in Michigan, that one crime could be a conviction.
[00:22:27] Speaker B: Yes, that's right.
[00:22:28] Speaker A: But it could only be one.
[00:22:29] Speaker B: That is something that has. So in virtually every state, they have more some way to get non convictions expunged. And in most states, they now have some way to get convictions expunged. But the, the details of the laws really vary very widely across the states. And, and in some states, you know, it's only misdemeanors you can get expunged. In Michigan, it always did extend to pretty significant felonies, but only one.
And that in some ways, from a research coding perspective, that meant that made some things easier for us. It was easier to deem people eligible or ineligible.
But it does mean that when we think about our results, we, we're basically estimating results in all of these categories, whether it's the rate of obtaining a set aside or the outcomes thereafter.
You should think about it as people. People who have pretty limited criminal records and records that were at the time of the expungement at least five years old because that's the. The.
Was the waiting period in Michigan. Michigan still has pretty long waiting periods, but it has liberalized those number rules to some extent. And it also has like it used to be that even if you're. Even if you had two, a two count conviction stemming out of the very same incident, that itself disqualified you. And so we looked at some points, we were doing some comparison. We were using some comparison groups with people who have.
This is kind of for secondary analyses, but comparison groups are people who have one extra misdemeanor that was really old or maybe two convictions stemming from the same.
From the same incident. But basically for uptake though, we wanted to code people who were eligible and then we just asked. We just basically report the percentage of them that received a set aside Eligibility can't be coded perfectly. We don't have out of state convictions, but we looked at some data on how often people who have a record in one state moved to other states and stuff and made some background assumptions about the extent to which we might be over coding eligibility. And I think it's not, you know, we have various alternative estimates in our paper based on different assumptions about stuff like that, but pretty much we think our best guess is about 6.5% get it within five years. And I think we said maybe about 11% might get an expungement over the course of their lives. So the majority of those who are getting it are getting it within the first five years. But the overall lifetime rate under the regime where you have to apply, at least at the time, was pretty low.
So that's the uptake result.
[00:25:40] Speaker A: Yeah. And jj, just a couple of details about that. First of all, the five year period that you studied, what did that mean in terms of the calendar you're looking at? So is that sort of 2000 to 2005 or what time period was that in terms of when you're doing these analyses or what you're looking at?
[00:26:00] Speaker C: Shoot, Jim, I don't remember exactly when the range was.
[00:26:04] Speaker A: See if I could stuff you once.
[00:26:06] Speaker B: 99 to.
[00:26:08] Speaker A: Yeah, so around that. Right. In other words, right around the turn of the century is when we about.
[00:26:11] Speaker B: And then we track them. So then there's the five years that they wait are before they become eligible.
[00:26:18] Speaker A: Yes.
[00:26:19] Speaker B: And then there's another five year period. Whatever we defined it as, whatever our cohort was, it actually may have been a less than five Year cohort. It might have just been a couple of years, three years or something like that.
We defined it in a way that we would have enough outcome data and remember, as I said, we lost the date field. We needed to measure when they got a set aside in 2012. And so we counted backwards from that to define what we were, the cohort we were looking at.
[00:26:51] Speaker A: Got it. Okay. And then, and so we've got, we've got a, basically, you know, one way of putting it as sort of a, you know, best estimate, best point estimate, somewhere around six and a half percent, you know, over five years, but very, you know, maybe somewhere between 5 and 8% over lifetime. Around 11% of folks are getting. Another way to think about that is that 90% of the people are eligible or not are not getting this, this expungement. And so jj, maybe you could just give it, you know, this is, you know, that we care about access to justice here on proof over precedent. This seems like possibly a pretty severe access to justice problem. Can you just give us a real broad overview? What was, what was required to go through this petition process? And do we, you know, is there reason to think that that might be the culprit in terms of why so few people.
[00:27:36] Speaker C: Certainly. So one of the things we do in the paper is we spend some time talking to practitioners in the area about what in their view seems to hold things up, what are the sources of friction that lead to this. And not surprisingly, one of the biggest, and we should come back to this when we talk more later about automatic expungement is just awareness that it's even possible.
So certainly a large fraction of people don't understand that there's such a process as expungement. And it's something that they can pursue if they have the money and the time and the energy to do it. And so they're sort of off, off, off the trail right from the beginning. But even if you learn of it, there are financial hurdles, psychological hurdles, process related hurdles, and the financial ones are significant. People who are looking for expungements are often doing so because they are struggling to get the kind of job that they want or the kind of housing. They typically don't have the money they need to hire an attorney to help them navigate the process, or they're kind of following if there's self help available. They're trying to do that, but oftentimes they are flying by the seat of their pants.
And that's scary and it's intimidating. And some of the requirements are Often exacerbating some of these features, like the need to get fingerprints at the police station or returning to court for a hearing, even in a situation where there was unlikely to be any objection. And so people who've had a bad experience with the criminal justice system, the psychological burden of returning to a police station to be fingerprinted and to run an FBI search on your record to verify your eligibility, all of that can be traumatic and unpleasant. And so if you think about people being people, it's the kind of thing you put off and you hope you won't need to do.
And that's pretty much what we, you know, not surprisingly, we find that, you know, there are 10 to 12 steps that you have to go through, including various fees, not like a single fee, but different fees, fees for fingerprints, fees for the FBI, criminal record search, and court related fees as well. And so, you know, they add up. But on top of it is just the complexity of the process and the fact that it's overwhelming for many people who are in the category of individuals who are probably the most likely to benefit from this kind of relief.
[00:30:10] Speaker A: And one of the things I was struck by, and this also dovetails with some additional research in Kansas, the Access Labs published a Kansas Law review piece that will feature in a later podcast about the process in Kansas. It also has between, based on how you count, somewhere between eight and 15 steps that you have to go to.
It resonates with Yalls earlier published article 2020.
One of the things I was struck by was that you have to give the court information that it already has that is beyond what it needs to identify the transaction. In other words, I can understand a need for the court to identify. Okay, you want to have some record expunged, you need to tell us enough for us to identify the transaction that we're going to, about which we're going to suppress the information. Okay, I get that. Right. You know, maybe the court case number or something like that. Right.
[00:31:03] Speaker B: But beyond certified record of conviction, they have this.
[00:31:08] Speaker A: That's exact. There is exactly where I was headed. Why does the court system need to have you feed it information that it both already has and in fact it has the official copy of that information. It is the official source of that information. You need to go to the court to acquire the information, obtain it, and then give it right back to the court? I mean, I'm, I'm a little bit lost on that one. Was there ever any explanation?
[00:31:34] Speaker C: Yeah, Michigan is decentralized, so a court would have to go and get that from the Michigan State Police. And the Michigan State Police were also the agency tasked with running the, the national search for records. And so it is. There is a sense in which, at least in Michigan, although I'm sure that it would be quite easy for a judge or a clerk at the court to get that same information, they would have to do something more than just check their own case management system.
[00:32:04] Speaker B: Well, the criminal record, that's true, but the certified record of conviction that they had to go to the court for.
[00:32:10] Speaker C: I see, I see. The actual conviction they're trying to print.
[00:32:13] Speaker B: Record, that part is for a background check. That. That's true. The court can't, at least at the time, it didn't have a means to do it itself. I don't know if it.
[00:32:20] Speaker C: I mean, the thing I find interesting about it is the, the focus on identity, Jim. So I don't know whether or not, if you're petitioning in the states that you've reviewed that one of the things you need to do is establish that the application is being filed by the person who is likely to benefit. And one question is, why is that necessary? I mean, I've thought about this in the context of other work where people are seeking relief. Like, what is the theory that you might put together a poor expungement application and that there would be, as a result, some negative prejudicial consequence from that? That seems unlikely to me, that. But, you know, nevertheless, you actually have to go out of your way to prove that it's actually you that is applying for it. Even though, I mean, it's hard to understand who's going to be out there paying lots and lots of money to, to expunge some particular person's record that, you know, that, you know, that's not that person, or at least somebody who's on their side. So, you know, there's a. There's sort of process identification features that carry over that might not make a lot of sense in the context, you know, that other, cheaper forms of ID verification might be just fine.
You know, for example, just having a license showing that your name is the.
[00:33:37] Speaker A: One that's on the state ID card. Something like that. Yeah, exactly.
And again, I want to move on, but just one bigger, bigger picture. The, the bit about needing to go to the state police. Of course, if we had data systems where the state police could talk to the, to the Michigan court system in a way that it would allow easy, easy acquisition of records or, you know, a unified record system, you know, we wouldn't have that problem. Either.
[00:34:02] Speaker C: And that's a lot of what's been designed. And one of the big. I mean, those are the big price tag for automatic expungement in Michigan was of course, Michigan State Police coming up with a way to evaluate this while also integrating with all of the different courts where these records existed. And although I don't know the details, I'm not sure if anybody really knows the details of how it all works now in Michigan.
Certainly much more integrated than they were in the past, thanks in large part to some great leadership at. At scale, the State Court administrative office here in Michigan.
[00:34:37] Speaker A: Super.
Okay, then, then. So let's, let's take. So basically then going back to. Let's talk about outcomes. But how did we. What sort of, you know, estimation strategy did you apply to the outcome? So you create a cohort of people who are eligible for expungements but didn't get them. And you create. Then you have a set of people, you know, proverbially, because these are all. It's all de identified. You have a set of people who were eligible for expungements at the same time and did get them.
And then basically you compared the two on these various outcomes using some econometrics, some statistical models in order to try to adjust for differences. Is that. Do I have that roughly correct?
[00:35:22] Speaker B: Sort of. Except that our main analysis. So the only one that we really tried to do a causal analysis on was, was the economic outcomes. And that's partly because on the crime outcomes, we couldn't really use the same before and after type of analysis that we did because by definition, you had no crime before the before.
[00:35:46] Speaker A: That's back to that one conviction rule in Michigan.
[00:35:48] Speaker B: Right, exactly.
So the same methods of measuring trends before and trends after would lead to the misleading suggestion that except expungement increased your chance of crime, since you couldn't have gotten it if you had any crime before. Right.
So for crime, we basically just look, you know, we had their full criminal records because that's. Because we got those from the state police. And so we could see the rate of both subsequent arrests and subsequent convictions.
And we could.
And then what we did was for the same years corresponding to the cohorts we were analyzing, we pulled over, like, we could look at our comparison groups to see their rates of rearrest and reconviction. But given that we were only trying to make a general descriptive claim, I think a more interesting claim that the more interesting conclusion we were able to draw is that our group of people who received expungements actually are arrested at lower rates than the average person in Michigan.
Now remember, of course, the average average is across people who never commit crimes in their lives and people who commit a lot of crimes in their lives.
It's not a rate of zero, but it's still quite a low rate.
And we did that just by pulling the general data that the police report for the frequency of arrests. And we were able to, you're able to look at it age stratified and to make certain comparisons within group. But basically no matter what type of comparison we drew, the set aside recipients ended up being less risky on arrests, on convictions, and really very low risk when it came to the types of things that we might care the most about, like felony reconvictions or violent crime reconvictions. Super, super low risk.
And so we, and you know, we don't want to make a causal claim about that because it could be that they were just very low risk to begin with and that it's not that the, that getting an expungement helped them to stay, to stay low risk. Right.
And that's.
But, but it could also be that getting an expungement helps them to do things like get jobs, as we show with our economic outcomes, and that that itself reduces crime risk. We can't really disentangle those two things in our results.
But the idea that they're at least quite low risk to begin with is consistent with a large criminology literature on desistance from crime, which basically shows that when people have a long period in their life in which they don't commit another crime, like the, the big risk of criminal recidivism is shortly after somebody has been convicted and punished for a crime. And if they're going a minimum of five years, which is what the waiting period was in Michigan for a set aside and in practice on average several years longer than that, then their risk is going to be quite low to begin with. And so that's probably some of the reason for the low crime rates then for the economic outcomes.
[00:39:33] Speaker A: Oh, actually the. Hold on, but just got a couple of questions before we go into the economic outcomes. But jj, it sounds like you were going to add something there. Go ahead.
[00:39:40] Speaker C: Yeah, I mean, in some sense we were in looking at recidivism, trying to address one of what we think of as the key arguments against expungement, which is that this information is somehow helpful to public safety. And it's actually difficult to find somebody who sort of makes that claim in an explicit way as opposed to we just have a right to know. We're not necessarily saying that we can use the information in a way that would make ourselves safer. It's sort of surprising, but we view it to be that that's the implicit claim. And so to discover that at least with a waiting period.
Right.
That you achieve essential desistance.
Our hope was to knock that argument out of there. I mean, so there is selection into this class. People might be on the up and up, getting better jobs, getting better housing, but for a system that has a waiting period, the people who are opting into this petition based system are low risk. And so you needn't worry that that information is going to be particularly valuable to you if your focus is on public safety. And I think our view is that there are a lot of people out there who are interested, of course, in the rights of people to start again or the dignity interest of living without this criminal record hanging over your head for the rest of your life. But that many people are focused on public safety and whether or not this is likely to make them safer or less safe. And so we wanted to address that by looking explicitly at recidivism rate. So regardless of the cause, whether it's selection or some kind of causal story, the main message was you shouldn't be worried with an expungement system set up like Michigan has, that dangerous people are having their records removed from the public view.
[00:41:41] Speaker B: And I would say we testified in favor of the expansion of Michigan's law. And I would say most of the questions that we got were about the crime issue because that's what people are worried about. And in some ways you could push back and say, look, this was under Michigan's old, very stringent law and it's the people who jumped through the hoops to get the set aside. So this is like a self selected sample. So it's not very surprising that they're low risk. But what happens when we expand access to expungement to a larger group of eligible people and automate it so that we're no longer filtering people out by their hoop jumping? Right.
And I think that our answer there is you can. It is probably the case that the crime rates that you will then observe among people who have received expungements are not going to be as low as the ones that we observed. It is a self selected sample that we observed.
But if we think logically about potential causation in this situation, there is no theory out there and there is no empirical evidence out there in any study that suggests that making somebody's criminal record or keeping somebody's criminal record public indefinitely protects the public from their future crime. From their future crimes. Right.
It is true that like maybe some potential landlord out there might do a background check and say, well, I'm not going to rent to this person because they have a criminal record. And maybe that kind of move you could think of as protective in some way toward that landlord. Right? Or toward a single employer. But the, but the people in question, the people with the records, they're still out there in the community, right? But now they have no house and no job. Right.
If they don't have access to records. And so if you think in the aggregate about public safety, are we better off as a society by taking people who are not behind bars, they are released into the community and making and erecting essentially permanent, if not total barriers, substantial hurdles to them finding adequate housing and employment? Is society going to be safer?
Our research, I don't want to overstate it, doesn't directly answer that question. But the broader literature on predictors of criminal recidivism in criminology and economics doesn't support that theory. It supports the idea that in general it's good to get people stable jobs and housing.
[00:44:27] Speaker A: Just one quick note. Am I correct that it's not just a self selected group, but also a court selected group? So for example, if the court is extremely good at predicting who's going to stay in the same on the, on the straight and narrow going forward, then that might also raise some of the, some of the questions that you've just articulated potential answers to. Sonia, is that correct?
[00:44:49] Speaker B: Yes and no. So it is in theory a court selected group. In practice, the Michigan State Police told us that of all applications submitted, about 3/4 of them are granted. And we think that most of the ones that we don't have hard numbers on this, but most of the ones that are not being granted, it's because of the background check discovering that they weren't legally eligible. They weren't eligible by all anecdotal accounts, like when we talked to experts in the field about why some people were not getting set asides, it was basically never the judge said no or the prosecutor came in and fought it or something. In general, when people who are legally eligible pursue them, they might run into, as we've discovered with some of our later survey based research, they might run into some technical problems like an error in their criminal record in the data set. But if they, but once those technical problems are resolved, they end up getting the Set asides. Now it could be that somebody who committed a particularly notorious crime or the type that they knew was going to make the judge look badly upon them. Maybe they wouldn't apply in the first place.
So that might be part of the self selection process is kind of anticipation of what a judge would be likely to do do.
[00:46:15] Speaker C: And by the way, I, I'll just add that our, our more recent work where we worked with Detroit Project Clean Slate, which is a, a government agency in Detroit that that essentially represents applicants if they're eligible. They tend to, to, to represent them. They don't usually, they don't screen them out because you know, they have, let's say a particularly bad eligible conviction and they almost always, I mean I, I think it's like 99% if you know, if the person's eligible, they wind.
That's some other evidence that supports our claim in the earlier paper that really judicial discretion, prosecutorial discretion is not a big part of the story here.
[00:46:57] Speaker A: And by the earlier paper you mean this paper?
[00:46:59] Speaker C: Yeah, that's right.
[00:46:59] Speaker A: Correct. Yeah, exactly right.
[00:47:01] Speaker B: We've been doing some subsequent research to try to follow up on that in various ways.
[00:47:06] Speaker A: But, and I want to be conscious of time, we may want to ask you back to talk about the subsequent research if you're willing to do it before. But let's see if we can, we can cover the employment outcomes because there is a key assumption there that I just want to make sure we get out on the table as well as the outcomes themselves. So let's get the outcomes themselves with respect to employment. You use the wage data from the state equivalent of the state Department of Labor. And then jj, let's hear from you, what were the top level results there? First of all, many of these are in terms of an odds multiplier.
So what are odds and what is a. Basically a multiplier and how should we interpret that? How do we understand that?
[00:47:47] Speaker C: Well, shoot, I haven't looked at the table in a while. So but we are looking at, we're basically using a dichotomous outcome here, at least for employment. And so you can use logistic or probit or whatever and you turn your estimates into some sort of ratio that allows you to, to think about it in terms of percentage change. And what we find is that employment rates for people who mostly moving from people who are unemployed to having some form of employment, there's a 25 percentage point increase in employment. So from our perspective that's really a huge finding. And I think it's important to Add a few caveats. For example, it might be that some of these people were actually working in the informal sector and moved over to a more formal sector, which would then be subject to unemployment insurance.
[00:48:48] Speaker A: Actually, let's stop there. So if you, if you were doing app ride or, you know, if you're driving for Uber or doing app work or if you're being paid sort of cash under the table, that would not appear in the data set. So it's not a problem with the analysis, it's just that doesn't appear in the data. Because. That's right, that doesn't get reported to the Department of Labor.
[00:49:05] Speaker B: Hoover did not exist in this era, I think.
But yes, informal employment wouldn't get. But you know, we consider it in some ways like a win for society to move people from under the table.
[00:49:20] Speaker A: Employees, among other things.
[00:49:21] Speaker B: You get unemployment insurance, it's because it's better for them. And you know, then society gets to tax them and regulate it.
[00:49:30] Speaker A: And then we also see self employment. Right. Certain forms of self employment are also covered that you wouldn't appear on the data set. But anyway. Go on, jj. Go on.
[00:49:38] Speaker C: Yeah, and you know, then. So of course, one of the big challenges for something like this is just identification. So the fact that somebody gets a set aside at an expungement at a particular point in time, that could be either because, listen, something bad happened in their life, they lost their job, things look, look pretty terrible. And so for a couple of months, and then they decided, well, things are bad, I should, maybe I should get that expungement. But then things just return to the baseline. So you had so some sort of regression to the mean. And we attribute the change, the improvement, to the expungement rather than to an actual improvement in their actual status that's due to the expungement.
[00:50:23] Speaker A: Got it. And so just again, put that in as lay terms as possible. If it turned out that a heavy percentage of your folks were applying at the same time and motivated by the same time, they were looking for additional for better employment, and that was sort of driving both the timing of the employment search and the expungement, then that might cause problems with your estimation techniques.
[00:50:51] Speaker C: It's hard to know which way it goes too, because people who actually anticipate the ability to get their record cleaned up may start investing in themselves as a result of it. So they start getting education, they start preparing for the job market.
And that is also a function of the existence of this form of relief. So it is pretty complicated. And this is Actually one of the great things about automatic expungement now, right. Because automatic expungement brings an important challenge, which is at least with petition based expungement, we know that people are aware or mostly aware, let's say that, that, that they either are applying for it or that they received it. Some of our recent survey evidence on this question emphasizes that people are still oftentimes confused about what's going on with their, with their case, but, but they at least know that they're trying to work on this. Whereas with automatic expungement there are presumably lots and lots of people whose records suddenly become clear at an almost random time, but are unaware of the, but conditional on being aware of it. They are aware of it, then you have almost a random time at which it applies to their record. And so this may give us more insight in the future about the causal story here.
[00:52:16] Speaker B: Yeah, I think about the challenge of self selection, or rather of the fact that our sample is so selected. The fact that, that it's not just that these are people who jumped through the hoops to get the set aside, but that they chose to do it at a particular time. I think about it as posing two different types of challenge to the interpretation of our result. One is a causal inference challenge, an internal validity challenge, and the other is a challenge, an external validity challenge, a challenge of extrapolating the results to a different population. Right.
So on the causal story, you know, we have like, we'll get a cleaner answer to the causal results when we are able to carry out the same analysis on automated expungement recipients. And that's, and that will help with both concerns. And that's one of our next goals.
But with the results that we already published, I mean, we did a couple of supplementary analyses to try to.
We have basically a couple of reasons to think that the results are probably causal. One is that we see them kick in only once the person actually receives the set aside and not when they applied for the set aside, which is typically about six months earlier. And so if it were all about regression to the mean or about the motivation that they had at the time that they applied, we would think we might start seeing some of those positive effects kick in earlier. And then we also looked at a subset of our sample that were people that applied for the set asides immediately when they became eligible for the set aside, that is after five years, within the first quarter of a year after becoming eligible. And those people we think of as more like a random sample because it accounts for a Large share of the people applying for set asides, way more than would kind of randomly apply that apply in subsequent quarters. And so we think that there's a group of people who they really want set aside. So they are very motivated. But the motivation, the thing that's triggering them at that time is not that they're in the midst of a job search or that they just lost a job or something. It's just they just passed that eligibility threshold. Right. So they may not be any more motivated then than they were three months or six months or a year earlier. Right. So when we look at that sample, we see like when we looked at our larger sample, we basically see a declining employment trend that turns into an upward employment trend. Like a V shaped graph. Right. And that kind of graph gives you some sense of worry that the turnaround may be just bouncing back from something bad that happened earlier. But when we look at those early appliers, we see a flat trend leading up to the point when they applied and then a steeper upward trend. And the total change is similar to the other group. And so, so that group we feel a little stronger making causal claims about, but we still have to be cautious. I think the external validity concern is more serious because even if we are able to draw causal conclusions about the group that we studied, they still are the group of people who jump through the hoops. They are people who believed expungement would help them. And it's not very surprising that people who believed expungement would help them strongly enough to go through the process of doing it were in fact helped by expungement. But if we think about the larger potential sample that automated expungement applies to, which includes some people, you know, probably many people for whom, if you ask them, will expungement help you? They would say yes, but they aren't the people who went out of their way to try to get it. I think we should expect some positive effect. Like, you know, there's tons of research saying that having a criminal record is a barrier in your life, so removing it should have some positive effects, but we shouldn't expect them to be as large as the effects that we found for the self selected group.
And that is my prediction. We haven't done any of the analysis of automated expungement yet, which we've, we've gotten another grant to research and we're, we're just in the data acquisition phase for that right now.
But if you ask me ex ante, what do you predict? I would predict that we will see some positive effect, but it won't be as big as what we found earlier.
[00:57:04] Speaker A: Just to comment on the internal versus the external explanation that you just provided, Sonia. The internal was one of the things we're attempting to solve with a randomized study.
[00:57:15] Speaker B: Exactly.
[00:57:17] Speaker A: If we are successful at pulling off the randomized operation, then we may be able to shed some more light solving that problem. It will not solve the external problem, you know, with a, with a, with an, with a use. When you're using a petition based regime or when you're analyzing a petition based regime, you're stuck with that one.
Let me just say that's all the time we have. Y' all have been extremely generous with your time, both of you. So any. Jj, start with you and then Sonia, final thoughts on this. I mean, one idea is expand the remedy, make it automatic. That comes with assumptions about the external issue that you just identified. Sonia, along with others, sometime I want to have you back and talk about the ban the box analogy which you talk about in the paper. I'm actually, I think I'm more scared of that than y' all are. From the sound of your, of your law review article.
I'm actually personally terrified of that when, along with the potential concerns about not danger to the public, not from recidivism in the area, but from reductions in accountability on official entities. So, for example, how much harder it would be to hold police officers or police departments as a whole accountable when official information about what they do starts to disappear or starts to become very difficult. These are, I think, broader debates, difficult to obtain. I mean, I think broader debates. And so I'd love to have them sometime. But let me give you all the last word and along with my advanced. Thanks, jj. Last word from you.
[00:59:04] Speaker C: Yeah, I'm going to spend a little bit of time talking about bluebird monitoring.
I think that notice is huge now as we move to a world in which sort of recognizing that, you know, if we're serious about offering people access to certain kinds of remedies that, and it's the kind of thing that we can figure out on our own. By we, I mean the government, you know, just, just making the decision to, to grant them is, is going to make, you know, we're going to move. As you, as you point out, 90% of people who wouldn't have otherwise gotten one will now have one. And the question is, how do we communicate that to them and how do we allow them to kind of maximize a potential upside of that from an employment perspective and housing and the other sorts of advantages to it. And I think that's an ongoing area of research as well. So conditional on having an automatic system in place that is going to be essentially reviewing records over time and deciding that at some point they ought to be removed from the public eye because they're no longer useful, maybe even distracting and counter, counterproductive.
How do we work with people who have those records to make sure they know about it? How do we allow people to maybe find their own records online? So coming up with ways, and this is where identification validation really does matter.
If you're going to allow people.
[01:00:30] Speaker A: To.
[01:00:31] Speaker C: Learn about their own records so they can find out where they are, you have to make sure that people who they don't want to learn about their records can't use the same technology.
But, but notification and figuring out how to reach people, especially this population, justice impacted individuals is a real challenge. And so I think that's the next frontier. Assuming we think this is a good thing to do, how do we actually implement it in the future so that we don't have something similar to the uptake problem, the lack of awareness problem that we might otherwise have.
[01:01:03] Speaker A: Super Sonia, last word from you.
[01:01:06] Speaker B: So I want to raise a different type of caveat basically, which is that virtually all expungement systems have substantial waiting periods, including both automated and petition based expungement systems, but especially automated.
And even though we found quite positive results, as I said, those are probably partly because of self selection. And there is some research coming out lately from other people about things that are not expungement but are other ways of mitigating the harm of a criminal record.
Things like the Fair Credit Reporting act or California has moved some felonies down to misdemeanors. So various forms of record relief. And there have been some pretty discouraging studies coming out recently with showing limited or no effect on people's subsequent employment hours outcomes. So this is like Amanda Agan and co authors and a couple of the pieces for example.
And I hope that we will find better results when we look at expungement, which is a more sweeping form of relief, even if expungement is automated.
But I do worry that like some of these studies raise, I think very plausibly the argument that once, once somebody is past the waiting period at which some of these things kick in in some ways that they're past the critical point at which they most needed help.
And so I think it's really important to that, you know, if assuming that waiting periods remain in these criminal.
In all of these criminal record clearing proposals, that, which I, which I expect is likely to remain the case that we really think as a society about ways to help people get on their feet right after they're convicted or right after they're released from incarceration, which is when they're facing tremendously high recidivism rates and just a huge risk of a revolving door going back into prison. So that might be more things like job placement programs, better transitional housing programs.
So record clearing, it is part of the picture of how to enable successful reentry, especially in the long term. But it can't be a replacement for thinking about how to strategically help people right away. And helping them is also helping society. Right. Because that is when they're at the public safety risk is highest. And so we should think about it as a, as society's problem to figure out what to do in that in those critical years.
[01:04:00] Speaker A: Terrific. We'll leave it there with tremendous thanks to both of you for the generosity of your time, for the excellent work that you did, and I hope we'll have you back sometime to talk about some of the bigger picture policy issues as well as the continuing research that both of y' all are pursuing. You mentioned some of it in this area in the future, so thank you so much.
[01:04:20] Speaker B: Thanks for having us.
[01:04:20] Speaker C: Thanks very much.
[01:04:22] Speaker A: Proof Over Precedent is a production of the Access to Justice Lab at Harvard Law School.
Views expressed in student podcasts are not necessarily those of the A J Lab.
Thanks for listening. If we piqued your interest, please subscribe wherever you get your podcasts. Even better, leave us a rating or share an episode with a friend or on social media.
Here's a sneak preview of what we'll bring you next next week. But I have three folks with me today talking about expungement in Kansas and in particular how difficult it is to achieve record clearing in Kansas without a lawyer.
