SIO313: US Police Killings Undercounted by More than Half, According to New Study

Transcript

This transcript was generated automatically and may contain errors — the audio is the record.

0:00:12 Thomas: And welcome to Serious Inquiries Only. This is episode 313. I'm Thomas. That's Dr. Lindsey Osterman. How you doing?

0:00:18 Lindsey: Oh, I'm wonderful. How are you?

0:00:20 Thomas: I'm doing fine myself. Huh, On one hand, feeling like we were very vindicated in our past episodes on police violence, but also a super depressing vindication, so I'm not... But either way, I mean, if you went back, if you listened to, at the time, our series on police killings, police violence, so I think you really should, police bias too, as well as all those things, um, did a fantastic job of breaking down some of the science involved in that.

And throughout that whole time, I believe I said multiple times that, like, also keep in mind, no matter when, what we're talking about when it comes to police violence, that so much of the stats that we can get access to, or that anyone can who's running studies and stuff, so much, much of that is police reported, often. And, and so much of even the news reporting will be like a reporter talked to the police who told them something. And it's like, we have, should have learned, unless you're somebody coming from a totally different political perspective, we should have learned by now that that is just not reliable. And the big example was George Floyd, you know, that would have just been, oh, medical, medical related death while

0:00:28 Lindsey: Mm hmm. Yeah. Oh, yeah. Yep.

0:01:33 Thomas: Well, yeah, a guy had a heart attack. You know, that was how that would have been coded, reported, and maybe still was after we go through this article in the study. But anyway, this you may have seen in The New York Times what we're going to talk about today, which is at least that's where I saw it. A new study says that more than half of police killings are mislabeled and the number of police killlings is basically double what you would have thought it was based on how it's been reported. Which, yeah, I mean, that's kind of what we were saying. And

0:01:34 Lindsey: History of drug use and heart condition, right? Uh huh.

0:02:07 Lindsey: Yeah,

0:02:07 Thomas: Whenever there's a study involved in something, that'S when we can have science person, best science person in the world, Lindsay Osterman, to explain it to us.

0:02:15 Lindsey: That's so kind. Yeah, no, this was this was a great article, this this New York Times article. And, now that I've read the original study, and I went back and read the New York Times article again, like, there's really not many things to gripe about in terms of the the writing here. It was it was quite well done. You know, in addition to summarizing this the findings in this study, they're also sort of highlighting a number of these these anecdotes, right, in which there was clearly a death that was caused by police-related violence and then it was misreported on the death certificate. So one of the things they talk about was Elijah Mclean, who was put in a chokehold, a choke hold that is now banned by most police departments, so chokehold by police, and then injected with ketamine. And in the death certificate, the cause of death was, quote, "undetermined."

0:02:28 Thomas: Nice. Mm-hmm. Mm-hmm, yeah.

0:03:00 Lindsey: Yeah, I have a few guesses.

0:03:00 Thomas: Hard to tell. No way to know.

0:03:01 Lindsey: Yeah. I know. Totally hard to get. Yeah. Yeah. Right. So I think I'm glad that they brought up those specific cases to illustrate, because I think it triggers the right intuition about why what these researchers did in this current study really matters. It matters that these deaths are being misclassified, because the cause death on the death certificate really matters in terms of whether charges are brought, right? I mean, you know, taking the George Floyd trial, for instance, I'm sure that folks who followed that remember that a lot of the defense's case leaned heavily on the medical examiner's judgment of that history of drug use and the pre -existing heart condition being contributors, right? So this matters, right.

0:03:01 Thomas: Yeah.

0:03:40 Thomas: There's some other tidbits from the article. And apologies if, you know, I don't want to scoop what you're going to talk about. Out, but some things that caught my eye just preliminarily from the article here, the killings have just gone up, like up and up and up when crime has gone down. Now, obviously 2020 was, you know, a very special year. Crime did murder, went up, homicide, all that, for, you reasons that I think we'll, you'll find we're involved due to COVID and other things going on. But that's an outlier year. For the most part, crime is at all time lows. I think it's still at all -time lows, even relative to the 90s and stuff, the peak in like the early 90s or late 80s.

0:04:01 Lindsey: Yeah,

0:04:18 Thomas: And even still, police killings going up. And I believe it said the discrepancy of the killings is also increasing. You know, like black Americans, 3.5 times as likely to be killed. So this is something that as much as the knee-jerk conservative reaction is always, well, don't be a criminal. It's crime. It is like, well, okay, crime has done nothing but go down until last year when it went down dramatically and police killings are going up. There is a problem here.

0:04:17 Lindsey: There is definitely a problem, and the results from the study are pretty compelling in that regard. So, yeah. Do a breakdown of this very fascinating

0:04:53 Thomas: Let's get into it. I am really excited to learn more about it from

0:04:55 Lindsey: Study? Yeah, yeah, no, this is really cool stuff. As you say, depressing stuff, but the methods that they're employing here I think are really interesting. So, study, right, this was published in The Lancet, which is a well-regarded major medical journal published just this year, 2021. the authors of this article had three major goals here. So the first was to examine the extent of underreporting of police violence in this particular national database of vital statistics. And vital statistics are things like, you know, records of births and marriages and divorces and deaths, obviously. So there's this this national database called the NVSS, or the National Vital Statistics System, and the NVSS maintains this publicly available data set on all these vital statistics. And anybody can go right now and generate reports on deaths according to demographics or whatever. I've actually used this data set a lot for some of the research that I've done. So it's a trusted data source.

It's assumed to be a reliable and accurate record of not just the number of deaths that that occur in various regions in the U.S., but also the demographics of people who are dying and their causes of death. And the authors note in their supplementary materials for this article that the NBSS really does have like virtually 100% completeness in terms of recording all deaths,

0:05:27 Thomas: Really

0:06:15 Lindsey: But there is this well-documented issue with civilian deaths due to police violence being misclassified in the database, which amounts to essentially underreporting of this specific cause of death, right? So one interesting thing here is how these deaths get misclassified. So the cases in this NVSS database originate from death certificates. So someone from the state or local jurisdiction, this is usually a medical examiner or coroner, fills out the death certificate, which includes information like sex and race and ethnicity and cause of Death.

0:06:56 Lindsey: And they input it into the system. And so cause of death is categorized according to these things called ICD codes, or I think it's International Classification of Death Codes. And that's the classification system used by the WHO. This classification system is pretty elaborate. There are, you know, codes for deaths by suicide, and then subcodes for different methods. There are codes for death by infectious disease, and then subcodes for different, you know, types of diseases. And there's also one set of codes that corresponds to deaths due to legal interventions, which just means deaths due to injuries inflicted by police or other law enforcement agents in the course of arrests or suppressing disturbances or to maintain order, and so on. So if a civilian died because of a gunshot wound from another civilian, it would be classified as a homicide.

0:07:47 Lindsey: But if someone died from a gunshot wound in the context of a legal intervention, it would be classified in this separate category for legal interventions, right? So there's a couple of things about this process that introduce significant risk of error and could lead to misclassifying a death from a legal intervention as a death by homicide or of something else, right. So first of all, depending on the jurisdiction, the person who is filling out the information on the birth certificates could be a physician trained in forensics, or it could be a physician with no forensics training, or it could be the sheriff, Or yeah, or it could be a mortician.

So the authors of this current article point out that this is actually more common than you might think, especially in smaller and rural counties that don't have the resources to employ a full-time medical professional with this forensics background and training. And this is significant because it means that there is wide variability in the medical expertise and the training and the personal interests of the people who are filling out out these death certificates. And so, you know, if they aren't trained well, they might not know that it's important to include, you know, in the cause of death, whether it happened during a legal intervention, or the person reporting this may be the sheriff and not want to report that

0:08:22 Thomas: Right.

0:09:01 Lindsey: This was, you know, a death that was caused by one of their officers, right? So that's the first potential source of by or source of error, right, is that there's this wide variability in who's making the reports and writing this stuff down. The second major source of error here potentially is that these fields that they're filling out in the death certificate, they're just these like free-response text fields. So for like cause of death, for instance, it just has a blank box that says cause of death. And the person filling out the death certificate is just writing something in that blank, like gunshot wound to the chest, homicide. And it's those responses that get coded in the NVSS database by, you know, independent people who work to maintain the NVSS database. So in my mind, this creates a lot of room for misclassification, because the accuracy of the classification depends on how accurate and how detailed the information in that field is.

0:09:00 Thomas: I-

0:09:56 Lindsey: So if the coroner reports that the cause of death was a gunshot wound, but just neglects to report that it was a police officer that fired the gun, the death will end up being misclassified as a homicide that didn't take place in the context of legal intervention. And then when people run statistics on fatal police violence using the NVSS database, that case ends up not being counted. And apparently this happens a lot, and it's a well-known issue.

Think it's important to make clear, just right at the top, that the system by which deaths are being classified here is creating exactly the kind of ambiguity where subtle bias kind of thrives, right? If the field makes it ambiguous as to which information is relevant to put in, then it's really easy to leave out critical details. Like if they, instead of a blank field, if they had like a series of check boxes that corresponded to the details that get coded later, like these intent categories, which breaks up suicides and homicides and legal interventions and so on, if they have those check boxes, it would be much more difficult for things like the involvement of an officer to get left out. And so, you know, the New York Times article interviewed a former medical examiner, and he said that he'd been calling for years for a change to the reporting system for cause of death and moving, you know, to the sort of checkbox system because it leaves far less room for error, right?

0:11:16 Thomas: I could imagine that being an awful lot of check boxes though, I mean, there's a lot of different ways to die. Yeah.

0:11:22 Lindsey: Potentially. That's, that's potentially true. I was thinking about that. I mean, if you if you were doing this in an electronic system, you could just have sort of like drop-down menus that kind of branch in structure. I mean, you could set it up in a way that would be efficient, I think. Yeah. Yeah. So all of that is a relevant background for explaining that first goal of the article, which was to estimate the extent of misclassification of deaths that are due to police violence. The second goal of the study was to offer a statistical method for correcting for the underreporting in these these official data sets. And then the third goal was that based on this method that they're proposing, they also present revised estimates of how many people in the US actually died between 1980 and 2019 as a result of police violence. And that's where that headline came from, that they're being undercounted by about half, right?

0:12:13 Lindsey: They start their article by talking about like a few facts that just sort of communicate the gravity and scope of the problem with police violence in the U.S. This is not going to come as a surprise to anyone who listens to this show or pays half attention to the news that, yes, no, yeah. But yeah, police violence is a huge problem in the US, and it's a problem that disproportionately harms Black Americans, right? They discuss a 2019 report from something called the Global Burden of Diseases, Injuries, and Risk Factors Study, or GBD for short. And that's this large -scale study of mortality that's been going on for about two decades now. And they release reports annually on estimates of mortality from various causes across the globe, right? So in the GBD's 2019 report, they indicated that police conflict and executions accounted for nearly 300 ,000 deaths globally between 1980 and 2019. And of those global deaths, the United States accounted for for

0:12:24 Thomas: I- would hope not. We're-

0:13:14 Lindsey: 13.2% of them, despite only accounting for 4% of the global population. That's, yep.

0:13:20 Thomas: Number one, we're number one, baby. American

0:13:23 Lindsey: That's I got that in my notes, actually. Yeah. Number one, America. Yeah, yeah. So that illustrates Yeah, just how well we're doing on this issue. And, and we've talked about this in past episodes. There's lots of past research indicating that as a proportion of the population, black men are disproportionately more likely to be killed by police than white men. At something like 2.5 times the rate for black men versus white men. And black women also are about 1.4 times more likely to be killed than white women, right? And so the The authors note, and again, we've done several episodes on this, that this is the result, or that this is influenced by contributions of both systemic racism, right, and direct racism, right? One thing that these authors point out is that in this 2019 GBD report, report. The estimates of deaths due to police violence were calculated differently than they were in past years. And the specific thing that they did differently was to correct for this systemic underreporting of deaths.

So that was the first year that the GBD used this correction, but it will be applied to all future iterations of the GBD because this is such a well-known issue. And this current study that we're talking about is using a statistical strategy that resembles one that the GBD employed in order to correct for this underreporting bias, right?

0:13:24 Thomas: Exceptionalism at work. Yeah. Wow.

0:13:54 Thomas: Oh.

0:14:40 Lindsey: So here's what I was particularly interested in, or here's what I was particularly interested in knowing when I first saw that New York Times headline, right. Was curious about what alternative data sources for police killings they were using to compare to the NVSS in order to get a sense of the underreporting rate, right, because the utility of this kind of analysis kind lives and dies on the quality of their alternative data. And I really wasn't sure what that would look like. But as it turns out, there are a number of non -governmental databases that track deaths due to police violence. And so the three that they used in their study were called Fatal Encounters, Mapping Police Violence, and both of these are maintained by nonprofits. And then the third one was called The Counted, which is a project that is maintained by The Guardian.

0:15:27 Lindsey: All three of these databases are open source, meaning that they're built from publicly available information. So namely like news reports and public record requests. And I read the about pages for each of these websites for these databases in which they describe, you know, how they are building and maintaining these databases just to get a sense of their rigor. And actually found them pretty impressive. So yeah, yes,

0:15:48 Thomas: Yeah, totally. I mean, this is a, this is a major thing that, that least one of those websites you're talking about in the past. I also know there was a, I think there was one that was like Washington Post related, or there was like a newspaper affiliated one they've also used before.

0:16:00 Lindsey: Yes. And I guess I was looking to to see, like, you know, are they doing anything to sort of verify or estimate, like how complete these data sets are, right, and how accurate they are, and they are doing that work. So like, yeah, so all three of these databases appear to be using similar methods to populate their cases. And yes, they are largely sort of pulling cases from news reports. And then, as you just mentioned, they also accept tips from the public. So somebody can write in And report either a new case that's not in their set, or they can provide additional details for a case that is in their setup, but has some information lacking about it. And so the cases they get from the news reports or tips from the public are then vetted by paid researchers who verify the details by looking at reports from multiple news outlets and things.

And then they all record information about the demographics of the person who was killed and the method by which they were killed, and they store it in this database. But the piece that I found really persuasive in terms of this being high quality, or verified as high quality, is that Fatal Encounters has been doing an enormous amount of work to estimate the accuracy and completeness of a data set that's generated with this strategy of using the news reports and public tips.

0:16:26 Thomas: Wow.

0:17:12 Lindsey: So they are verifying their cases one state at a time through public records requests for death certificates, and then checking the entries in their database, which is very expensive and very time consuming. But it's also excellent in terms of verifying, you know, the completeness of their sets here. So like, if they're auditing, you know, their Arkansas cases or whatever, they would have to pay a processing fee for each death certificate they want to obtain. And then someone has to verify each entry against the death certificate. So that's that's a lot of work. I mean, yeah. So but after evaluating the records from 11 states so far, Fatal Encounters has verified a 97% completeness rate of the the original data set that was constructed from the news reports and the tips from the public. That's pretty good.

0:17:59 Lindsey: That's an understatement, actually. That is really awesome. Yeah, so the point is these folks are doing some really careful and impressive work to build a more complete sort of picture of police killings than the NVSS is providing, right? Together, so far, these three databases alone, and, you know, they are not the only ones out out there, have have 20 years of data on police killings from the year 2000 to 2019. Right. They're not the only non governmental open source databases that are tracking police violence, but they are the only three that these authors found that included some critical details that they wanted to analyze.

0:18:32 Lindsey: So These three in particular include both firearm and non firearm related deaths. Some databases only have information on police shootings. Yeah, and they also include information regarding the state in which the death took place and the race and ethnicity of the person who was killed. So those were the most of the critical variables that they wanted to look at in in their analysis. Right, so those are their alternative databases. And what these authors are going to do is compare the information from these databases to the data in the NBSS in order to estimate the degree of underreporting in the official government statistics.

0:18:32 Thomas: OK, so We already know the punch line, but take us through it, I guess.

0:19:12 Lindsey: Yeah, I think it's I think it's worth getting into the details here. There's some there's some really interesting stuff. So so here's here's what they did. So the so for each of these four databases, one and these three open source ones, they extracted the following information about each death in the set, right? So they got age and sex of the deceased individual, state in which the death occurred, the year of the death, and then race and Hispanic ethnicity. All of this is, you know, about the deceased individual. They then obtained estimates of death rates as a function of these demographic variables across these different datasets. And this allowed them to estimate the gap between these national vital statistics and these higher fidelity open source datasets for the years in which they had both the national vital statistics information and the open source data.

0:20:04 Lindsey: But it also allowed them to make predictions about the true rates of police killings in years for which they don't have the open-source data yet, right? So again, like the NVSS database has cases for 39 years from 1980 to 2019. But collectively, the open source data sets only have information about 20 years out of that 39-year time period. And so for the missing years, they're essentially imputing predicted rates of deaths due to police violence based on the relationships that they're observing for the 20 years that they do have both the national vital statistics and the open-source data. Um, does that make sense, or does that sound like magic stats hand waving?

0:20:43 Thomas: No, that makes sense. I much of that did they actually verify? Did they have to verify the whole thing or do they do some statistical sample that tells them all that?

0:20:43 Lindsey: Okay.

0:20:55 Lindsey: Okay, so they've got, they've got actual information on 20 out of those 39 years, right? And they're the the those should should be complete reports in terms of the open source things. Um, so for the, for the remaining 19 years, they, they don't have the open-source data, and that's where they're doing sort of the statistical prediction to get those values. Um. Yeah, so that's, that's all clear. I had like a, a bunch of elaboration on that, but it might not be necessary.

0:21:00 Thomas: OK. OK. Gotcha. Gotcha. Well, I mean, it's I think it's probably the same principle as or guess I'll ask the question, I means the same basic principles of like polling, you know, where it's like you you ask a survey, 2000 people or whatever. And somehow that provides you with reasonable confidence of what the whole country thinks or whatever, is it that kind of statistical?

0:21:41 Lindsey: Yeah, basically.

0:21:41 Thomas: Statistical, yeah, yeah.

0:21:42 Lindsey: I mean, it's a, it's little bit more complicated in terms of the stats that they're doing because they're, you know, they're using relationships between the death rate um, like different states and in these different uh, sort of social categorizations regarding race and ethnicity, they're using relationships between the death rate and all of those different types of things in the observed years, and then they're, they're creating like a, a prediction for the rates in the years that they don't have observed data for, like, based on the the relationships that they did observe. I feel like I'm being a little recursive there, but hopefully that makes sense. I don't know.

0:22:13 Thomas: Well. As a statsy person, tell us what you think the accuracy of this is? I remember what, or what I guess, what the math tells you the accuracy if it is.

0:22:25 Lindsey: Yeah, no, this is definitely, this works. This is not magic at all, right? Like, so more commonly, no.

0:22:32 Thomas: Yeah. It's not Nate Silver just getting the whole election wrong kind of thing. Why?

0:22:39 Lindsey: No, right. No. If you have a bunch of variables that you know are related to the thing that you want to predict, you actually can predict, predicts, you know, pretty accurately within some margin of error, what, you know, the estimated case counts are actually going to be in those missing years that you're looking at. It's the most common place that this happens in like psych research, it's called imputation, data imputation.

It is like, you know, say we're doing a survey and somebody fills out like a 10 item measure of self-esteem or something and they forget to ask to answer one of the questions. You can reasonably impute what their answer to that 10th question is based on their answer to the other nine items because self-esteem items tend to be highly correlated with each other, right? Yeah, this is, this is not uh, this is not an unusual thing to do, this is As far as I know, right, the way that they did this looks totally legit um, so I think, yeah, can consider this reliable. But yeah, so they estimated both um, rates of death due to police violence and they estimated the degree of underreporting as a function of gender and race and ethnicity. So, for example, they examined whether Black Americans are killed by police at higher rates than white Americans, and they also examined whether the NBSS is underreporting police killings differentially for Black versus white Americans. So what did they find?

0:24:00 Lindsey: I guess, yeah, we gave the game away way a little bit, but so here's what they found. I'm going to do a little bit of quoting here. So here's the, the, the headline finding top line. So quote, the misclassification of police violence in the NVSS data is extensive from 1980 to 2018. The NVSS did not report 17,100 deaths out of the 30,800 deaths that we we estimated, accounting for 55.5% of all police violence deaths from 1980 to 2018.

0:24:02 Thomas: Don't you tell me what I'll ask it, you tell me. What did they find, so It was 2018, okay.

0:24:39 Lindsey: Well, they had 2019, but they were doing something kind of different with the 2019 results. They kept breaking those out separately to sort of give you a sense of what the rates looked like for that year.

0:24:43 Thomas: Okay. Gotcha.

0:24:49 Lindsey: But yeah, they did have 2019 results, so yeah, so more than half, 56 % of deaths due to of police violence not classified as such in the official national statistics.

0:24:49 Thomas: Yeah. That's Wow. That's a failing grade on a test. Well, unless you're taking the bar exam on opening arguments, that's about what I would get. They should do better. Medical profession, whoever's in charge, coroners, whatever it is, they should have to do better than me on the bar exam on Opening Arguments. That should, that is my new policy.

0:25:07 Lindsey: Mm -hmm.

0:25:17 Lindsey: Totally agreed. If they had some checkboxes, they, I mean, I know that seems like a stupid detail, but like that really is the kind of thing that creates less room for people to leave stuff out.

0:25:20 Thomas: Yeah, yeah. Oh, knowing human behavior. Yeah. If you've ever been in charge of any like processes within a company of like, Like, all right, when you take an invoice, you got to, you know, make sure to do the following thing. And it's like the that people mess this stuff up is staggering. It's very, the easier you can make it, the more streamlined, the More Accurate, you're going to get the data.

0:25:41 Lindsey: Mm -hm. Yeah, yeah, exactly. Right. And it removes like sort of plausible deniability, right? If there really is some sort of explicit bias that's operating here, it's a lot harder to sort of justify leaving out certain details to yourself if you know it's very clear in the question that you need to answer whether an officer was involved or whatever it is, so

0:26:03 Thomas: Yeah. One of the top sayings of any accounting, whether it's accounting or like government or whatever, in my position, it's always been garbage in, garbage out. They always say garbage in garbage out, you can't, if there's, if they're messing up the the data on the way and there's really not a whole lot you can do to fix it short of doing this massive study comparing an actual good data source and but you need that, you need the reliable numbers somewhere to even be able to do that. If we didn't have, if they had no second data source, we would never know this, right? If it were, if it weren't for these other non-governmental entities keeping track of this, we would never know.

0:26:03 Lindsey: Yep. Yeah. Yeah. That's, that, that is right, that's why I was so curious about where they were getting all these numbers. Um, all right, so that was, that was sort of the the top line finding. They also had some some additional interesting findings, uh, sort of broken up by demographics. Um, so here's another of the quote. "Total underreporting is also disparate by race and ethnicity." No surprise there.

0:26:56 Thomas: That's Interesting, huh?

0:26:57 Lindsey: Yeah. So continuing from 1980 to 2018, the greatest underreporting of deaths was among non-Hispanic black people with 5,670 deaths missing out of an estimated 9,540 deaths, which is 59.5% misclassified. That's, that's higher than that, that average there, right? Right. For for non-Hispanic white people, 56 percent were misclassified for Hispanic people of all races. Fifty percent were misclassified. So there's... there's lots of misclassification happening across the board. But it is interesting that you you get this this higher rate of mis classification for deaths involving involving black people. So, thing that I thought was interesting is that under reporting rates also varied by state. So the top five states with the most severe underreporting were, in order, Oklahoma with 84 % underreporting.

0:27:06 Thomas: Yeah. Yeah. Wow. Well.

0:27:52 Lindsey: 84%, yeah.

0:27:53 Thomas: I was, my guess was gonna be smaller states, just because, you know, like it sounds like maybe, maybe in cities where you have you more employees handling this kind of thing, um versus like the small town sheriff, whatever, doing, not giving a shit about filling out the paperwork correctly, that was, is going to be my hypothesis. I don't know if that'll hold up here, but Oklahoma so far feels like a good hit on that.

0:28:03 Lindsey: Mm-hmm. Yeah. Yep, I think I got a few more hits for you. You. So also, so the next one is Wyoming with 79%, Alabama with 77%, Louisiana, 76%, and Nebraska, 73%. A lot of rural area out there, I think. So yeah.

0:28:17 Thomas: Hmm. Yeah. Yeah, Louisiana is a little bit interesting, but I like I don'T think of it as quite the same as like Nebraska or Wyoming, right? But weird.

0:28:34 Lindsey: Yeah. No, I mean, it's Louisiana, but but you still, you've got, you're welcome.

0:28:41 Thomas: Yeah thank you for clarifying that it's not the same state as Nebraska or Wyoming. I always get that mixed up.

0:28:48 Lindsey: Yeah, me too. Yeah no, but I mean but you know, There's still, there's a lot of rural areas in Louisiana. I know they've got, you know, a few big cities, but still, this is, it was not surprising to me to see it on that list. So those are our top five states. Five states with the least amount of underreporting were Maryland with 16%, Utah with 20%, New Mexico, 26%, Massachusetts, 33%, and Oregon with 36%.

0:29:16 Thomas: That's interesting.

0:29:17 Lindsey: Yeah, I thought so too.

0:29:17 Thomas: Wow. 16%. That's almost like having real data. You're, you're approaching accurately reporting police killings. What a miracle. Still way off, but

0:29:23 Lindsey: Yeah. Yeah, I know. I was proud of them too. That's great. Overall they estimate that there have been 32,000 deaths from police violence from 1980 to 2019. And their estimates indicate a 38% increase in the rate of deaths due to police violence from the 1980s to the 2010s, right? So an increase from 0.25 per 100,000 to 0.34 per 100,000, which is interesting. So not surprisingly, they also found that deaths due to police violence are disproportionately higher for men than for women.

0:29:29 Thomas: Yeah. Yeah.

0:30:04 Lindsey: So they estimated 30,600 deaths for men and 1,420 deaths for women, so that's more than a 2,000 percent difference in mortality between men and women, right? Right. And there are also disparities by race, which is, you know, this is a reprise on stuff that we've talked about a lot. So for non-Hispanic black people, the rate was 0.69 per 100,000 versus 0.2 for non-Hispanic white people. So that's three to three point five times higher for Hispanic people of any race. The rate was point three five per hundred thousand versus, again, point two for non-Hispanic white people, and the rate of non-Hispanic indigenous people was 0.38 per hundred thousand, which was 1.8 times higher than that of non-Hispanic white Americans. So definitely seeing racial disparities here that we've seen in past studies. Thing I thought was interesting here is that they looked at trends over time as a function of these different racial and ethnic categories. And I encourage people to actually look at the article. We'll link it. It's very hard to podcast graphs, I've noticed.

0:31:09 Thomas: Yeah, that is one limitation. Yeah?

0:31:12 Lindsey: Yeah. But it's an interesting graph. There's a number of things about it that are really interesting and, obviously, from the 1980s to the 1990s, it looks like rates of deaths due to police violence actually declined significantly. And this was particularly pronounced for non-Hispanic Black people and Hispanic people of all races, right? Then from the early aughts to the 2010s, those rates appear to be increasing again. And that's true across races. But it's also appears to be particularly true for racial minorities, right. Other marked thing that just just jumps out at you immediately from this graph is that the of killing of non-Hispanic Black people in the U.S. is just like orders of magnitude higher than all of the other racial categories.

0:32:00 Lindsey: It's like there's some clustering toward the bottom of the graph for all the other racial categories that they looked at, racial and ethnic categories they looked at. But consistently, there's like a very large gap there in terms of Black people being killed at much higher rates. So that really does leap out at You. You. So the take home here is that across this 40 year period, rate of police killings are disproportionately high for black people. We knew this compared to other racial groups by a very wide margin. And police killings do seem to be trending upwards since the 90s into the 2010s. And this trend is most pronounced for black people. And that's according to these estimates, right?

0:32:38 Thomas: As the crime rate was dropping dramatically during that same time, right? Yeah.

0:32:42 Lindsey: Yep, we don't have a problem, it's fine. Yeah. Yeah, all right, so so those are the those are the main findings, which I I did find very compelling. So, um, you know, sort of takeaways here, this National Vital Statistics, uh, data set, right, misclassified and underreported again more than half, more than half of the estimated deaths from police violence in this this 40 year period. And the underreporting was not uniform. There was significant underreputing across, you know, all racial and ethnic groups, but the highest underreporting occurred for black people. Right.

0:32:45 Thomas: Yeah.

0:33:04 Thomas: Doesn't Really seems like that's something that explicit bias might explain, where people or police or whoever, you know, they are more apt to try to cover up a killing of a black man, essentially, is that, that makes sense? Yeah.

0:33:32 Lindsey: That that's a reasonable hypothesis. I mean, it's, you know, I think there's a lot of individual motives that could be going into that. And again, you know, because of the stupid reporting thing, it's it's probably very easy for people to sort of explain away why they're reporting things in the way that they are. Yeah, I mean, of all the things in this set, right, that it seems like explicit bias is probably a good candidate to explain that one.

0:33:57 Thomas: Or I guess if not even bias, I suppose it could just, again, it could be like more wanting to play down that type of police killing, right? Which I don't know, I guess you could call that bias, but just to clarify, like I still systemic racism, it's just not like you, the process by by which you're seeing it as a little different than like an explicit racial bias kind of thing.

0:34:05 Lindsey: Yeah. Mm hmm. Yeah. I think I think that's a good point. Right. Yeah. Could definitely see somebody saying to themselves, like, this is reported this way, this is going to be this is going to racialized, right? And that playing into their their desire to downplay it or whatever. But I don't know, it's it's impossible to tell. Yeah, so we have a we have have a problem with police violence in this country, and it is disproportionately hurting black American men in particular, in disproportionate way, right? So yeah, a solid study. They did talk about a couple of limitations that I think are worth spending a minute on. So one thing that they said, you know, would be important as an avenue for future research is to fill this gap in the current study, the fact that they didn't do any analysis of non -fatal injuries?

0:34:29 Thomas: Right.

0:35:03 Thomas: Yeah. Yeah. I was going to bring that up.

0:35:06 Lindsey: Yeah. Right. I mean, yeah. Deaths is not the full scope of the problem.

0:35:09 Thomas: Well, I imagine that's just such a huge data challenge, right?

0:35:13 Lindsey: Yes. Right? I mean there, there's a lot of things here that are, that are huge challenges, but, um, but would be very valuable, I think. So, so yeah, I, I would be very interested to see something like a replication of this that, that did include non, non -fatal injuries and stuff. Limit that they talked about was that they did not analyze analyze police officers killed by civilians or state executions. So these are sort of, they're in the linked category with the police killing civilians, right? Didn't look at civilians killing police or state executioners, and they say that it's because it was beyond the scope of their analysis, which it is, right, they're interested in the underestimation of of police killing civilians. But I would have been very interested.

0:35:56 Thomas: Yeah. To see how accurate that is.

0:35:58 Lindsey: Exactly. Right. Like, it seems like if you had

0:36:01 Thomas: Boy, one for one, a hundred percent. I bet that would be my prediction. Close to a hundred%.

0:36:03 Lindsey: My yes, No, absolutely. My prediction is absolutely that you're going to see far less underreporting for those other two categories of deaths that occur in the context of legal system. Right? Yeah, I would have been really interested in them using the same sort of statistical techniques, technique, to estimate that and then comparing the under reporting rates. But I'm guessing that nobody is maintaining an open source database of civilians killing police or state executions.

0:36:21 Thomas: Yeah. Yeah. Yeah.

0:36:32 Lindsey: So yeah, this.

0:36:33 Thomas: Well, maybe executions, but yeah, who knows, um, and and if they could have done that, it would have been a way to help rule out the incompetence excuse, you know, oh, weird you're able to fill out the form on this thing, hmm, I wonder why, but

0:36:36 Lindsey: Yeah. Yeah, no, I mean, it would definitely be another mark in favor of this, this, again, being a place where subtle bias can kind of operate and, and change responses. So no,

0:37:00 Thomas: I guess we don't have that data, so I can't, that's that's all just me guessing that that would be what would result from that, but I suppose we can't say for certain, maybe they would be just as incompetent at that data collection, I don't know,

0:37:08 Lindsey: We can't, you know, it's hard to tell. And there's other there's other, you now sort of categories of causes of death that that might show similar kinds of underreporting biases. And I'd be interested in all of those. These two in particular, I think would would be really interesting for comparison, just just to be able to compare this data to. So I'm glad that they stuck this in there, because it seems like, you know, these these nonprofits that are maintaining these open source database of the police killing civilians might be interested in sort of taking that up as well, just so that we have sort of comparison data, which would be interesting, right?

That I thought was mentioning here is that they noted an issue with the classification of like sex and gender. So death certificates in the US only report sex in a binary way, and they don't distinguish between sex assigned at birth and gender identity, and the open source databases apparently all contained, you know, some number of death records for which the gender was listed as transgender or non-binary, and so in their analysis they had to recode those cases to binary designations, and

0:37:30 Thomas: Yeah, yeah, yeah.

0:38:20 Thomas: Interesting.

0:38:20 Lindsey: Yeah, right, and it's probably just because it was such a low frequency, you know, that they wouldn't have been able to do a comparison there, but you know, they're, they admitted that like this does erase, you know, the non-cisgender individuals in the set, and it could potentially, it does obfuscate the higher rates of violence against transgender people and transgender black people in particular, right, yeah, exactly, so so they, you know, called for for more research to be done that that doesn't have that kind of problem associated with it. Right, so those are those are the main limitations that I thought were worth bringing up, are very important to disclose.

0:38:38 Thomas: Right. That Oh, right. Yeah, that would be something that it would be great to have data on to know.

0:38:53 Thomas: Yeah. Yeah yeah. Well, to good science, you know, we want to be clear about what we can and can't know. So I suppose, again, a very dark dismal subject. But in terms of the science here, this this gets your seal of approval as good science thingy. Right.

0:39:04 Lindsey: Yeah, exactly.

0:39:14 Lindsey: I as I can tell, yes. Right.

0:39:15 Thomas: A top notch science thingy?

0:39:17 Lindsey: Think I it's pretty, pretty compelling, pretty top notch and and very practically useful. Right? So I don't want to lose the fact that they weren't just trying to find out what was happening between 1980 and 2019. They were doing this in part to provide a statistical strategy for correcting the NVSS data, right? So when researchers are using this data, they could potentially apply this kind of correction in order to produce more valid

0:39:20 Thomas: Yeah. Yeah, mm hmm. Mm hmm. Yeah? Yeah? And for

0:39:51 Lindsey: Results, which is great.

0:39:51 Thomas: And for lay people, we can just be like, yeah, multiply by two. That's pretty easy.

0:39:54 Lindsey: Hey.

0:39:55 Thomas: Do that in my head. Twice as bad as whatever the the numbers said.

0:39:57 Lindsey: Yeah. It's true Yeah. Apparently, yeah.

0:40:02 Thomas: Right. Well thank you for taking us through all the science there, Lindsay. A very important follow-up to our police episodes. Again, if you haven't listened to those, go back in the feed. You'll find them. I think Lindsay did a fantastic job with them. And this is necessary follow-up to say, again, talk about limitations. You said, and you mentioned, I was going to mention this too. This still says nothing about police interactions. We still don't know.

0:40:27 Lindsey: Yeah.

0:40:27 Thomas: A lot of people on the conservative side like to take that number. Oh, what's that? Only 0.32 out of 100,000 or whatever the number you said. Like, oh, you don't even have to worry about that. It's like, okay, yeah, those are the ones who were killed. Imagine how often you're getting pulled over for driving while black. You're getting surged. You know, anything short ordered have killed. You're getting beaten. You know, like all those things are, you we've covered them other places as best we could. But in terms of the data, that's none of that is here. And so don't just minimize this problem down to the those who lost their lives. You know, it's such a bigger problem than that to keep in mind.

0:40:35 Lindsey: Right. Absolutely. Yes. I could not agree more. Oh,

0:41:08 Thomas: All right. Well, thanks for the breakdown, Lindsay. Thanks so much for listening, everybody. I hope you learned a lot. I know I did and keep those questions. Oh, I got to post the October. I think it's you know what? I already know, the October Patreon Q&A thread is going to be a spooktacular Q&A. I'm just calling it right now. So I'll post that spooktacular Patreon Q&A thread so that if you have any questions about this episode or any others, put it there, patrons, and we'll answer them.

0:41:18 Lindsey: Yeah. Oh, I can't wait

0:41:37 Thomas: It's not ghosts. Yeah, we can just answer. It's not ghost. And do that on patreon.com/seriouspod. Thanks so much for listening. We'll see you next time.

0:41:32 Lindsey: Ghost questions Very welcome. Yeah,

0:41:54 Clip: Thank you so much for listening to Serious Inquiries Only. If you like the show, the absolute best thing you can do is support us at patreon.com/seriouspod. The second best thing you can do, is please share it on social media or by word of mouth. Speaking of social media, follow me on Twitter at Serious Pod. If would like to be Facebook friends, send me a request at facebook.com Thomas podcast. If not up for that level of intimacy, that's fine. Maybe you want to join the discussion at facebook.com/groups/SIO members. Thank you again for listening. We'll see you next time.