NIJ-Funded Research on Firearms Violence in Urban Cities Advancing Scientific Evidence to Inform Practice

DARYL FOX: Good afternoon, everyone, and welcome to today’s webinar, “NIJ Funded Research on Firearms Violence in Urban Cities: Advancing Scientific Evidence to Inform.”  At this time, I’d like to introduce Basia Lopez, Social Science Analyst with the National Institute of Justice.

BARBARA (BASIA) LOPEZ: Good day, everyone. Thank you very much, Daryl, for the introduction. Yes, I’m a Social Science Analyst for the—for NIJ, NIJ’s Research and Development Agency of the United States Department of Justice. My primary research portfolio at NIJ focuses on firearm violence, and for the last few years, I’ve had the privilege to work with our renowned panelists that will speak to their research during this webinar. So they will present a series of papers summarizing the newest finding—findings of the research projects on criminal offenses with firearms in urban areas. The others will discuss issues concerning gun involvement among youth living in urban areas, the impact of demolishing or habilitating vacant and decaying buildings on firearm violence, and a relationship between the built environment and socioeconomic traits on firearm violence. 

The studies presented here were supported by the list here—Awards from the National Institute of Justice. Note that the opinions, findings, and conclusions or recommendations expressed here are those of the authors and don’t—do not necessarily reflect the positions or policies of U.S. DOJ.  So let me introduce our first speaker, Rachel Swaner. She is the Research Director at the Center for Court Innovation. Dr. Swaner’s recent work focuses on research with hard-to-reach population, exploring their experiences with interpersonal, social, and historical trauma, the relationship of these to violence and the criminal legal system, and methods of resistance, resilience, and healing. She’s also a Professor of Public Administration at New York University. Today, she will present her research findings of the NIJ-funded project titled “The Gun Epidemic Reconsidered: Creating a Foundation to Reduce Firearm Violence Among Urban Youth.”  Rachel, welcome. And the floor is yours.

RACHEL SWANER: Great. Thank you, Basia. And thank you for—NIJ for the—funding this project and for hosting this webinar today. So I’m going to talk to you a little bit about a study that we conducted in the last couple of years.  And our main goals are really just to understand why are young people carrying guns? So our research questions are why are they carrying guns, what are their networks like, and how are guns a part of these networks, because a lot of the research that involved youth gun-carrying had been done decades ago. There was limited qualitative research that had been done to really understand the complex confluence of factors that influence the decision to carry a gun. And there were—there were limit—methodological limitations and sample limitations of past studies, where they might be talking to people in schools, which overrepresents—or people who carry guns at this age often are not found in schools or disconnected from schools. Or they were talking to people who were already in detention which may overrepresent the criminal intent behind a lot of the thinking behind picking up or carrying a gun. So we wanted to talk to the young people themselves. So we interviewed 330 people 16 to 24 years old in three neighborhoods in New York City, neighborhoods that were chosen because of their high rates of gun violence compared to the city as a whole. The eligibility requirements were they had to be in this age group, they had to be from one of these three neighborhoods, and they had to have carried or owned a gun or been shot or shot at recently. So we use respondent-driven sampling which is a mean—a network means of recruiting a population where you recruit an initial couple of seed interviews. So we call them seed interviews. So they’re initial interviews. And then they help recruit other people in their network. 

So this little graphic next to—here is the three different colors, represents the three different neighborhoods. And the big circles are the seed interviews. So that’s where we have to expend most of our energy recruiting. And then all of the little dots after that are those people that they recruited. So each person was given three coupons to then bring other people from their network who were eligible for the study. So they were from these neighborhoods, met the age requirements, and met the eligibility requirements. Where did we recruit these people from at the start was, one, Cure Violence programs, which are health—public health programs to—they try to interrupt the transmission of violence, so they use credible messengers to go out in the community and try to stop retaliation, shootings, violence interrupters. So we worked with Cure Violence programs. Each of the neighborhoods we chose had a program like that, and they helped us do some recruiting. But a lot of it was also from just being out in the community in—NYCHA is our public housing authority in New York City. In all of these NYCHA public housing developments, there are open areas, spaces. So doing ethnographic research, spending time in these communities, seeing who the gatekeepers are, identifying hotspots. So doing recruiting there. And then as well as getting into indoor gang spaces like trap houses and—so you might ask, well, how did you, Rachel, get into these? 

Well, I want to talk a little bit about the ways that we approached this study, because it’s a part of the findings itself, that participatory methods and trust building were absolutely essential to this study. So we had to hire a field—we started off with typical research assistants that we might hire. Some of my grad students or things like that. But we soon realized that that wasn’t going to get us access to the population that we wanted to study nor were we—when we did somehow get access, going to get honest responses.  So we ended up hiring field researchers that reflected the study population, both demographically and in lived experience. And they—we hired them, trained them as researchers, and they were consistently present in the places that were important to youth. So like these outdoor spaces and like the trap houses. And in order to get into those spaces, we had to engage gang leaders who then granted us access. We—you know, we assured them that we weren’t the police. And because of the population that we hired to do some of the research, they were able to kind of talk to them and use cultural codes to indicate that they understood street life. They would spend long hours in the community and across the different three neighborhoods, we had different approaches. So something that worked in The Bronx didn’t necessarily work in Brooklyn where people felt less comfortable being outside in these spaces. So we talked to 330 young people, ages 16 to 24. Majority men, Black, living in public housing. Over a third had children. Forty percent were currently in school. But, as I said, we went up to the age 24. And they had significant criminal justice involvement. You can see here, almost all of them had been stopped by the police just in the last few years. Almost 90 percent had ever been arrested. And for those who were arrested, they were arrested an average of seven times. And 63 percent had ever spent time in a juvenile detention facility, a jail, or a prison. So significant criminal justice involvement among this population that we talked to. 

And I want to kind of frame their—remember, the research question that we want to ask is what—the answer is why are these young people carrying guns. So I want to kind of frame this in—with some contextual factors that violence was a near universal experience. Okay. That’s a little bit cut off, but 88 percent had been shot or shot at, 81 percent had been attacked, 70—or 81 percent, had known somebody that had been shot or shot at, may have been attacked with other weapons. So you can see that this violence was a near universal experience for them. 

In addition, they really felt disconnected from the mainstream economy and they often turn to the alternative economy as survive—survival, and guns often came out as part of this alternative survival strategy. You can see here a quote. I’m going to try to mix in some of the quantitative and qualitative data. The interviews that we did were a mix, so we asked lots of closed-ended questions but also lots of open-ended questions as well.  So we got the numbers and the nuance behind it and the full picture. So we found that, really, people—young people in our study were mostly caring to increase their feelings of safety. As I said, they had—many of them had been shot or shot at, attacked physically with a non-firearm weapon, had somebody close to them, a family member or close friend, having been shot or shot at. So everybody was talking about they were caring to really increase their feelings of safety, and they had a pervasive sense of neighborhood distrust and a feeling that they could be victimized at any time, right? So this is more of just a general feel—fear. And then they had more localized fear that they were—they were really fearing that they could be hurt from retaliation. And this quote, the second quote here is from a female in our study. And I think that when—the female gun carriers, things took on a little bit more nuance when they—they especially felt like they had to prove themselves to other men in the community. And then, finally, people talked about fear of the police. So you can see this quote. They really felt that they could be shot at any time by the police and that the police were there not to protect them but were harassing them, arresting them for minor crimes like marijuana smoking or being in the wrong place without an ID or jumping the turnstile. So they were kind of over-policed for minor crimes, but then felt like the police weren’t there to protect them and their community from real acts of violence. And they also felt that they could be shot at any time by the police. And not necessarily because there were things that were happening in their neighborhood, but often because there were things that were happening in a larger racial ethnic community across the country and things they saw in the media and social media almost daily. 

They also talked about—so the big finding, they’re carrying because they’re scared.  They’re afraid for various reasons. And they also talked about care—being very careful and being very cautious around gun use. So I mentioned that, and I’m not going to talk about it today because of time constraints, but one of our other questions were what were these people’s networks like and how were guns a part of these networks. So, in the study, we talked a lot about their gang involvement and—in gangs, and oftentimes they were talking about them and their friends in their gang and their colleagues and their gangs, and sometimes the gang leaders were really very cautious about using guns and kind of from the top was coming down that you’re not really supposed to use them unless your life was threatened. So people talked a lot about how they exercised restraint. And I thought this was interesting because we asked about their drug and alcohol consumption, and they did make a distinction saying that oftentimes the more alcohol that they had, the more likely they were—they were like—more likely they were to pull out their gun, but marijuana had a calming effect that made them kind of help them regulate their emotions so that they weren’t going to use their gun. 

Another context factor is they had extreme distress of the police. As I said, many of these people came from communities where their families had been broken up because of mass incarceration. They felt harassed by the police. They felt like police saw them, especially if they were gang members and the police, when they were gang members, police would use words like animals and terrorists and things like that to describe them and they didn’t take time to get to know them or the people in their community or understand where they were really coming from. So there was an extreme distrust of the police. They didn’t feel like—as you can see, these numbers here are very low. Only 15 percent saying that police have good reasons for when they arrest people and thinking that they really want to understand the community. People talked about their experiences where they felt that the police would increase the problem, where instead of coming in and helping defuse the situation, they would amplify it. And most people talked about like the ways that police think about them are really, really negative and—so they think that they’re out there shooting people no matter what, and—but they aren’t.  As they were talking about, they talked a lot about the caution that they exercise around their gun use. They really felt that—like I said they had this fear of being shot at any time and often had a racialized factor being like, “Okay, we’re young and Black, and the police—we just expect the police—when we exit the house, that the police are going to harass us.” 

So key takeaways. Youth mostly carrying to increase feelings of safety, there was a widespread belief that they could be victimized at any time, and gun served to protect them from these real and perceived threats from other gun carriers, so from rival gang members, from residents of different housing developments that were often in conflict with each other, and the police. And these street-involved young people are really caught in double binds, they’re potential targets from other youth and at the same time other—from the police that is theoretically in place to protect them. So further efforts that they take to provide for their own safety merely serve to increase their vulnerability from both these sources. And also another theme that came out, the key takeaway is this institutional abandonment. They felt there was like a lack of any pathway to methods of survival, just economic, physical, emotional, and educational. And when institutions fail to provide and people are less and less able to access the necessary supports for themselves and their family when there aren’t other mechanisms to survive. They turn to the street economy and guns—and guns to ensure their safety. They felt they were out there on the streets because they can’t get jobs, they can’t go to school, they don’t have safe and stable and affordable housing. And police strategy is that they’re sort of perceived as nonexistent for serious offenses and militaristic for minor ones, and more policing was a tactic that increased their paranoia, fear, and sense of imminent harm, which only lead to more gun carrying. And it’s not that they don’t see a role for the police. It’s just that the more aggressive and more warlike the institutional response, the less these youth have any ability to back down or to pursue things that would make them safer. 

So our recommendations. So we want to, first, bring services to spaces important to youth. And this is literally. Participants, they often pointed to the restrictions on their movement. Given the concerns over violence, they didn’t feel like they could go between different neighborhoods. They couldn’t—sometimes within their own neighborhood, they couldn’t pass a different housing development. So community organizations must bring their necessary services, which these youth actually want.  They have to bring them directly to the localized basis where these young people congregate. And that’s why we were able to access over 300 young people is because we did that. Hire more credible messengers. So I mentioned that we did some recruiting from Cure Violence programs, which is just one program that utilizes the credible messenger model. So, you know, community organization seeking to engage high-risk youth, they have to hire a frontline staff who the street experience and cultural knowledge to connect with these young people. Invest in safety strategies outside of law enforcement and—so given the deep distrust of law enforcement, community strategies to ensure safety have to be developed that foster trust and encourage healing, and it cannot be done only by the police. And then, finally, as I mentioned, a lot of these young people were engaged in the street economy as a method of survival, so we need to create job programs, specifically for youth and justice-involved people. They often talked about—I mentioned their extensive criminal justice involvement. They often talked about not being able to get a job. They even said, you know, they are not supposed to look at these criminal records anymore but they’re still doing it. So create programs specifically for these young people and justice-involved people that sort of—are designed around the social realities, based by these young people who must work to build—they have to work to build their qualifications and experience, and these jobs have to provide concrete—to—pathways to living wage jobs and long-term career paths and not just be “We’re going to help with your resumes,” that, at best, lead to very low wage jobs and sometimes no jobs at all. And if you want to access these young people, you have to engage gang leadership. So I mentioned one of the research questions were—was what are these young people’s networks like, and gangs play a significant role in their lives and—in a positive and negative way. And given the role that gangs play in these young people’s lives, they have to be partners at the table for any discussion of community services and safety if you really want to access this population.  So I mentioned there’s lots of other things that are covered in the report. You can read more about it at this link. And you can also—there’s a podcast that we did with some of the participatory researchers whose lived experience reflect that of the study population, and you can hear more about how we did it. And feel free to email me with any questions. Thank you.

BARBARA (BASIA) LOPEZ: Thank you, Rachel, very much. Big applause to Rachel for her presentation. We appreciate that. We are ready to move to our second presenter, Rose Kagawa. She is an Assistant Professor in the Department of Emergency Medicine at the University of California, Davis. Her current research is on comprehensive background check policies, the built environment and violence and poverty reduction interventions. Dr. Kagawa received her PhD in Epidemiology and her MPH from the University of California, Berkeley. Today, she will present her research findings on the NIJ-funded research project titled “Preventing Firearm Violence: An Evaluation of Urban Blight Removal in High Risk Communities.” With this, I will give the floor to Rose.

ROSE KAGAWA: Thank you so much, Basia. I am really excited to be a part of this panel today because it’s a great opportunity to broaden the conversation about gun violence in America and think about the environments, social and physical, in which firearm violence frequently occurs. A lot of times when we talk about preventing firearm violence, conversations can focus pretty narrowly on firearms or perhaps individuals with firearms. And that completely makes sense. All firearm violence evolves firearms and a person pulls the trigger. Next slide please. However, the firearm and the person don’t exist in isolation. They have a personal history. They live in a place. That place has a history. And when we think about the broader drivers of rates of violence about the environments both social and physical in which firearm violence takes place, we open up our opportunities for prevention and sometimes for large scale prevention.  Next slide. 

Place matters. Where we live determines who we interact with on a daily basis, our friend networks, the services we have access to, the activities we engage in, the quality of our kids’ educations. And this is not just by chance. Policies such as redlining, restrictive covenants, and even urban renewal in the 1950s and ’60s sometimes explicitly divided people into different neighborhoods based on race and wealth.  Neighborhoods and just about everything in them impact individual health risk. Slide. 

People may have heard of the fact that life expectancy varies greatly by ZIP code. And, in fact, neighborhood is one of the best predictors of life expectancy. This is a map of Wayne County in Michigan where Detroit showing the disparity in life expectancy by neighborhood. These stark differences we see here exist, in part, because neighborhoods reflect and maintain the unequal distribution of resources in society. Next slide. 

Criminology has shown us similar concentrations in place, but this time we’re showing crime in Detroit. And these are often referred to as hotspots. So these locations often share similar physical characteristics that might be around alcohol outlets or abandoned buildings. Next slide. 

Post-industrial cities in the Midwest often experience rates of firearm violence that far exceed the national average. Following industrial restructuring, a lot of these cities lost jobs, incomes fell, the tax base shrinked, and the people who could, sometimes left. And with further shrinking budgets, city services are strained, opportunities shrinks, and crime increases, and, again, the people who can, sometimes leave. So we see this spiral of depopulation, the shattering of business districts, increasing abandonment, foreclosures, and long-term vacancies, and increase in crime. In foreclosures, and perhaps particularly the vacancies that follow foreclosures in an unhealthy housing market, are known to be associated with crime. Next slide. 

Research has shown that interventions on these places may reduce rates of firearm violence locally. And one of the most rigorous studies to date, vacant lots in Philadelphia, Pennsylvania were randomized to receive a light touch greening intervention. It involved trash removal, the planting of grass and trees and regular mowing. Or they were assigned no intervention. And the study was led by Dr. Branas of Columbia. Estimated that this greening intervention led to significant reductions in shootings in the treated areas relative to the control areas. So this is a really intriguing concept. Next slide. 

Now, Detroit saw ballooning foreclosures in home abandonment following the 2007 financial crisis. In response, Detroit has applied a range of legal and policy tools to address this issue, including the demolition or restoration of dilapidated homes. And since then, Detroit has spent millions of dollars on the demolition of vacant and decaying properties, primarily as an effort to stem the foreclosure crisis and prevent further neighborhood destabilization. But recent research suggests interventions to remove dilapidated buildings from the landscape, a much more intensive intervention than the greening I described a moment ago, may provide cities an additional means for preventing violence. Next slide. 

The primary objective of this study then was to estimate the effect of demolish—demolishing vacant and derelict buildings in Detroit, Michigan on the prevalence of firearm violence in 2017. Next slide. 

Detroit has among the highest rates of fatal violence in the country. In 2019, the homicide rate was forty-one per hundred thousand. That’s thirty-five per hundred thousand more deaths than the national average. And from 1960 to 2000, the population of Detroit declined by more than 40 percent. As a result of this and subsequent continued depopulation, Detroit has a high concentration of vacant and abandoned homes. It’s estimated that nearly one in four homes in Detroit are vacant.  Detroit also has one of the most extensive demolition programs in the nation, having demolished more than 20,000 buildings since the demolition program began in earnest in 2014. Every green dot on this map is a demolition. So these factors make demolition—or Detroit an ideal place to study the effects of demolition on firearm violence. Next slide.

 

Today, I’m presenting some preliminary results for our first two outcomes. These are the FBI-defined part one violent crimes, and they include homicide, rape, robbery, and aggravated assault. I’ll show—I’ll also show results for these violent crimes that involved a firearm. So these data are from police reports and they’re publicly available from Detroit’s Open Data Portal. 

The unit of analysis here is the block, so all measures are aggregated to the spatial unit.  And by block, we literally mean like a neighborhood block. The map on the slide gives you a sense of the size of our units and what they look like. The orange lines show the violence of each block. And there are just over 15,000 blocks in Detroit. Next slide. 

We have a very specific research question and that is what would have happened to the prevalence of firearm violence had there been no demolition in the preceding quarters of 2017. So we haven’t actually tapped them, in terms of demolition and crime for each quarter. In quarter one, a certain number of blocks experienced demolition and the following quarter the crime rate was X. Next. 

In quarters one and two, there was a certain amount of demolition, and by quarter three, the crime rate was Y. Next. And so on. As such the observed prevalence is our measure of firearm violence in the context of demolition. But what we want to know then is what would have happened had this demolition not occurred. Next. 

So for each quarter, we obtained an estimate of what would have happened had the preceding demolitions not taken place, and then we can make comparisons between the observed prevalence of crime and its estimate of the counterfactual prevalence. Next slide. 

The process of foreclosure, home abandonment, and demolition is complex and reciprocal demolition is an intervention that is both affected by and likely has an effect on many of the same factors that impact violence and possible inference in this scenario can get really tricky. We used a method called longitudinal targeted maximum loss-based estimation. That’s a mouthful, so people call it LTMLE. LTML—LTMLE is a semiparametric, doubly-robust substitution estimator, and it allows us to incorporate flexible machine learning method called Super Learner to fit our models. And what all that boils down to is that this approach allows us to account for dynamic relationships among our variables in a way that traditional regression approaches often cannot. Next slide. 

The other benefit of LTMLE is the ability to incorporate machine learning in order to model nonlinear relationships. So we’re using over a hundred variables that describe every block in Detroit, either primarily sourced from parcel level administrative information. These variables include information about the presence and types of buildings on the blocks, whether they’re industrial, commercial, et cetera, characteristics of those buildings such as age and size, and indicators of the health of the local market, so sales values, foreclosure rates, vacancies, whether things are renter versus owner-occupied. And we couple these data with information from the American Community Survey on demographics, education levels, and other socioeconomic indicators. And then using Super Learner, we allowed for more flexible relationships among these variables. So most regression-based techniques assume an additive linear or log linear relationship between the independent variables and the outcome, but, in reality, the probability of the outcome likely has a more complex relationship with many of these explanatory variables. 

For example, perhaps crime increases with increasing vacancies, up until, say, 60 percent of homes are vacant. And then maybe actually crime decreases with increasing vacancies because there are fewer people that perpetrate or be victims of that crime. So that would be a nonlinear relationship that this method is more likely to capture. 

Also, I just want to note that we control for the number of demolitions occurring in a block prior to 2017 as well as spatial lags for demolition and the outcome, and we adjust our standard errors to account for clustering of blocks within the census tracts. Next slide. 

To give everyone a sense of Detroit in our time period, the left here shows a density of vacancies and you can see some large areas with relatively high levels of vacancy in the darker blue and on the right we have the results from k-means cluster analysis with education, age, home values, and other socioeconomic variables. And the gold and blue areas in that map tend to have higher education levels, lower unemployment rates, higher home values, and they’re more racially diverse. But gold—relative to blue areas, the gold areas have few—they tend to have fewer kids under 18 in the home and slightly higher population density. And in the pink and green areas, unemployment tends to be higher. People are less likely to have a bachelor’s degree. Home values are a lot lower in these areas. Tend to be predominantly African-American. And what distinguishes the pink and the green is that the pink areas are slightly more disadvantaged in terms of those same measures and they also tend to have a larger household size with a few more kids and be a bit younger in average. Next slide. 

There were more than 2,600 total demolitions in about 1,700 blocks in 2017 alone.  Nearly all of the demoed buildings were sourced from foreclosures. Just prior to 2017, Detroit launched the next phase of its demolition program, placing priority on the areas that had not yet received demolition, marked in blue on the map on the right. And you can see a greater concentration of demolitions in these areas in the map on the left. So they lined up pretty well. Next slide. 

The slide on the right here shows just more clearly which blocks received any demolition during the study period, emphasizing that with the exception of downtown, in northwest Detroit, demolition occurred just about across the city. Next slide. 

And here you can see the distribution of firearm violence in the city over the study period. We see a somewhat higher density of violence in West and East Detroit, as well as the area just northeast of downtown. Next slide. 

So our estimate of the counterfactual, we call it the prevalence of violence had there been no demolition, suggests it would have looked just about like it did in reality. So this table shows differences in the prevalence and the confidence around—confidence interval around those estimates. And the results are fairly precisely naught. So we’ve repeated these analyses in a number of ways using block groups, which are just a slightly larger geographic unit than blocks, and among blocks and block groups with no previous demolitions, and these models have yielded largely similar results to what I’m showing here. The study is still in progress so additional results are forthcoming. For example, we’ve not yet tested whether there’s an effect on firearm violence in areas receiving a greater dose of demolitions or whether these results vary by neighborhood characteristics. There’s more to come. Next slide. 

I also just want to note a few limitations. First, with model assumptions in mind, we had to limit association so that demolitions and neighboring blocks are not included in effect estimates. Second, we were only able to obtain crime data with accurate location information for 2017. And we had planned to study the impacts of demolition since 2009, but without crime data, this was not feasible, so it’s very possible that results would be different for a different time period. By 2017, the city had already demolished more than 10,000 buildings, and perhaps the most dangerous places has—had already been removed. We also know anecdotally that demolition efforts were first targeted to places with the highest population density. So by 2017, these demolitions may be in areas with relatively lower population density. And the last limitation I’ll mention here is that we’re estimating effects for a lower dose intervention than previous research suggests is effective. So I mentioned this is a work in progress and we have more to do. We’re—we’ve got two more outcomes to include. One is looking at drug crimes and also crimes that are likely to occur around abandoned properties, and then we’re also doing this study—the same study in Cleveland, Ohio. I just want to finish by saying it’s important to recall that only two of nine previous studies of demolition and violence have not found an association. So it’s also important as we move forward to think—to do some additional work to try to understand what may be leading to these different results, because it’s actually this information could be critical in designing and targeting demolition programs. And last slide. 

I just want to acknowledge the National Institute of Justice for the funding and my magnificent research team without whom this would not be possible.

BARBARA (BASIA) LOPEZ: Thank you, Rose. Again, great applause to Rose for the presentation. Thank you very much for sharing these findings with us in the audience here. Next presentation will be—will be provided by Max Griswold. He is a Policy Researcher at the RAND Corporation whose work analyzes behavioral risk factor policy and social determinants of health. His research has appeared in The Lancet and JAMA, and his work on alcohol use has been featured in The New York Times and Washington Post. His current project investigate the role Crime Prevention Through Environmental Design policies have on eviction rates in cities and the limitations of the National Instant Criminal Background Check System’s database for research purposes. Today, he will present findings of the NIJ-funded project titled Understanding Socio-Environmental and Physical Risk Factors Influencing Firearm Violence. With no further ado, Max, the floor is yours.

MAX GRISWOLD: All right. Thanks. Thanks for everyone coming out today. So as far as about a research project, I’m going to show you today what’s part of a broader work on the relationship between the built environment firearm crime. My component, which has this rather pedantic name that you see here, is looking at what a relationship between building types, like banks and schools, so if you find both banks and schools in a census tract, how does that relate to violent crimes independent of just finding a bank or just finding a school. You also see from what—someone they call Luke Muggy to be presenting after me on how distances from different building types can relate to violent crime. And we also have another part of the team [INDISTINCT] where we spoke with community members across study sites to see how they think the built environment might influence violent crime. So even though I’m going to be focusing on these latent environmental determinants in the segmentation, I might be drawing on some aspects as broader work. I also want to recognize this was a team effort and not just my own work. So shout out to my co-team members, Priscilla Hunt, Rosanna Smart, Sean McKenna. All right. 

So getting into it. So this work is rising on of a theory on built environment, how it influence and modifies the incidence of crime. So what do I mean by this built environment? Anything built by people we’re going to encounter when we walk through a city. So it could be buildings, like banks or schools, restaurants, but it could also be building features like street lamps, fences, or road signs. Tons of different hypothesis on how the built environment might influence crime, like Crime Prevention Through Environmental Design or CPTED, social disorganization theory, reconnectivity theories. These theories are tending to focus on built features. So things—like in CPTED, discuss how light contributes to natural surveillance. But there’s also a large fraction of academic work on how buildings specifically affect the crime. And I’ll give you some examples of this paper in a second. But, mostly, these papers are focused on single types of buildings. So how do—how does a public park relate to the incidence of crime, or schools, or street intersections? It’s a great starting point. But setting this relationship on just single building types and crimes could lead to wrong conclusion. So, for example, this image right here, you can see planning zones within Los Angeles. 

So when a given zone is for a specific purpose, we might expect to find similar buildings within that zone. So a mixed residential zone, we might have checked—there could be restaurants or bars, so on and so forth. But if we’re studying the effects of restaurants and crime rates, if we’re not accounting for zoning, we might not be analyzing the effect of restaurant. We might think it’s highly correlated the analyzing effects of that zoning. So given these correlations across building types that could arise from zoning or some other unobserved process leading to how a neighborhood has specific buildings, we need to be careful as researchers when we study the built environment spectrum crime. If we’re not necessarily adjusting for all these built environment features or clustering, we might arrive at the wrong conclusion. And so knowing this information could be really helpful for us to see what the effect is of zoning on crime or just understanding how neighbor characteristic contribute to crime reduction and help us maybe to plans some safer cities. 

So what did we do specifically in our project? Well, for one, we want to determine if we have these latent built environmental factors and if they associate with firearm crime.  So what do I mean by the word latent? So these are variables that we infer rather than observe. A classic example here from economics is something like quality of life. We can’t measure quality of life but we can observe variables like income, educational status, marital status, health status, whatever it might be. And in tandem, we can use those to infer something about quality of life, but we can’t actually measure quality of life itself. So for this branch, I think, well, we could measure zoning, and that those maps exist. I just showed you one. While it’s true, just because a building—an area has been zoned for something specifically, what’s actually built in there could differ from that zoning plan. So we’re interested in what naturally occurs given that zoning rather than what had—it had been zoned for which is more abstract. And, lastly, which encompass the overall project but also this specific aim was how suitable is open-source administrative data that you can get from something like data.gov or other databases for conducting academic research on the built environment. 

So let’s talk about our data in these slides. We can assume conclusion from that as well. So I’m not going to spend too much time on this, but this is a well-researched area, going back until—well, pretty far, but really a lot more quantitative data starting their analysis in the 2000s. These studies mostly come in one of two ways, observational studies or quasi-experimental methods. I’m using in this work observational study, but some of the critiques that I’ll mention would apply to all these papers regardless of the specific method they used. I’ll also point out two papers here on the far left that came out in 2020, which is a similar type of approach to what I’ll be showing you in a second. And we came to results that were very much in line with these papers, which found that the built environment and violent crime can depend heavily on the city in which it occurs. So what that relationship actually is.  

So I tried to make this talk a bit less technical. There’s a lot more steps that go into this approach than what’s here but I’ll just briefly go over what we were doing. So, again, our goal is to analyze the relationship between violent firearm crimes in both these built environment features and factors like zoning or neighborhood characteristics that are based on those features. So we did that by extracting a whole cost of data, which I’ll show in a second, and this is pertaining to your data on the built environment, socio-economic variables or firearm crime variables. We were using these data on a built environment socio-economic to perform what’s called a factor analysis, and I’ll briefly explain that. But it aims to construct a series of factors which are linear combinations of these built environmental variables that explain covariation across variables. So it’s kind of similar to something like a PCA-type approach or principal components analysis. 

So as part of this method, we have to determine which specific buildings we want to include within this data analysis, and some of them might not correlate well with other variables. And we also have to figure out how many factors we want to produce. I had methods for those that I can go into. It’s called KML scores and parallel analysis. Happy to explain those if there’s interest. Once we produce these factor variables from the factor analysis, we then use these as predictors with our regression models to try and predict firearm crime. All right. We—so we only have the set of latent factors based on the built environment—the built environment itself and then socio-economic variables for recent adjusters. Within each location, we’re doing about 24 different models that include variables in slightly different ways, and then we’re ranking these models based on how well they performed with data that we did not give the model upfront to see. We then ranked those models and we looked at the relationship for the best performers. So what data was informing this approach? 

So you can see on the right all of the different open data systems we use, which was a ton, across the four sites which were Pittsburgh, New Orleans, Detroit, and Los Angeles. We’re getting report of firearm crimes involving a firearm—I’m sorry. Report of crimes involving a firearm, specifically violent ones including homicide, robbery, and aggravated assault. We’re doing specifically violent crimes given known issues using incident report data in terms of how it’s coded and we’re extracting this from a city’s open datasets. These are reports between 2015 and 2018 depending on the city and the time of data collection. We get about 30,000 observations for crimes in Los Angeles all the way down to about 2,000 in Pittsburgh. 

Our data on the built environment, I mentioned one here but we actually tried two different sources. So we got it from either the county officer or the assessor where they do property assessments and code buildings into different building types for every parcel within a city. So depending on the city, we have anywhere from a hundred thousand to a million observations. We also got data from mayor’s offices on specific building types. So there might be a dataset on just the location of that [INDISTINCT] so we collected a ton of those. We ultimately only used the Office of the Assessor’s data because it had more consistent definitions, but we did collect a large amount of data that did not undergird this specific field analysis from other data systems. So one issue we ran into specifically with these open datasets that was an issue for us was the cost of living data that a lot of cities are using. 

So these are databases where the data is constantly being changed and uploaded through time. So if you look at the data one day, it could be completely different the next. And this is the only dataset that’s being made available for these open data portals, which makes it really difficult for research purposes, because we would love to be able to explain variation that occurs through time to identify a causal effect. But we’re not able to do that if we’re getting only a slice in time, and it’s constantly changing. It also makes it more difficult for research purposes when the data set that I’m using could be really different than the data set you use next year to try and replicate my work. So it would be great if all these open data systems could both provide that loading dataset and then slices of that dataset through time of historical records. But right now those living data systems which was a lot of these data sets made it much more difficult for us to do quasi experimental methods that we would have—would have preferred. 

And then lastly, we got some socioeconomic status variables and population variables from the American Community Survey five-year series, which has averaged over the study period for data collection. We’re using a bit of mixtures in years, but our argument here is by averaging over those time periods, we can get more consistent trends in something like the instant reports. 

So here’s kind of what our data look like. This is just showing the firearm crime patterns across the time period of 2015-2018. We have pretty different patterns, depending on the city. Los Angeles, it’s very much centered within central and south central area, while Detroit is pretty uniform across the city, and Pittsburgh has far less data points. And as I mentioned, we did end up actually analyzing New Orleans, but you can see the patterns there. So again, our goal of the models will be producing using these latent factors is going to be to replicate the patterns. 

So, to briefly walk you through what a factor analysis actually contains. So, these latent constructs are a bit clear. So on the right, you see what’s called a Factor Table. And this is—on the Y-axis these would be built environment features, so accounted them. So one, two, three, four, five sport fields to be included in the model. And then the columns themselves are the factors that we produce. So it’s just a big table full of numbers. Each columns corresponding to just one factor and there’s a cell in each value, which is called a building value for the football environment features. So if you look at the second column for public pools, the value there says 0.53. So if we have one more public pool in the census tract, the score for factor two in that census tract increases by 0.53. But the obvious question is well what is factor two? In general economy factors based on the tables, pretty intuitive, explained, so let’s just, kind of, take a peek at that. So looking narrowly on two we can see and the lighter the blue the larger the value for that built environment feature. So for this one, we have large values for sport fields, tennis courts, public pools, public parks, and playgrounds. So what does this factor—the interpretation? Well, it’s just outdoor activities. So that’s pretty intuitive in terms of what actually means. This wasn’t always the case and this status is off of Pittsburg by the way. So this is the first factor variable, which has—explained a significant amount of the covariance amongst buildings, but it looks really weird. It’s got UPS, university and colleges, transit, tobacco retailers, government buildings, FedEx, credit unions. Like what the heck could that be? Well, we have to do more analysis to, kind of, figure it out on occasion. This one ended up having a really intuitive interpretation. So I just plotted that score variable for each census tract on a map. What does it look like? Well, this factor is just picking up on a downtown core area. And so even though the factor loadings are kind of, strange, when you put on the map it makes it pretty clear what this variable is actually taking off. So we’re going to have all these factor variables, we’re not going to have just for Pittsburgh. For example, we used factor living in L.A. In general, we found the factors that we produced tended to correspond to what looks like zoning, that’s particularly the case in Los Angeles. And they were not really consistent across cities. So these would depend pretty heavily on the specific site you’re looking at to see what kind of general relationships across buildings exist. So here, I’m not—every single cell would have a value, but for visual purposes, I’m only showing the ones with a score above 0.5, which is why factor eight has no loadings here. And in general these, kind of, corresponds to mixed residential use or industrial uses. And so we would—included these variables within our models to then predict final balance. 

So what are those models kind of look like? So I’m going to go briefly for this table on the Y-axis. So what I did is this is just a regression table, but I’m displaying it visually. And I did that using an established simulation approach from Gary King’s work. So the Y-axis here is showing how—from our models in each color of line is a different model. And coefficient value is the slope here. So what was the predicted number of firearm crimes, depending on the number of buildings on a specific type in that tract, so just, kind of, looking at movie theaters here in the upper right and with that green line. So if you go from zero to, I believe, there’s eight movie theaters in a census tract, our model predicts you would go from 20 to 60 firearm crimes, so it would increase the amount of predictive firearm crimes by 40, by having eight more movie theaters. Now, again, this observational study, so these are associations not a causal effect. And—but the big takeaway here is depending on if you include these variables as those factors versus as just the built environment themselves, you can have a radically different interpretation of how that building environment contributes to firearm crimes. So something like—let’s go look at restaurants, which I’ve had in the literature pretty mixed reception in terms of what their relationship might be to crime, but it’s all over the place. So if you’re including the restaurant just by themselves, you find this negative relationship, so the more restaurants, the less crimes. But in our fully adjusted model, which is the pink line, you can see while it keeps going up and up and up. And in general, some of the single family homes, which has had a pretty consistent relationship, we found actually varies a lot if you’re including these in terms of that factor—the factor, it’s just something like zoning, rather than including it purely as a single family home within your model. We also found that these relationships were really inconsistent across cities. And so for something like Los Angeles and this is only displaying a small fraction of the variables we actually included, but we found for something like ours. The relationship there is different than the relationship in Pittsburgh in terms of both the sign and the magnitude. And so it depended very heavily on your site location in a way which include these variables. Okay. 

So our models do a great job of actually recovering firearm counts based off of the specifications. So this was all the out of sample Goodness-of-Fit. So basically we’re able to capture about 77 percent of the variation in firearm crimes. And on the left-hand side you can see the observed number of firearm crimes within census tract. And on the right, you can see what we predicted. They match really, really well. In general, for any given census tract, we’re off by about 15 crimes. In Los Angeles, the median census tract had about 50. And on top of that, we’re missing that when we’re in locations with a high amount of firearm crimes, so South Central Los Angeles. We tend to under predict quite a bit. But for small values, like in the Valley or in parts of Pittsburgh, we get those spot on. And so when we’re off by that 15, it tends to be in locations with a large amount of crimes. But in general, these variables are pretty robust. 

Two brief comments, we found that the inclusion of socioeconomic variables substantially improves how well we’re able to recover predictions in firearm crimes than including your own environment or including the factor variables. You go from an R-square of 0.25 to 0.77 out of fit. We also found that when you include the variables as these leading characteristics like the zoning ones, you do a much better job of predicting firearm crimes than including the buildings themselves, which leads me to believe that that’s kind of the mechanism that likely undergirds. And again, I’m speculating here because its observational data and we did not actually identify causal effects, but it tends to improve the model fit pretty largely instead of the likely source of what’s generating some of those crimes rather than the buildings itself, kind of the general takeaway here. 

But it has some limitations. So obviously, observational study results are very correlated. We tried to choose a model type that allowed us to best control for relevant adjusters, this is also true of the quasi-experimental approaches, where it’s only as good as the assumptions that are made. But we endeavored to capture as much as we could to make the interpretations on building environment variables more robust. We also found that the factors that we produce, they were very robust to alternative ways of estimating them or how we optimize the scores. But they can be difficult to interpret concretely and they differ by city. And then, the underlying datasets themselves have some pretty significant limitations. Like the assessment dataset, while each census tract has codes for the buildings that are included in that, so which kind of building it is, like a bank or credit union, we do not actually have a codebook of those definitions to know why they chose a specific definition for a location or not. Similarly, there’s well known issues of infinite dataset based on—well, if many crimes are unobserved. And so just because it’s in the incident dataset that might correspond to where law enforcement is patrolling rather than the location of all crimes. So there are some limitations. But I think in general, we found that the measure of data has some really useful information. But things like the living dataset or just the variety of different databases, encodings in them made it difficult to use for research purposes, but it’s definitely informative. We also found that these built environment factors greatly improved of predicting firearm crimes at the census tract level. And based on how variable these relationships are, despite a wide series control, I think it’s unlikely these previous studies are uncovering robust causal estimates, particularly because they’re not including that correlation between built environment variables within a tract or within a zone around them. So that’s all I have for you. Thanks for listening and I’ll pass it back to whoever’s the next.

BARBARA (BASIA) LOPEZ: Thank you, Max. Again, a big hand of applause for Max, for the presentation. Thank you. Our next presenter Luke Muggy is an Operations Researcher at the RAND Corporation with a broad range of experiences. Luke obtained a PhD in industrial engineering, focusing on improving humanitarian logistics system—systems, I’m sorry, through geospatial analyses and mathematical optimization. At RAND, Luke has conducted research for the US Army, Air Force, and Marine Corps, helping the Department of Defense improving its supply chain. After Hurricane Maria, Luke worked with FEMA to evaluate cost estimates for reconstruct—reconstruction projects in Puerto Rico. And more recently, Luke’s research has sought to improve FEMA’s approach to environmental and historic preservation. Today, he will present findings of the NIJ funded project on firearms violence, titled “Understanding Socio Environmental and Physical Risk Factors Influencing Firearms Violence.” Thank you very much. And Luke, the floor is yours.

LUKE MUGGY: Wonderful. Thank you very much for the introduction and thank you very much to all the other panelists for their fascinating research. Before I get started today, I wanted to thank my colleagues, Max Griswold, who just gave an excellent presentation before me. Tyna Eloundou Nekoul, Sean McKenna, Rosanna Smart and Priscilla Hunt, and of course, I’d like to acknowledge the funding from NIJ that made all of this possible. Thank you very much. Next slide, please. 

So a little bit of outline of what I’m going to present to you today, I’m going to start with the background and motivation for our work. I’m going to jump into the methods and try to explain how our work builds upon other great research in the past. I’m going to tell you a little bit about our data, which is—which is similar data to what Max presented. I’m going to present our results across four U.S. cities and then hopefully provide a little bit of context and discussion toward the end. Next slide. 

So to get started, motivation. We all know why we’re here. Firearms violence is a very serious problem in the United States and here’s some statistics to sort of emphasize that point. In 2018, there were over 57,000 incidents of reported firearms violence, which resulted in over 28,000 injuries, and over 14,000 willful or accidental deaths, those—that’s in the United States and that does not include suicides. Compared to some of our other peer nations like Canada, United Kingdom, and Japan, the U.S. experience is far more firearm related homicides per hundred thousand citizens. We experienced 12.21 in the United States, whereas Canada experiences two. So, six times higher than Canada and nearly forty-eight times higher than United Kingdom. 

Putting things in perspective, we aren’t quite as bad as Honduras, which is at 60, and Venezuela, which is nearly 50 per hundred thousand people. In addition to the substantial loss of life and pain, these homicides also caused an enormous amount of financial damage. It’s estimated that the U.S. society, the United States, all together, we pay the social costs are about $8.6 million per homicide. Pretty substantial. Next slide, please. 

We also know that firearms violence is not necessarily widespread. We know there’s a long history and a fruitful history of literature on place-based crime, it goes all the way back to 1961 and maybe even further. Though more recently, we have focused on hotspots. This research has helped showed that shootings tend to be heavily clustered in some small spatial areas and are not in general uniformly spread out across the nation or even across a certain city or neighborhood. So we would like to understand a little bit more about where and why do these crimes occur, specifically focusing on different types of built environment features, and by that I mean banks, gas stations, bars, schools, things that like Max said humans built. And knowing the potential influence that a built environment feature could have on firearms violence would help law enforcement—law enforcement focus patrols in certain areas, we might also help city planners design better use of land, physical development, and supporting infrastructure to, well, actually try to influence crime in certain areas. And next slide, please.

So a little bit more about place-based crime in the literature specifically focusing on that which tries to quantify the influence of certain built environment features. Previous studies have been fruitful, though largely focused in the New England area, at finding certain risk factors and—for firearms violence. So namely, they found that retail businesses, at-risk housing, bars, liquor stores, and dance halls included—are also are attractive to firearms violence. Also, grocery stores, bus stops, and residential foreclosures are risk factors for firearms violence and other types of crime as well. At the same time, they found certain built environment features are not—do not seem to attract firearms violence, gas stations, fast food, restaurants, laundry, and then also schools are seen as not, in the studies that have been done before, are seen to not attract firearms violence. 

This particular study, as we try to better understand the influence of built environment features on the incidence of firearms violence in the general vicinity. We focus on four United States cities, Pittsburgh, Detroit, Los Angeles, and New Orleans. And we do make some, in addition to looking at more geospatially spread out and diverse cities, we also make a contribution to analyzing how socioeconomic status plays a role within this, which I’ll get to in just a moment. We test new features. And we also implement—our method also implements the shortest path distance instead of as the crow flies distance, to try to get more precise measurements of the actual end—of the distance at which different built environments might exude some kind of influence. Next slide. 

Okay, so now a little bit about our methods. We rely on a Monte Carlo simulation and two equations. I’m not going to get into too much of the math but focusing on the equation at the bottom this is called the Network Cross-K Function for Stochastic Spatial Events. We are measuring, we’re using this to basically count the number of built environment and—or sorry, number of firearm incidents that take place within increasing distances of certain features. So can you bring up the next—the boxes? A couple more, you can click through. Yeah. There. Perfect. 

So the main parameter in this equation is P, sub i, jt, which is a binary parameter, which just takes on a value of one, if feature i is within distance T of shot J. And we’re adding all of these up for every feature of the same type. So this summation in the bottom is really just adding up the number of firearms incidents that take place within increasing distance thresholds of a feature. And then we multiply this count by a little fraction on the outside, which helps normalize across different networks. That fraction includes the cumulative edge length of the network, the number of features of a particular feature that we’re looking at, and then also the number of instances of firearms violence. 

We are using this equation here in—twice. First we are applying it to observe data on the actual locations where firearms violence occurred. And then we randomly distribute through Monte Carlo simulation, the same number of firearm incidences as actually occurred uniformly or randomly across the street network. And then we use—we leverage the Cross-K equation again, but this time applying it to the simulated data. We run the simulation a whole bunch of times, 200 times to be specific, which gives us a range of simulated Cross-K values. And then lastly, we compare the observed Cross-K values with the range of simulated Cross-K values. And we find that if the observed values are greater than the simulated range, we call this an attractive influence. And if the observed values are lower than the Repellent Range or lower than the simulated range, we call it a Repellent Influence. Because the Cross-K function is a cumulative distribution function of the count of firearms violence is at distance, we could have applied a KS test, Kolmogorov-Smirnov test to analyze the statistical difference between these. However, in practice, it didn’t work out very well so we had to use a general observation instead. Next slide. 

So we use that Cross-K function that I showed you there to assess whether there is a significant difference between the observed value and let—in the simulation occurred randomly. Once we know whether or not there’s a significant influence one direction or another, we’d like to measure the risk at varying distances from these different features.  To do this, we’re using something called the Shot Density per facility measure. It could also be firearms incident density measure. This S sub t, which, in the numerator, go ahead and click twice. The numerator is again, the—I guess one more time. Is the count of firearms violence that occurred in—within a distance T of these different features. And then we divide that count by number one, the distance value T, to try to address the fact that influence of a particular feature would dissipate with distance. So the further you are away from particular feature, we hypothesize the less influence that feature would have on firearms violence. And then the other thing we divide by is the number of features so that we’re able to make comparisons between feature types that have lots—maybe one feature type has a small number of locations, and another feature type has a lot of different locations. And by dividing by the number of features-locations, we’re able to compare it, make a comparison there. So just as an example of this Shot Density equation, let’s pretend we had two restaurants some—in a given city. Suppose one firearm incident occurred within a hundred feet of one of the restaurants, well, then the Shot Density at a hundred feet would be point .005, 1 over 200. Next slide. 

So we’re going to use these two equations to try to first assess whether or not their given built environment feature exerts a significant influence and then try to quantify the risk at increasing distance thresholds. To do this, we need data on firearms, violence, and on locations of built environment features. We use the city’s open data portals to gather this, and, of course, there’s lots of complications and limitations to the data. But in summary, we had a whole lot of features. You can see across Pittsburgh, New Orleans, Detroit, and Los Angeles we have between 25 and 36 feature types per city. The—we also have the number of firearms incidents over different time periods. Unfortunately, we were unable to find a single time window in which all four cities had consistent firearms data and built environment data. It would have been really nice if we could have the same time window for all cities. But unfortunately, the data didn’t allow that. Los Angeles, you can see, has a huge number of features and firearms incidents over our time period, which did present some comments some problems with computational tractability in the simulations. Next slide. 

Okay, so you might recognize these from Max’s presentation. You can see the geospatial distribution of firearms violence here. Couple just things to point out, I think it’s—you can—you can see the clusters present themselves well in New Orleans toward kind of the downtown area and also in Los Angeles toward the downtown areas. Detroit was interesting in that it was so uniformly—the firearms violence was so uniformly spread out that we think this might have contributed to our finding that there weren’t too many built environment features that were especially attractive in Detroit, because it just seems so uniformly spread out. Next slide. 

So Results, what did we find was attractive? What did we find was repellent and how much? Shoot. Well, it appears that when we turned these into—these slides into PDFs, it took out the simulation ranges. So you’re gonna have to take my word for it on the top there, I’m really sorry, there’s a simulation band missing from each of those graphs. But the going—in Pittsburgh, the tobacco retailers were attractive from zero to four hundred feet. So, in the direct vicinity of the 427 tobacco retailers. In the bottom there, you can see the Shot Density plot, which is a classic illustration of what an attractive feature looks like where there’s a high shot density value for low values, for low distance values. And then the further you get out from the—from the feature, the lower the shot density gets. So that’s kind of what we would expect to see in a lot of our attractors. However, it’s not interestingly. In New Orleans, the 316 lodging locations, which include hotels, motels, and B&Bs, and hostels, was the greatest attractor of our entire study. In fact, it was attractive the full zero to a thousand feet that we were looking. However, we are looking for context here a little bit. And so we considered that perhaps it could be colocation of these lodging facilities with bar areas, such as the French Quarter and touristy areas that maybe it’s the colocation of these things that’s actually drawing the firearms violence. However, we’d like to submit that these touristy areas are also very highly patrolled. And oftentimes, hotels will have their own private security. And so we’d like to submit that perhaps it’s not the—actually the lodging areas in the touristing—touristy areas that are actually the attractors. But that remains to be seen and we need some further analysis to confirm that. 

Lastly, on this slide, transit stations, which are just rail stations in Los Angeles, just the Metro, 57 stations, these were attractive in the direct in the immediate vicinity, specifically from zero to 360 feet from the station. For this one, we—for a sub sample of these results, we were able to run our simulations independently on the lowest 25 percent socioeconomic status locations, transit station locations. So we did just those lower SE—just those locations in low SES areas and ran our simulations. And then we also did it on the high—the highest quartile SES locations to try to understand how do our results vary by socioeconomic status. You can see here in gray that it’s actually the high socioeconomic areas that present the greater risk when it comes to rail transit stations. The low SES areas are relatively low risk compared to the high ones. I can tell you anecdotally that living in Los Angeles, I’ve seen higher security and police presence at rail transport at Metro transports specifically in cities like Santa Monica and Culver City, some of the more—the higher SES areas. Next slide, please. 

And again, I’m sorry that there’s no band for simulation here. There were so many convenience stores and supermarkets in Los Angeles that we were not able to run our simulation on the full set. But we were able to run our simulations on the lowest quartile SES and the highest quartile SES locations. We found that both of them are attractive to firearms violence, the full zero to a thousand feet strongly attractive, in fact. This may be, and be in part, due to the potential for robbery and the potential for crossing paths of predator and prey and the opportunity. However, it could also be things that have to do more with what’s going on near gas stations or near convenient stops such as drug dealing and prostitution. Interestingly, and counter intuitively, again, we see that is the high socioeconomically status areas that present the higher risk when it comes to convenience stores and supermarkets and the prevalence of firearms violence nearby.  As you can see—as you can see in the graph below, the gray scatterplot is much, much higher than the black one. Okay, next slide. 

Okay, so you can’t see it, but in the—in the simulation bands for these ones, the simulation bands are above the observed line, meaning that these particular features are repellents. Firearms violence does not occur near them. In Pittsburgh, this was public parks. In New Orleans, these were schools. In Detroit, this was sport venues.  And in Los Angeles, this was a Veteran Affair Centers. We actually find that more features than not are repellents of firearms violence than attractors. And I—and I think there’s individual reasons as to why each feature might do that. For instance, sport venues, we heard during our interviews that these—across cities are uniformly repellents of firearms violence and we think that this is because when they are in use, they are highly—there’s security forces and police presence. And when they’re not in use, they are empty parking lots and there’s not much potential for robbery or theft when they’re not in use. So, just one example. Okay. Next slide. 

Lastly, we have a little discussion. So, we are looking at average or aggregate results across the city where we aggregate all features of a similar type together. We find that the majority of them exhibit a repellent influence, as I mentioned, and I gave a little bit of a reason why we think sport venues might be in there. So, as I mentioned, there’s probably individual reasons for each type of built environment. We did find some attractors namely lodging in New Orleans, which was the strongest one and a novel result. We did find the alcohol outlets in New Orleans were attractive, which was also found for—in Newark, New Jersey and in Irvington, New Jersey in previous work. And lastly, we found that transit stations, rail transit stations, particularly in the high socioeconomic areas, are attractive. Bus stations, I believe, were also found in—to be attractive in Irvington, New Jersey. And lastly, we found no significant influence of fast food restaurants, which is—aligns with previous work in Newark, New Jersey as well. So, a little bit of new and a little bit of corroboration of previous results. Next slide. 

Last thing I’d like to mention is that this is not the end of the story. We need to do a better job of contextualizing these. I’m not a criminology expert, but I’m going to be relying on some of my colleagues to contextualize some of these results. Specifically, we need to consider firearms incidences that occur along the border of cities and how to treat these, how to deal with the number of features or the physical size of features, you know, comparing a giant university with a small gas station, that kind of thing. Considering socioeconomic status as we’ve presented and approached here, but certainly there are improvements and new iterations of other approaches that might be appropriate. We should consider the ambient population around certain built environment features. And last but not least, we should consider collocation of multiple features, maybe reinforcing an influence, mutually reinforcing some kind of influence. But that’s what I have to share with you today and thank you very much for your attention. And if you’d like to send an email a question, my email address is [email protected]. That’s [email protected]. Thank you very much.

BARBARA (BASIA) LOPEZ: Thank you very much, Luke. Let’s give a virtual applause to Luke for his presentation and definitely let’s give a virtual applause to all of our panelists. If you could turn on your cameras, this—let’s just—requested to Rachel, Rose, and Max.  We will have a session now for Q and A. We already have several questions in the queue. I will read them. In the meantime, I strongly encourage the audience to ask the questions as we will carry on with this discussion now.

First question comes to Rachel. “Great talk and really important findings. I noticed that most of the participants had children and I’m wondering if this came up at all in the interviews or if you can say anything about how that plays into their behavior and experiences.” Rachel?

RACHEL SWANER: Great. Thanks for the question, Dara. So, it was 37 percent of our sample had child—reported that they had children and we didn’t go too much into the open-ended questions about it, but for a lot of them, especially for the males in the sample, the child was not always living with them. So, excuse me, but some did talk about the need to engage in the alternative economy like to survive precisely because they needed money to provide for their family. So, we didn’t—we didn’t go as in depth into it because it wasn’t really specifically part of our research question. But now that we have the finding that over a third of these young gun users are parents, I think it’s something that future research should really address thinking through how that might change their behavior.

BARBARA (BASIA) LOPEZ: Thank you, Rachel. The next question is to Rose, “How long did greening vacant lots reduce violence? So for how long did greening vacant lots reduce firearm violence? And was it a short-term or did it—did it persist?”

ROSE KAGAWA: I’m guessing this is about the study by Dr. Charlie Branas and they looked—they did an 18 months prior to intervention compared to 18 months after. So, the outcome would have been averaged over those 18 months. But I also just want to add that the literature on demolition effects looks at a lot of different time periods, so some models are using days, some are using weeks, months, quarters and even years. And there are—there’s a lot of variation in sort of what you’re seeing there in a generally shorter time period of showing larger…

BARBARA (BASIA) LOPEZ: Thank you, Rose. The next question I think is to Max. “Why do you think these factors were so much from city to city and could you try to create city archetypes to increase consistency?”

MAX GRISWOLD: So, it’s difficult to say concretely what could be causing them exactly.  There’s two main sources in mind that I would guess. One would be the distribution of firearm crimes within the city. We’ve mentioned that’s kind of in Detroit where you can kind of see it’s a uniform and so it’s difficult for the model to pick up on. There’s just not enough variation in terms of whether those firearm crimes are occurring to take upon the effect that those buildings might be having and so it’s going to be a lot more variable for that location. So, it could be the distribution of firearm crimes within the city. The other places, it could be the datasets itself that we’re using pretty radically different data from city to city based on how the administrative data is being collected. So, this could be coming from what was the intensity of law enforcement in both the coding for specific crimes over controlling within a city and the strategies they’re using or could be coming from the assessment data in terms of what they’re choosing or not to code, in terms of the building itself. So, there’s a lot of ways in which that could change from city to city. This idea of making archetypes makes a lot of sense to me. Something I would immediately think to do is perhaps we can map studies based on the distributions. They have a fireman crime around some of these tracks so if there’s a lot of clustering like in Los Angeles, there’s likely other studies that have that kind of clustering, and so if you could match those together, maybe you could find more consistent patterns and say comparing that to Detroit where you set more uniform pattern. So, I think it’s a really interesting idea, that’s something we didn’t explore here, but it seems like a really natural extension.

BARBARA (BASIA) LOPEZ: Thank you, Max. Another question that we have is for Luke. “With lodging having such a wide radius, could there be another factor not identified?”

LUKE MUGGY: Yes. Absolutely. So, some hotels are really large and you can imagine that a firearms incident taking place near it means on any side of this huge thing, which makes getting precise distances difficult. I would—I would think that there’s other types of tangential crimes that might be occurring near lower—in lower cost hotels that can’t afford private security such as prostitution and drug dealing that might be attracting the firearms crime itself. Though I don’t have any specific data to confirm which hotels are—which types of hotels and lodging are the ones that are attracting firearms violence, I need to do some deeper diving on that. And, yeah, so basically I need to find—I need to find that out. But good question.

BARBARA (BASIA) LOPEZ: Right. Thank you. We still have a couple minutes left. So I want to encourage everyone to continue submitting their questions. Let me see here.

LUKE MUGGY: There’s a question for me. I’m happy to answer the next.

BARBARA (BASIA) LOPEZ: Yes. Go. Yes. Yes. So, go ahead.

LUKE MUGGY: Oh, okay. So, the question is, “Do you know if a pattern of high firearm crime in other cities with well-developed mass transit or is this a more feature for Los Angeles?” We did test all—I believe all of our cities, we tested rail transit stations. And it was only Los Angeles for which this was a strong attractor. The other cities were neutral and actually, in Pittsburgh, if I remember right, the high—there’s only transit stations in the high SES area and in New Orleans, there’s only transit stations in the low SES areas. And so that was kind of interesting to see. Yeah. So in general, Los Angeles was the only attractor, the other ones were neutral or repellent.

BARBARA (BASIA) LOPEZ: All right. And I think that—so we have time for two more questions, so I have to be selective. But there is a question to Dr. Swaner. You mentioned the importance of engaging gang leaders for more effective interventions. I am wondering if you think that this is a particular feature of the New York context. And considering the fractured nature of groups in other cities such as Chicago where gang leadership no longer exists in the classic sense.

RACHEL SWANER: Right. Great question. And, yes, I would say over the last two years, especially since the pandemic but you see fracturing even within traditional gang network, so like traditional like Bloods and Crips, they are now the Woos and Choos and there are members of each gang in that, so there’s definitely throughout all gang networks around the country, there’s a change in how these networks are working together or not and in their leadership. So, it is a good question. I do not though think it is particular to New York because we do have those fracturing happening in New York as well. And we are right now conducting—almost the same team conducting a similar study around the sociocultural roots of gun violence in five cities. So, we ended up going back into Brooklyn and a different neighborhood than where we conducted this study. And then we’re doing this—the study in Detroit, New Orleans, Wilmington, Delaware and Baltimore, Maryland, and in all of those studies, we also have to have been engaging—like, we wouldn’t be able to have access to these communities and it’s another participatory research study so that the people that we are hiring are members of this gang leadership and we have been able to access these different populations through them. And that doesn’t mean that their legitimacy in their network transfers over to a different network. But we are—we are trying to kind of hire a diverse group of people or people who have the cultural knowledge as well as the, like, higher level historical legitimacy in those communities, but I will say that right now we’re up and running in Wilmington, Delaware and in Detroit and we have not found it to be an issue. We have found that the participatory nature and engaging the gang leaders including in the work that we’re doing is not an NIJ-funded study. It’s funded by the National Collaborative on Gun Violence Research through RAND.

BARBARA (BASIA) LOPEZ: All right. Thank you very much. So, we are top of the hour actually.  It’s already 2:31. So I want to thank the panelists and I want to thank the audience for the questions. If we did not answer your question, you can email the presenters directly. With this, I want to thank you on behalf of the National Institute of Justice for your participation and attention to this important topic. Please visit nij.gov website for further information about firearm violence research and other topics. With this, thank you very much and goodbye.