Speaker Abstracts & Bio Sketches: The 2nd IOF Workshop on Gun Violence: A Statistical Approach

<< Back to Event Page: The 2nd IOF Workshop on Gun Violence: A Statistical Approach

Session 1: Emerging and Evolving Data Sources for Studying Gun Violence

Susan Parker (Northeastern)

Non-fatal Firearm Injury Surveillance in the U.S.: An Update

Abstract: No comprehensive data source exists that tracks all firearm injuries in the United States. Lack of firearm data infrastructure prevents basic descriptive understanding of the causes and consequences of firearm injury and gun violence. Hospital data systems measuring firearm injuries treated in emergency departments could provide access to reliable surveillance of firearm injuries at local, state and national levels if injury intent coding could be improved. Police agencies can provide nuanced detail on criminal assaults, which account for the majority of nonfatal gun injuries, but lack comprehensive coverage. We provide an overview of ongoing reforms and necessary updates to create comprehensive nonfatal firearm injury surveillance.

Susan T. Parker is a postdoctoral research fellow at Northeastern University, affiliated with the Community to Community (C2C) public policy initiative and the Bouvé College of Health Sciences. She is a quantitative social scientist applying causal inference and machine learning methods to research in violence reduction, and in particular gun violence and policing. She holds a PhD in Health Policy with an economics cognate from the University of Michigan’s School of Public Health as well as an MS and MPP from the University of Chicago. Previously, she worked as a senior data scientist. Her research focuses on violence reduction, and in particular firearm policy.

Charles Loeffler (University of Pennsylvania)

Using recovered ammunition to study constraints in the supply chain for gun violence

Abstract:This paper provides an empirical test of the hypothesis that the illegal market for ammunition may be an example of a “thin” market where buyers and sellers have difficulty transacting. Using data from a widely available but previously unused source of information on illegal ammunition—recovered illegal firearms themselves—this analysis introduces two measures of ammunition thinness—weapon fullness and ammunition “scrounging.” Examination of these two measures reveals that ammunition is seemingly quite accessible even in jurisdictions with restrictive ammunition and firearm regulations. This finding suggests diversion of ammunition from the legal to the illegal market, like other diverted non-durable goods (e.g., cigarettes), is not as restricted as hypothesized by some models of ammunition diversion. This analysis also provides preliminary evidence for another posited illegal firearm possession behavior—carrying unloaded firearms for the sole purposive of defensive brandishing or coercive threatening. The included results provide minimal empirical support for these posited behaviors. Taken together, these results suggest that access to diverted ammunition is sub-optimally constrained by existing regulatory and enforcement actions and that most illegal possessors appear both interested and able to fill their firearms, albeit with unknown search and financial costs.

Charles Loeffler is an Associate Professor of Criminology. He holds a Ph.D., Sociology, Harvard University, which he received in 2011 and in 2002 recieved his M.Phil., Criminology, Cambridge University, and A.B., Social Studies, Harvard College, received in 2001. Professor Loeffler studies life-course criminology, the effects of criminal justice institutions, and topics at the intersection of crime and technology. He is currently conducting research on the effects of criminal record expungement and the effects of processing juveniles as adults. He has recently completed research on the spatiotemporal patterns of gun violence and the prevalence of wrongful convictions among prisoners.

Iris Horng (University of Pennsylvania)

Probabilistic Record Linkage of Two Gun Violence Data Sets

Abstract: The Gun Violence Archive (GVA) and the National Violent Death Reporting System (NVDRS) are two gun violence data sets with different strengths. While GVA provides neighborhood-specific data, it lacks demographic details of the perpetrators and victims of the crime, whereas NVDRS offers comprehensive contextual details but lacks precise location data. To enable studies to combine the strengths of both the GVA and the NVDRS, we use probabilistic record linkage to link the two data sets. We employ fastLink, an open-source record linkage package in R, to identify matching records between the datasets. Our framework achieves a 90.12% accuracy rate in linking GVA incidents with corresponding NVDRS records, serving to benefit future research on gun violence dynamics and disparities.

Co-Authors: Iris Horng, Qishuo Yin, Dylan Small

Iris Horng is a mathematics undergraduate student at the University of Pennsylvania, and she will be pursuing a PhD in Statistics and Data Science at the Wharton School at University of Pennsylvania starting in fall 2024. Her research interests include statistical methods and causal inference, as well as biomedical applications to areas such as neuroscience and public health. In the past, Iris conducted research on analyzing brain connectomes using topological data analysis. Currently, she is exploring ideas at the intersection of topological data analysis and statistics and is also engaged in an independent study on probabilistic machine learning as part of her Directed Reading Program project. She has served as the Founder and Director Bridge to Math at Penn and as the President of the Association for Women in Math.

Session 1 Chair/Discussant: Wayne Osgood (Penn State)

Dr. Wayne Osgood came to Penn State in 1996 and retired as Professor Emeritus in 2017. He is especially proud of the 16 Ph.D. students whose dissertations he supervised here, who have gone on to make valuable contributions to the field of criminology. Before joining the department of Sociology and Criminology at Penn State, he was a faculty member in the Sociology Department at the University of Nebraska, Lincoln (1987-1996), and a researcher at Boys Town, NE (1986-1987), the University of Michigan's Institute of Social Research (1980-1986), and the Behavioral Research Institute in Boulder, CO (1977-1980). Dr. Osgood's research centers on delinquency and other problem behaviors during adolescence and early adulthood. His published research addresses the contribution of routine activities to offending, friendship networks, peer influence, sources of age differences, criminal careers, and the generality of deviance. He also conducted research on a variety of programs for juvenile offenders (including prevention, diversion, and residential programs), and he has written about statistical issues for the analysis of deviant behaviors, of longitudinal data, and of program evaluations.

Session 2: Patterns in Police Use of Force

Lucas Mentch (UPitt)

Racial disparities in fatal police shootings

Abstract: Fatal police shootings in the United States continue to be a polarizing social and political issue. Clear disagreement between racial proportions of victims and nationwide racial demographics together with graphic video footage has created fertile ground for controversy. However, simple population level summary statistics fail to take into account fundamental local characteristics such as county-level racial demography, local arrest demography, and law enforcement density. In this talk, I look at fatal police shootings between January 2015 and July 2016, and implement straightforward resampling procedures designed to examine how unlikely the victim totals from each race are with respect to these local population characteristics if no racial bias were present. I present several approaches considering the shooting locations both as fixed and also as a random sample. In both cases, I find overwhelming evidence of a racial disparity in shooting victims with respect to local population demographics but substantially less disparity after accounting for local arrest demographics.

Lucas Mentch Lucas Mentch is an Associate Professor at the University of Pittsburgh in the Department of Statistics. He was previously a Postdoctoral Researcher at SAMSI & NC State University from 2015 to 2016. While at SAMSI, he worked alongside leading researchers from across the world to investigate these issues, paying particular attention to issues related to confirmation or contextual bias and also in developing quality metrics for latent pattern evidence based on new image decomposition techniques. His core research is concentrated at the intersection of statistics and machine learning. His specific research interests are in Statistical & Machine Learning, Variable Selection and Non-parametric Inference, Statistical Computing, and Applications to Crime, Law, Forensics, and Sports. He received his M.S. (2013) and Ph.D. in Statistics at Cornell University (2015).

Marie Oullet (GSU)

Bent Badges and Bullets: Unpacking Gun Violence Through Networks

Abstract: Looking at gun violence through the lens of networks shifts attention away from individual shootings to the broader patterns of relations that connect these events. How people are connected within networks shapes the spread of gun violence and how it can be interrupted. Yet the data sources used to map gun violence networks are rarely interrogated. This study unpacks the building blocks of networks used to study gun violence, distinguishing between behavioral networks and social networks. Using policing as a case study, we show that concentrating on behavioral networks restricts our view to risky relationships and masks the broader social connections that shape gun violence patterns. Together our findings suggest that focusing on one type of network can present a skewed reality of how shootings spread or stagnate."

Marie Oullet is an Assistant Professor in the Department of Criminal Justice and Criminology at Georgia State University. Her research focuses on delinquent groups, including how they emerge and evolve, and how networks structure this process. She is currently conducting a longitudinal study on police networks to better understand the informal structure of policing, including organizational cohesion and fragmentation within departments, and the consequences of these network structures on the diffusion of behaviors and attitudes. Ouellet’s work has been published in Criminology, Criminology & Public Policy, Journal of Research in Crime and Delinquency, and Justice Quarterly.

Justin Nix (Nebraska)

Open-source data quality on police shootings

Abstract: Historically, a lack of comprehensive data on police shootings has hindered a nuanced understanding of this critical issue. However, in the past decade, the research landscape rapidly evolved with the emergence of improved data facilitated by crowd-sourced, non-governmental initiatives. In this chapter, we explore the transformative potential of such data, reviewing recent findings that have significantly advanced our collective knowledge on the topic. We also emphasize the caution required when navigating this newfound wealth of information, outlining some of the inherent challenges and potential pitfalls associated with the analysis of crowdsourced data. By recounting a few cautionary tales, we emphasize the need for methodological rigor and contextual awareness in leveraging crowdsourced data to understand the complexities surrounding police shootings. In doing so, we hope to contribute to more informed public and academic dialogues on police shootings.

Justin Nix is a Distinguished Associate Professor in the School of Criminology and Criminal Justice at the University of Nebraska Omaha, where he teaches classes on policing and coordinates the Master of Arts degree program. He earned his Ph.D. from the University of South Carolina in 2015. His research interests include police legitimacy and officer decision-making. To date, Justin has authored or co-authored more than fifty peer-reviewed journal articles on these topics, as well as several book chapters, research briefs, and op-eds. He has served as a consultant to the National Policing Institute, the COPS Office, and the Department of Homeland Security. In 2019, Justin was one of four early career researchers selected by the National Institute of Justice for its LEADS (Law Enforcement Advancing Data and Science) Academics pilot program. He is also a member of the Crime and Justice Research Alliance’s expert panel, and frequently engages with local and national media on issues pertaining to policing and criminal justice.

Session 2 Chair/Discussant: Claire Kelling (Carleton College)

Claire Kelling is an Assistant Professor of Statistics at Carleton College. She has recently completed her Dual PhD in Statistics and Social Data Analytics at Penn State. Before that, she earned degrees in Statistics and Economics at Virginia Tech. Throughout her academic, professional, and extracurricular activities, she has systematically chosen positions and experiences that will allow her to develop the analytical skills, political awareness, and theoretical grounding to inform public policy. She has extensive experience in applying statistical methods to research problems that work towards social good. In her second year, she was an NSF Big Data Social Science IGERT (Integrative Graduate Education and Research Traineeship) Fellow. Her past and current research lies at the intersection of criminology, public health, public policy, spatial statistics, and computing. She analyzed and created new spatio-temporal models of crime, while also incorporating social proximity through a network approach to neighborhoods. This Big Data Social Science Fellowship has allowed me to work with statisticians, political scientists, sociologists, geographers, computer scientists, and many more.

Session 3: Causal Modeling of Gun Violence Policies

Avi Feller (Berkeley) / Eli Ben-Michael (CMU)

Statistical methods to estimate the impact of gun policy on gun violence

Abstract: Gun violence is a critical public health and safety concern in the United States. There is considerable variability in policy proposals meant to curb gun violence, ranging from increasing gun availability to deter potential assailants (e.g., concealed carry laws or arming school teachers) to restricting access to firearms (e.g., universal background checks or banning assault weapons). Many studies use state-level variation in the enactment of these policies in order to quantify their effect on gun violence. In this chapter, we discuss the policy trial emulation framework for evaluating the impact of these policies, and show how to apply this framework to estimating impacts via difference-in-differences and synthetic controls when there is staggered adoption of policies across jurisdictions, estimating the impacts of right-to-carry laws on violent crime as a case study.

Avi Feller is an associate professor at UC Berkeley working at the interface of statistics and data science and the social sciences. His research focuses on developing practical, transparent methods that can be applied at scale, and on deploying these tools in a range of policy domains, with a particular emphasis on education and criminal justice. His research has appeared in top methodological and social science journals, including Journal of the American Statistical Association, Journal of the Royal Statistical Society, Econometrica, Proceedings of the National Academy of Sciences, and Nature Human Behavior. Feller has received multiple awards for his work, including the COPSS Emerging Leader Award and the SREE Early Career Award. Feller received a PhD in statistics from Harvard University and, prior to his doctoral studies, worked as a policy official in the White House Office of Management and Budget.

Eli Ben-Michael is an assistant professor in the Department of Statistics & Data Science and the Heinz College of Information Systems and Public Policy at Carnegie Mellon University. Previously, he was a postdoctoral fellow in the Institute for Quantitative Social Science and the Department of Statistics at Harvard University. He received a PhD in Statistics from U.C. Berkeley and spent his undergraduate years at Columbia University where he received a bachelors degree in computer science and statistics. His research focuses on developing statistical and computational methods to solve practical issues in public policy and social science research. He is particularly interested in bringing together ideas from statistics, optimization, and machine learning to create methods for credible and robust causal inference and data-driven decision making.

Bijan Niknam (JHU)

Design choices that involve small undefined geographic areas

Abstract: In this talk, we describe the challenges of evaluating neighborhood-level community violence intervention programs in a major metropolitan area and some study design choices to address them. The overall approach used is a difference-in-differences design comparing crime rates in intervention and comparison areas over time. However, that “simple” design requires substantial modification for use in this context. First, the intervention areas were not clearly defined geographic areas (such as census blocks). We describe our spatial disaggregation approach to subdividing the city into groups of treated and control areas with uniformly sized hexagons, and how we chose a spatial resolution resembling city blocks. Second, the program intervention may spill over onto neighboring control areas; thus, eligible control areas must be carefully chosen to mitigate such effects as well as account for the presence of similar, concurrently operating programs elsewhere in the city. Third, we describe the strengths, limitations, and overall performance of two alternative design approaches for balancing pre-intervention gun violence trends: multivariate matching, and the augmented synthetic control method. In addition to comparing the similarity of intervention and comparison sites in the pre-intervention period, we present maps to highlight that these methods can choose substantially different control areas. Finally, we discuss key differences between matching and augmented synthetic controls that, in consideration of the intended target audience, may influence the choice of design.

Authors: Bijan A. Niknam, PhD; Michael R. Desjardins, PhD, MA; Elizabeth A. Stuart, PhD; Daniel W. Webster, ScD, MPH

Bijan Niknam is a Postdoctoral Fellow at Johns Hopkins Bloomberg School of Public Health since Jun 2023. His research involves developing causal inference tutorials for the masses and studying the effects of gun violence reduction programs for the Bloomberg American Health Initiative. He is a healthcare researcher who operates at the intersection of statistical methods and quality measurement. He uses modern matching and weighting methods to study crucial problems in healthcare delivery, and develop methods that are designed to enable policy makers to more efficiently address them. A major goal of his work is to communicate the attributes and capabilities of modern adjustment methods to the broader research community.He received his PhD in Health Policy at Harvard University in 2023; a BS in Economics, Wharton School at University of Pennsylvania in 2011; and a BA in International Studies and Business/Russian Language, Literature, and Culture at University of Pennsylvania in 2011. His adviser is Elizabeth Stuart at JHU.

Dan Lawrence / Eric Piza (CNA/SIUE/UMD)

Assessing the Impact of Gunshot Detection Technology: Methodologies, analytical approaches, and insights

Abstract: A main staple of contemporary policing has been the emphasis placed on technological solutions to crime. A particular technology that has increased in popularity is Gunshot Detection Technology (GDT), which uses acoustic sensors to detect and identify the location of gunfire events in real time and initiate police response. This presentation discusses the different types of GDT that have been used by law enforcement, the associated research examining the technology, and considerations around statistical analyses for a law enforcement technology that is implemented in phases across large areas of a city over time. The presentation will conclude with considerations for future scholarship on the topic.

Dan Lawrence is a Research Scientist at CNA. Daniel Lawrence is an expert in law enforcement technologies and analytics as well as methods to improve police-community relations. His objective is to produce high-quality empirical research using innovative approaches that are grounded in criminological theory. Lawrence's research interests include police technologies (e.g., body-worn cameras, gunshot detection technology, public surveillance systems, machine learning and artificial intelligence), police legitimacy and procedural justice, police screening and hiring practices, and community policing. At CNA, Lawrence has led and supported numerous criminal justice grants and projects for federal agencies, including the Bureau of Justice Assistance and the National Institute of Justice, as well as projects with foundation funding from Arnold Ventures and the MacArthur Foundation. Lawrence uses rigorous evidence-based approaches and quantitative and qualitative analysis to assess law enforcement agency operations and organizational reform. Before coming to CNA, Lawrence worked at the Urban Institute's Justice Policy Center and RTI International's Policing Research Program, where he led multiple evaluations of policing technologies and practices. Lawrence holds a doctorate and a Master of Arts in criminology, law, and justice from the University of Illinois at Chicago. He also holds a Bachelor of Science in criminal justice from Northeastern University, Boston. Areas of expertise include Data and Architecture Management, Domestic Safety and Security, and Plans and Strategy.

Eric L. Piza, Ph.D. is Professor of Criminology & Criminal Justice, Director of Crime Analysis Initiatives, and Co-Director of the Crime Prevention Lab at Northeastern University. His research agenda focuses on the spatial analysis of crime patterns, evidence-based policing, crime control technology, and the integration of academic research and police practice. Dr. Piza has published over 65 peer-reviewed journal articles and 3 books. In support of his research, Dr. Piza has received competitive grants from funding agencies inclusive of the National Institute of Justice, Bureau of Justice Assistance, U.S. Department of State, Swedish National Council on Crime Prevention, and the Charles Koch Foundation’s Policing & Criminal Justice reform program. Dr. Piza has been a keynote speaker at conferences and seminars organized by government agencies around the world, inclusive of the U.S. Department of Justice, London Mayor’s Office for Policing and Crime, Swedish National Council for Crime Prevention, Carabineros De Chile (The National Police Force of Chile), New York State Division of Criminal Justice Services, and Massachusetts Association of Crime Analysts. Before entering academia, Dr. Piza served as the GIS Specialist of the Newark, New Jersey Police Department, responsible for the day-to-day crime analysis and program evaluation activities of the agency. He received his Ph.D. from Rutgers University.

Session 3 Chair/Discussant: Rosanna Smart (RAND)

Rosanna Smart is a senior economist at RAND, codirector of the RAND Drug Policy Research Center, and affiliate faculty of the Pardee RAND Graduate School. Her research is in applied microeconomics, with a focus on issues related to health behaviors, illicit markets, drug policy, and the determinants of gun violence. Her current drug policy research studies a variety of issues related to better understanding substance use behaviors in the context of complex policy changes, including evaluating the implications of evolving cannabis market dynamics, and understanding trends and patterns of polysubstance use. Her other strand of research focuses on informing effective gun policy in the United States, evaluating the differential effects of gun policy across different populations and communities, and identifying interventions that can reduce gun violence; within this work, she serves as codirector of RAND's Gun Policy in America initiative. Her research has been published in outlets such as the New England Journal of Medicine, Journal of Health Economics, American Journal of Public Health, and the Proceedings of the National Academy of Sciences. Smart received her Ph.D. in economics from the University of California Los Angeles.

Day 2: Friday, May 17, 2024 1–4 pm ET

Session 4: Point Process Modeling of U.S. Gun Violence

George Mohler (BC) Andrew Holbrook (UCLA)

Changes in the reproduction number of mass shootings in the United States following the COVID-19 pandemic

Abstract: The rate of mass shootings in the U.S. has increased since the start of the COVID-19 pandemic, however the mechanisms behind the increase are not well understood. We explore the extent to which the reproduction number, or contagiousness, of mass shootings has increased since March 2020. Using a Hawkes process model for the intensity of events, we find that the reproduction number increased after March 2020. This increase is observed in three datasets maintained by the Gun Violence Archive, Mother Jones and USA Today, though the estimated amount of increase in the reproduction number varies from 0.096 to 0.337. We then discuss potential reasons for the increase in the contagiousness of mass shootings.

George Mohler is the Daniel J. Fitzgerald Professor at Boston College in the Department of Computer Science. His research interests include statistical and deep learning approaches to solving problems in spatial, urban and network data science. Several current projects he works on include modeling and causal inference for overdose and social harm event data, fairness and interpretability in criminal justice forecasting, and modeling viral processes and link formation on networks using a combination of point processes and neural networks. He received a B.S. at Indiana University - Bloomington, and a Ph.D. at University of California - Santa Barbara.

Andrew Holbrook (UCLA)

Computing hawkes processes for gun violence research

Abstract: Self-exciting point processes, or Hawkes processes, elegantly describe random conta- gion dynamics and find use in finance, viral epidemiology, social networks, seismology, wildfire science, neuroscience, and criminology. Despite this success, this large class of stochastic process models faces a significant barrier to broader uptake: it is often difficult to fit a Hawkes process model to data. Likelihood-based computations largely scale quadratically in the number of data points, and the likelihood surface is sometimes nonlinear or multimodal. New scientific applications lead to model extensions that require algorithmic tailoring. Here, we discuss likelihood-based inference procedures for Hawkes process models and review the strategies that data scientists have developed to scale these models to larger quantities of data. We also demonstrate recent high-performance computing approaches that enable the application of spatiotemporal Hawkes processes to data produced by acoustic gunshot location systems in U.S. cities. On the one hand, a single graphics processing unit (GPU) can help scale Bayesian inference to 100 thousand gunshot observations, and 40 GPUs working in concert help scale inference to 1 million observations. On the other hand, fitting a spatiotemporal Hawkes model to big gunfire data leads to new statistical questions and technical challenges.

Andrew Holbrook is Assistant Professor of Biostatistics at UCLA. Andrew received his Ph.D. working with Babak Shahbaba and Dan Gillen at UC Irvine studying statistical methods for neural decoding and the analysis of longitudinal MRI. He completed his postdoctoral training at the UCLA Department of Human Genetics, where he worked with Marc Suchard to develop high-performance computing (HPC) tools for viral phylogeography. Andrew’s research focuses on HPC strategies for big data statistical inference from high-dimensional and highly-structured statistical models. He is a recipient of both an NSF CAREER Award and an NIH K25 Award for quantitative research in biomedicine. For Andrew, the “Data Science Laboratory” is a hub for interdisciplinary cross-pollination and central to 21st century science.

Yao Xie (Georgia Tech)

Gun violence incidence modeling through non-stationary spatial-temporal self-exciting point processes

Abstract: Analysis of gun violence in the United States in the literature has utilized a variety of models based on spatiotemporal point processes. Previous studies have identified gun violence as having a contagion effect, characterized by bursts of diffusion across urban environments, which can be effectively represented using the self-excitatory spatiotemporal Hawkes process. The Hawkes process and its variants have been adept at modeling self-excitatory events, including earthquakes, disease outbreaks, financial market movements, neural activity, and the viral spread of memes on social networks. However, the existing Hawkes models applied to gun violence rely on relatively simplistic stationary kernels, failing to account for the complex, non-homogeneous spread of influence and impact over space and time. To address this limitation, we adopt a non-stationary spatiotemporal point process model that incorporates a neural network-based kernel better to represent the varied correlations among incidents of gun violence. Our study analyzes a comprehensive dataset of approximately 110,000 gunshot incidents in the Atlanta metropolitan area from 2009 to 2023. The cornerstone of our approach is the innovative non-stationary kernel, designed to increase the model's expressiveness and preserve its interpretability. This approach not only demonstrates strong predictive performance but also provides insights into the spatiotemporal dynamics of gun violence and its propagation within urban settings.

Yao Xie is a Coca-Cola Foundation Chair and Professor in the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech, which she joined in 2013 as an Assistant Professor. She also serves as Associate Director of Machine Learning and Data Science at the Center for Machine Learning. From September 2017 until March 2023 she was the Harold R. and Mary Anne Nash Early Career Professor. She was a Research Scientist at Duke University from 2012 to 2013. Her research lies at the intersection of statistics, machine learning, and optimization in providing theoretical guarantees and developing computationally efficient and statistically powerful methods for problems motivated by real-world applications. She is currently an Associate Editor for Operations Research, IEEE Transactions on Information Theory, IEEE Transactions on Signal Processing, Journal of the American Statistical Association: Theory and Methods, Sequential Analysis: Design Methods and Applications, INFORMS Journal on Data Science, and an Area Chair of NeurIPS and ICML.

Session 4 Chair/Discussant: Lingzhou Xue (Penn State)

Lingzhou Xue is a Professor of Statistics at Penn State University. He received his B.Sc. in Statistics from Peking University in 2008 and his Ph.D. in Statistics from the University of Minnesota in 2012. He was a postdoctoral research associate at Princeton University from 2012-2013. His research interests include high-dimensional statistics, nonparametric statistics, statistical and machine learning, large-scale optimization, and statistical modeling in biomedical, environmental, public health, and social sciences.

2:15pm - Session 5: Zoom Breakout Room Discussions

30 Minutes Discussion, 30 minutes Reports

Breakout Room Topics (tentitive):

Emerging and Evolving Data Sources for Studying Gun Violence
Patterns in Police Use of Force
Causal Modeling of Gun Violence Policies
Point Process Modeling of U.S. Gun Violence

After the breakout discussions, groups will report highlights.

Session 6: Miscellaneous Gun Violence Papers

David Corliss (Peace-Work)

Statistical Analysis of School Shootings in the United States as Stochastic Terrorism

Abstract: This study presents a longitudinal analysis of school shooting attacks in the United States. Raw data from the US federal CHDS school shooting database was analyzed to exclude suicides, accidents, and threatening actions not resulting in injury. Exploratory data analysis indicates two historical time periods with distinct behaviors. Time series cluster analysis indicated a change in attack behavior about the year 1991: prior to that time, attacks and fatalities were relative few, normally distributed, and show little variation from year to year. Beginning about 1991, a second type of attack pattern is found superimposed on the historical pattern, with both continuing to the present. This second pattern is found to be consistent with stochastic terrorism, characterized by a highly variable annual attack rate, marked increases in deaths and lethality (deaths per incident), and a skewed distribution with a risk of a high number of fatalities driven by a small number of extreme events. This study illustrates diagnostic characteristics of stochastic terrorism, which are use to recommend an operational definition: (1) an act of terrorism, where
the attack targets a community as a whole and the specific victims unknown to the attacker (2) the attacker is motivated by an Instigator through mass communications encouraging violence in the expectation that unspecified persons will attack the targeted group.

David Corliss is the founder and Director of Peace-Work, an all-volunteer cooperative of statisticians and data scientists applying statistical methods to issue-driven advocacy in poverty, education, social justice, and providing analytic support for charitable groups. Human trafficking research is a major initiative at Peace-Work, with studies on socio-economic analysis, risk factors, and legislative consulting. With a PhD in statistical astrophysics, Dr. Corliss leads a data science organization in the automotive industry while continuing academic research in astrophysics and taking pro bono cases as a human rights mathematician. He is the author of Stats4Good, a monthly column on statistical analysis for the greater good, which appears in Amstat News, the membership magazine of the American Statistical Association. His research interests include Statistical modeling and analysis, Demographics and socio-economic analysis, Risk factors for human trafficking, Intersectionality, and Legislation.

Jonathan Jay (Boston U)

Using machine learning and aerial imagery to enhance research on the built environment and community firearm violence

Abstract: A continually-growing body of evidence links interpersonal firearm violence to built environment factors, such as abandoned buildings, liquor stores, and green space. Improving the built environment is a promising, cost-effective direction for community-level interventions, but additional research is needed to guide programs. To advance this field, researchers need large-scale access to built environment data and useful strategies for leveraging these data. This presentation will review two studies using aerial imagery, pre-processed with convolutional neural networks, to enhance conventional analytical strategies in the spatial epidemiology of firearm violence. The first uses imagery for matching case and control locations to study the spatial influence of alcohol outlets on firearm violence. The second uses imagery to identify categories of within-city landscapes, to test effect measure modification in a difference-in-differences analysis of the effects of demolishing abandoned buildings on firearm violence. These examples will illustrate the potential for computer vision to support future work to identify and refine built environment strategies to reduce firearm violence.

Co-Authors: Jorrit de Jong, Marcia P Jimenez, Quynh Nguyen, Jason Goldstick

Jonathan Jay is an Assistant Professor of Community Health Sciences at the School of Public Health. He studies urban health, especially youth exposure to gun violence. He works at the intersection of data science and community health, focusing on relationships between urban environments and health and safety risks. He is the principal investigator of a career development award from the National Institute on Minority Health and Health Disparities (NIMHD) to study multilevel strategies for reducing racial disparities in youth firearm injuries. He also leads Shape-Up, a project using analytics to help city residents reduce firearm violence through environmental improvements (winner of the $100k Everytown for Gun Safety Prize and a 2019 Solver with MIT Solve). He was a KL2 early career scholar of the BU Clinical & Translational Sciences Initiative. Before receiving his doctorate in public health (DrPH) from the Harvard T.H. Chan School of Public Health, Dr. Jay trained as a lawyer-ethicist and worked in global health policy. He received a BA with honors from Brown University, a JD cum laude from Georgetown University Law Center and an MA in philosophy from Georgetown University.

Max Kapustin (Cornell University)

We’re Building Something Here: Reducing Gun Violence Without Exacerbating the Harms of Policing in Baltimore’s Western District

Abstract: The rate of Americans shot and killed in 2021 reached a near record high. Much of this gun violence is concentrated in a handful of disadvantaged neighborhoods in U.S. cities. We report on a focused strategy to reduce neighborhood gun violence. This strategy—which directs police and social service resources to a small number of people thought to be actively engaged in gun violence—was implemented in an area with a per capita rate of homicides and shootings among the very highest in the U.S.: Baltimore’s Western police district. In the 18 months after its introduction, we estimate that the strategy reduced the rate of people shot or killed in the Western district by roughly a third, and carjackings by almost 40%, with no evidence of spillovers to other parts of Baltimore. Importantly, despite the central role played by police in this intervention, these large reductions in gun violence were not accompanied by a broad increase in arrests. We continue to investigate the mechanisms through which this strategy affected gun violence. Preliminary evidence suggests that targeted enforcement activity may have played a key role.

Co-Authors: Aaron Chalfin, Jeremy Biddle, Cristina Layana, Brian A. Wade, Ben Struhl & Anthony A. Braga

Max Kapustin studies interventions to improve the life outcomes of disadvantaged youth and adults in U.S. cities, particularly ways to reduce their exposure to violence. Using large-scale experiments and other causal inference methods, his recent work estimates the effects of efforts such as cognitive behavioral interventions and employment for men at high risk of gun violence, mentorship for youth disengaged from school, and data-driven management changes within police departments. Earlier work of his examined the effect of housing vouchers on children’s behavior and long-term life outcomes. His research is supported by grants from the National Institutes of Health as well as Arnold Ventures. Professor Kapustin received his PhD in Economics from the University of Michigan. Prior to coming to Cornell, he was a Senior Research Director at the University of Chicago Crime and Education Labs.

Chair/Discussant: Claire Kelling (Carleton)