Dr. Gia Elise Barboza-Salerno | ISSUES | The Ohio State University

Dr. Gia Elise Barboza-Salerno

Wed, 04 Nov 2026 10:45:00 +0000

Dr. Gia Elise Barboza-Salerno is an Assistant Professor whose work bridges social work, public health, law, and statistics. Their research examines how neighborhood conditions and environmental exposures shape violence-related harm across the life course—including child maltreatment, intimate partner violence, and gun violence—with a focus on equity and structural accountability. Methodologically, their work integrates Bayesian spatial and spatiotemporal modeling, geospatial machine learning, and remote sensing to translate complex data into actionable prevention and policy.

Dr. Barboza-Salerno has published on violence prevention, environmental and spatial determinants of health, and justice-oriented applications of data science, with recent work appearing in Applied Spatial Analysis and Policy, Child Maltreatment, Journal of Interpersonal Violence, and Journal of Urban Health. They lead collaborative, community-engaged projects that center marginalized populations and emphasize place-based, trauma-informed prevention.

Their research utilizes nationally representative datasets and administrative records, quasi-experimental designs, and advanced longitudinal methods, including Growth Mixture Modeling, Parallel Process Modeling, and Bayesian space–time approaches.

Download my resumé .

Additional appointments

Center for Regional and Urban Analysis (CURA), Affiliated Faculty, The Ohio State University
Translational Data Analytics Institute (TDAI), Affiliated Faculty, The Ohio State University

Certifications

Licensed Attorney, 2018 — State of Maine
Licensed Attorney, 2020 — State of Colorado
Licensed Attorney, 2023 — State of Ohio
Licensed Attorney, 2023 — Southern District of Ohio (Federal)

Awards and honours

Tony Tripodi Prize for Research Excellence, OSU College of Social Work (2025)
6th Annual Article of the Year, International Society for the Prevention of Child Abuse & Neglect (2023)
PhD Faculty Excellence in Teaching Award, OSU College of Social Work (2023)
Instructional Technology Award, Michigan State University, Virtual University (2005)
Outstanding Teaching Award, Michigan State University, Office of Graduate Studies (2002)
GlaxoSmithKline Scholar Award for Applied Statistics, Biomedical Data Sciences (2006)
Woman’s Law Caucus Scholarship, Michigan State University
Jurisprudence Awards (Constitutional Law I; Contracts II; Legal Theory; Conflict of Laws; Property)

Modeling Firearm Injury Risk Using Spatial Random Forests

Thu, 23 Oct 2025 00:00:00 +0000

Our presentation, Modeling Firearm Injury Risk Using Spatial Random Forests, will be delivered by Dr. Gia Barboza-Salerno at the Women in Statistics and Data Science on November 12, 2025 in Cincinnati, OH.

This study introduces a spatial machine learning framework for predicting firearm injury risk across Chicago census tracts. Using Spatial Random Forest Modeling (SRFM), we integrate social, built, and natural environment indicators, such as social vulnerability, housing burden, green space, traffic volume, and tree canopy, to identify localized structural determinants of firearm violence.

Results demonstrate that racial composition, economic hardship, and environmental stressors jointly predict spatial clustering of firearm injury, while local variable importance reveals that built environment features exert context-specific effects across neighborhoods. These findings highlight the value of spatially aware, data-driven approaches for informing equitable, place-based violence prevention and policy.

Abstract: Gun violence remains a critical public health crisis, particularly in urban centers like Chicago, where firearm-related injuries are among the highest in the nation. These injuries are not randomly distributed; they are spatially concentrated in neighborhoods shaped by structural inequality, including racial segregation, economic disinvestment, and environmental neglect. While random forest models have shown promise in predicting firearm violence and identifying neighborhood-level risk factors, conventional approaches often ignore spatial dependence, leading to biased estimates, inflated variable importance, and reduced geographic generalizability. This study applies spatially explicit machine learning to model non-fatal shootings across Chicago census tracts from 2020 to 2024. A curated set of predictors was compiled, including measures of social and environmental vulnerability, the built environment, and housing inequality. Three research questions guided the analysis: (1) Does incorporating spatial structure via Moran’s Eigenvector Maps (MEMs) in spatial random forests and spatially structured random effects in a Bayesian spatial model improve predictive accuracy compared to non-spatial models? (2) Which neighborhoods exhibit elevated firearm violence risk due to unmeasured spatial processes, as captured by spatial random effects? (3) Does prediction error vary across geographic space, revealing limits to model transferability? Findings suggest that the spatial random forest model outperformed the non-spatial model (pseudo R² = 0.98 vs. test R² = 0.50). Spatial cross-validation revealed geographic variation in accuracy (median nRMSE = 0.76), while spatial random effects from a Bayesian Poisson model identified unexplained latent risk. The enhanced predictive accuracy reveals hidden geographic risk patterns, suggesting that social, structural, and environmental factors can efficiently inform targeted violence prevention strategies.

Download the presentation here .

No Bright Line: How AI-driven Technology Blurs Legal Boundaries Between Prevention and Punishment in Family Regulation

Wed, 22 Oct 2025 00:00:00 +0000

This new article by Dr. Gia E. Barboza-Salerno argues that Predictive Risk Models (PRMs) can be effective when focused on prevention. The argument is grounded in two core principles. The first is a rights-based balancing approach that extends beyond privacy concerns to encompass due process protections tied to legally recognized status.
The second emphasizes that PRMs, rather than serving solely as detection or risk assessment tools, can help identify structural risk factors, guide equitable public investment, and promote accountability, without expanding surveillance or punitive measures that compound systemic harm.

Together, these principles challenge the false dichotomy between prevention and punishment that underlies much of AI’s deployment in family regulation. By integrating a legal and ethical framework rooted in rights protection and structural accountability, the article redefines what responsible and justice-oriented AI looks like in the context of family welfare systems.

Read the full article here: “Barboza-Salerno, G. E. (2025). No bright line: How AI-driven technology blurs legal boundaries between prevention and punishment in family regulation. Family Advocate, 47(2), 00–00. American Bar Association.”

Beyond Urban–Rural Binaries: Remote Sensing Indicators of Built Environment Risk for Firearm Injury in Cook County, Illinois

Wed, 22 Oct 2025 00:00:00 +0000

Presented at the 2025 National Research Conference for the Prevention of Firearm-Related Harms, this study develops a remote-sensing–based framework to identify built environment risk factors for firearm injury in Cook County, Illinois. Using satellite-derived indicators of nighttime light intensity, impervious surface coverage, vegetation (NDVI), and built-up density merged with firearm injury data from 2020–2024, we fit spatial random forest models and Bayesian spatial Poisson models to quantify environmental predictors of firearm harm.

Findings show that built environment characteristics—rather than traditional urban–rural categories—more accurately distinguish high-risk census tracts. High built-up intensity, fragmented greenness, and asymmetric nighttime activity remain strong predictors even after accounting for area-level deprivation. These results demonstrate the utility of remote sensing for identifying structural vulnerability and guiding place-based firearm injury prevention.

📄 Presentation

Download the presentation here .

GIS Day 2025

Wed, 22 Oct 2025 00:00:00 +0000

Congratulations to our undergraduates Oliva McLucas and Malcolm McCarthy for their abstracts accepted to the 2025 GIS Day at The Ohio State University on November 19, 2025.

Olivia McLucas: Introducing a Spatial Adverse Childhood Environments Index using Geographical Information Systems We extend the cumulative Adverse Childhood Experiences (ACEs) framework to the neighborhood scale with spACEs (Spatial Adverse Childhood Environments), an index of structural disadvantage in childhood. The spACES index is constructed from county-level measures in the 1990 Census, including poverty, educational attainment, employment structure, housing burden, single-parent prevalence, and racial residential segregation, which reflect the environments in which the Opportunity Atlas birth cohorts (children born 1978 to 1985) were raised. The Opportunity Atlas links childhood places of residence to adult outcomes, capturing household income rank, educational attainment, incarceration, and family formation (2010-2015), as well as credit outcomes (2020). Using GIS, we visualize geographic variation in spACEs and assess its association with long-term adult outcomes. We propose spACEs as a revised measure of childhood structural adversity that captures the enduring impact of neighborhood disadvantage across the life course.

McCarthy, M.: Mapping Traumatic Brain Injury using Geographic Information Systems Traumatic brain injury (TBI) is a critical public health problem linked to numerous mechanisms, including unintentional injuries such as falls and motor vehicle crashes and intentional injuries such as assaults. Despite the magnitude of this burden, no research has systematically examined the geographic distribution of TBI hospitalizations and their county-level correlates. We utilize county-level hospitalization data from the Washington State Department of Health, filtered to include all TBI-related admissions by year. These records are linked to the Area Deprivation Index (ADI) for the period 2016–2023, encompassing domains of economic hardship and inequality, educational attainment, and financial strength. The merged dataset enables the analysis of spatial and temporal patterns of TBI hospitalization risk and tests whether county-level social and economic disadvantage is associated with higher rates of TBI hospitalization. Using GIS visualization and regression models, we demonstrate how community structural conditions shape the distribution of TBI and inform strategies for prevention.

The presentations will be made available after the conference.

Enhancing Brain Injury Surveillance Using Large Language Models and Topic Modeling

Sun, 19 Oct 2025 00:00:00 +0000

Our presentation, Enhancing Brain Injury Surveillance Using Large Language Models and Topic Modeling, was delivered by Dr. Shockley McCarthy at the AI for BI Conference at The Ohio State University on October 17, 2025.

This work introduces an AI-driven framework for improving the detection and contextual understanding of brain injury cases using unstructured medical and administrative text. By applying large language models (LLMs) and structured topic modeling, we extract interpretable patterns from injury narratives and ICD-coded data to identify mechanisms of injury, intent, and underreported cases of intimate partner violence (IPV)–related brain injury.

Integrating methods from public health, law, and computational linguistics, our approach aims to enhance the precision of injury surveillance systems and inform equitable prevention and policy strategies addressing violence-related traumatic brain injury (TBI).

Download the presentation here .

Dr. Barboza-Salerno Addresses Grand Challenge to Prevent Gun Violence

Tue, 06 Jan 2026 00:00:00 +0000

Dr. Gia Barboza-Salerno addresses the Grand Challenge to Prevent Gun Violence.

An assessment of opioid overdose mortality risk across the urban–rural continuum: Integrating satellite-derived and socioeconomic indicators

Wed, 04 Nov 2026 10:45:00 +0000

Session: Built environment and community-based approaches to advancing health equity
Program: Public Health Education and Health Promotion
Date: Wednesday, November 4, 2026 | 10:45 AM – 11:00 AM

Disability-related depression inequities are driven by accessibility deficits: Evidence from crowdsourced sidewalk data and machine learning

Tue, 03 Nov 2026 14:45:00 +0000

Session: Mobility, Transit, and Built Environments for Disabled People
Program: Disability
Date: Monday, November 2, 2026 | 2:45 PM – 3:00 PM

Extreme heat and suicide risk among older adults: A longitudinal time-series analysis in an urban U.S. setting

Mon, 02 Nov 2026 12:30:00 +0000

Session: Climate, Culture, and Life-Course Contexts in Mental Health
Program: Mental Health
Date: Monday, November 2, 2026 | 12:30 PM – 1:30 PM

Improving surveillance of IPV-related head and neck injuries using AI-enhanced ED narratives

Mon, 02 Nov 2026 10:30:00 +0000

Session: Injury Epidemiology, Trauma Systems, and Health Care Utilization Posters
Program: Injury Control and Emergency Health Services
Date: Monday, November 2, 2026 | 10:30 AM – 11:30 AM

Identifying place-specific mortality mechanisms by integrating cause-of-death narratives and neighborhood typologies using machine learning

Mon, 02 Nov 2026 09:15:00 +0000

Session: Epidemiology Section: Conceptualization and Measurement
Program: Epidemiology
Date: Monday, November 2, 2026 | 9:15 AM – 9:30 AM

Predicting harsh school discipline in Albuquerque Public Schools: A machine learning analysis with temporal validation

Mon, 02 Nov 2026 08:30:00 +0000

Session: Public Health Social Work Perspectives on Health and Well-Being Across the Lifespan
Program: Public Health Social Work
Date: Monday, November 2, 2026 | 8:30 AM – 10:00 AM

New Publication: Profiles of Stressful Life Experiences in Pregnancy

Wed, 02 Sep 2026 00:00:00 +0000

We are excited to announce a new publication from the ISSUES Lab!

Profiles of stressful life experiences in pregnancy: a latent class approach linking early childhood adversity, intimate partner coercive control, and material hardship to adverse birth outcomes

Gia Elise Barboza-Salerno & Taylor Harrington

Published August 31, 2026 in BMC Pregnancy and Childbirth (Open Access)

Abstract

Acute antenatal stressful life events (SLEs) during pregnancy are well-established risk factors for adverse birth outcomes. However, little is known about how distinct constellations of antenatal SLEs, shaped by early adversity, material hardship, and intimate partner coercive control, combine to influence perinatal health.

Data were drawn from 5,280 participants in the Kansas Pregnancy Risk Assessment Monitoring System. Using latent class analysis, five distinct stress profiles were identified:

Family Member Illness and Death (16.3%)
Multiple Overlapping Stressors (5.2%)
Financial Insecurity (11.2%)
Residential Instability and Family Conflict (8.7%)
Low Stress (58.6%)

Compared with the Low Stress class, individuals in all high-risk classes had significantly higher odds of reporting ACEs, IPV, material hardship, and cigarette use before pregnancy. The Multiple Overlapping Stressors class showed the greatest risk of low birth weight, small for gestational age, and preterm birth.

Key Takeaway

Tailoring perinatal interventions to specific SLE profiles, particularly for individuals with histories of ACEs and IPV, may reduce poor birth outcomes.

Download PDF | View Publication

Gia's Journey through Oceania and the 2026 ISPCAN Conference in Melbourne, Australia

Mon, 31 Aug 2026 00:00:00 +0000

Dr. Gia Barboza-Salerno recently traveled through Oceania and presented at the 2026 ISPCAN World Congress on Child Abuse and Neglect in Melbourne, Australia (August 24–26, 2026).

Explore their full journey, including stops along the way, conference highlights, and reflections on the trip in the interactive feature below.

View Gia’s Journey through Oceania

Profiles of stressful life experiences in pregnancy: a latent class approach linking early childhood adversity, intimate partner coercive control, and material hardship to adverse birth outcomes

Mon, 31 Aug 2026 00:00:00 +0000

Detecting Hidden Mechanisms of Abuse-Related Injury in Children: Structured Topic Modeling of Clinical Death Reviews and Legal Opinions

Mon, 24 Aug 2026 00:00:00 +0000

Presented at the ISPCAN World Congress on Child Abuse and Neglect, Melbourne, Australia, August 2426, 2026.

Full title: Detecting Hidden Mechanisms of Abuse-Related Injury in Children: Structured Topic Modeling of Clinical Death Reviews and Legal Opinions Triangulating Multidisciplinary Child Fatality Reviews with Factual and Procedural Narratives from Case Law

Abstract

Existing child maltreatment surveillance systems miss or misclassify between 8090% of maltreatment deaths, and CPS records undercount fatalities by 5576%. Unstructured narrative text clinical death reviews, court opinions, and social work records accounts for 8090% of all organizational data produced, yet it is rarely leveraged for injury surveillance.

This presentation introduces a two-study AI + Structured Topic Modeling (STM) pipeline that integrates two distinct narrative corpora to recover latent, abuse-related injury mechanisms that structured administrative codes routinely obscure.

Study Design

	Study 1: CFRT Death Reviews	Study 2: CourtListener Case Law
Source	Colorado Child Fatality Review Team reports	Appellate opinions retrieved from CourtListener
Documents	223 child-level narrative records	751 opinions → 335 confirmed cases from Ohio
Text	Two free-text fields per report (clinical + administrative)	Facts + procedural history
LLM Extraction	GPT-4o-mini: caregiver behavior, injury mechanism, setting, diagnosis	GPT-4o-mini: 9 structured fields including perpetrator, victim type, legal outcome
STM Models	Risk & Contributing Factors; Circumstances	Injury Mechanism; Long-Term Consequences
Covariate	Incident year (temporal prevalence)	Judicial ruling (affirmed / reversed)

Research Questions

Can large language models transform unstructured child fatality narratives into standardized semantic features caregiver behavior, injury mechanism, setting, diagnosis?
What latent themes emerge, and what typology of risk and incident circumstances do they suggest?
Do the recovered themes reveal recurring patterns of head and neck trauma consistent with abuse-related injury?
How does the prevalence of these themes shift across incident years (20202025)?

Key Findings

Both corpora converge on craniofacial injury as the dominant mechanism routinely undercounted by ICD codes.
Strangulation emerged as a critical lethality predictor; soft-tissue damage can surface 23 days post-injury, requiring clinician and court training on delayed sequelae.
Records document long-term neurological harm (cerebral palsy, pituitary damage, blindness) in children often years before clinical follow-up.
Children represent a structurally distinct injury profile in both datasets, warranting specialized clinical and legal protocols.
“Accidental” injuries (falls, environmental contact) are frequently embedded in an abuse context screening for mechanism, not just diagnosis, is essential.
Appellate opinions and death reviews function as injury-surveillance sources beyond structured administrative fields.

Methodology

The pipeline applies:

PDF parsing (pdftools) and regex isolation of narrative fields
GPT-4o-mini for LLM-based structured field extraction
Structured Topic Modeling (STM) with incident year or judicial ruling as a prevalence covariate
BART zero-shot classification as an independent supervised-style validation check
FREX word scoring (frequency � exclusivity) for topic labeling

Lab Members

This presentation was developed by the ISSUES Lab team at The Ohio State University:

Dr. Gia E. Barboza-Salerno (Founding Director), Dr. Henriikka Weir, Malcolm McCarthy, Dr. Karla Shockley McCarthy, Dr. Amy Watson-Grace, Dr. Keith Warren, Esenia Cassidy, and Olivia McLucas with collaborators Scottye Cash and Suzanne Perkins.

📄 Presentation

Download the presentation slides here .

Congratulations to Keith Warren and Gia Barboza-Salerno for Completing the Agent-Based Modeling Certificate

Sun, 09 Aug 2026 00:00:00 +0000

We are proud to announce that Keith Warren and Gia Barboza-Salerno have successfully completed the Santa Fe Institute’s Introduction to Agent-Based Modeling certificate course through abm.complexityexplorer.org! 🎓

Agent-Based Modeling is a powerful computational method for simulating the actions and interactions of autonomous agents including individuals, households, and organizations in order to understand how macro-level patterns emerge from micro-level behaviors. ABM is particularly well-suited for studying complex social systems such as neighborhood dynamics, health behavior spread, child welfare intervention cascades, and community-level resilience.

The addition of ABM to the ISSUES Lab’s methodological toolkit opens exciting new avenues for our research on structural inequality, environmental justice, and social determinants of health.

Congratulations to Keith and Gia on this outstanding achievement! We look forward to seeing how these new computational skills enrich the lab’s work.

Congratulations to Malcolm McCarthy for Presenting at the American Psychological Association 2026 Annual Convention

Fri, 07 Aug 2026 00:00:00 +0000

We are thrilled to congratulate Malcolm McCarthy, an undergraduate psychology student in the ISSUES Lab, for presenting his first-authored research poster at the American Psychological Association (APA) 2026 Annual Convention, held August 6-8, 2026 at the Walter E. Washington Convention Center in Washington, DC!

Malcolm’s poster, titled:

“How Multiple Environmental Factors Influence Mental Health Outcomes”

explores how neighborhood-level environmental conditions including greenspace exposure and suicide death concentrations shape mental health and suicide risk across census tracts in Cook County, Illinois. The study integrates spatial data and statistical modeling to disentangle the complex relationships between depression burden, environmental predictors, and suicide mortality.

Key Findings

Neighborhoods with higher depression burden were not the same neighborhoods with higher suicide mortality, highlighting that these are distinct spatial phenomena.
Higher greenspace predicted greater depression while also being associated with reduced suicide incidence.
Walkability and greenspace together influenced the co-location of depression and suicidality.
Suicide locations were closer to greenspace boundaries than random control points, and suicide risk increased within 1 km of nearby greenspace.

Policy Implications

The research points to actionable environmental design strategies such as accessible greenspace, improved walkability, and targeted community support services as levers for reducing mental health disparities in urban areas.

Co-authored with Dr. Gia Barboza-Salerno and Dr. Karla Shockley-McCarthy, Malcolm’s presentation was part of the Society for the Psychological Study of Social Issues (SPSSI) poster session (Division 9).

📄 Download the Poster

This is an outstanding achievement for an undergraduate researcher. Malcolm’s presentation reflects the caliber of work being done in the ISSUES Lab and his dedication to evidence-based science in service of community wellbeing. We are incredibly proud of you, Malcolm!

What 25 Years of Jury Selection Data Reveal About the Questions Raised by the Karmelo Anthony Trial

Fri, 10 Jul 2026 00:00:00 +0000

In Batson v. Kentucky (1986), the Supreme Court held that prosecutors may not exclude potential jurors solely because of their race. Nearly four decades later, questions about racial disparities in jury selection continue to arise. They resurfaced during the 2026 murder trial of Karmelo Anthony, where the defense raised a Batson challenge after all eligible Black jurors were removed from the pool and an all-White jury was ultimately seated.

The Anthony case cannot be evaluated using the Mississippi data analyzed here. It occurred in a different state, a different county, and under different circumstances, and nothing below should be read as a claim about it. However, the case does bring up an important question that the criminal legal system should confront: if race-based juror exclusion is unconstitutional, do measurable racial disparities in jury selection still persist? To examine that question, I reanalyzed a publicly available dataset assembled by APM Reports containing jury selection records from 305 criminal trials in Mississippi’s Fifth Circuit Court District spanning roughly 1991–2017.

Five Key Findings

Racial disparity in prosecution strikes persisted after Batson. Black jurors remained roughly 37–38 percentage points more likely than non-Black jurors to be struck by prosecutors, even as overall strike rates declined.
All-White juries occurred far more often than a race-neutral process would predict. Across 218 trials with complete jury-composition data, 14 juries contained no Black members. Under race-neutral selection, the model predicts roughly 1.3.
The disparity was robust across statistical approaches. The odds of a Black eligible juror being struck are nearly eight times those of non-Black jurors, and the gap persisted after adjusting for county, defendant race, and trial-level clustering.
The disparity persisted even when defendants were White. The pattern cannot be explained solely as an effort to match juror demographics to defendant demographics.
Every county examined showed a statistically significant disparity. The magnitude varied; no county showed the gap disappearing.

These are descriptions of patterns in a particular, non-random sample of trials. They are not, and should not be read as, determinations of intent in any individual decision, nor as population estimates for trials in general.

Overview

In 2018, reporters working on the podcast In the Dark published a detailed statistical investigation of jury selection in Mississippi’s Fifth Circuit Court District. Their analysis documented that, across hundreds of criminal trials in seven counties, prosecutors used their discretionary “peremptory” challenges to remove Black prospective jurors at rates far exceeding those for white jurors. That investigation is available at: features.apmreports.org/in-the-dark/mississippi-district-attorney-striking-black-jurors/

This article independently replicates the findings reported by APM Reports using publicly available jury selection data and extends the analysis through multilevel modeling, county-level decomposition, time-trend analyses, and probability-based estimates of how often the observed outcomes would be expected under race-neutral jury selection. The article concludes by placing the 2026 murder trial of Karmelo Anthony in a broader historical context. While the Mississippi data cannot determine what occurred in that specific case, the questions it raised about Batson challenges, race-neutral explanations for juror strikes, and the seating of an all-White jury are consistent with these historical patterns.

Throughout, the analysis is descriptive and statistical. Racial disparities in aggregate data are documented as patterns; they are not, and should not, be read as determinations of intent in any individual decision.

About the Dataset

The APM Reports jury dataset and analysis are available at github.com/APM-Reports/jury-data. It covers 305 criminal trials across seven Mississippi counties that are within the Fifth Circuit Court District. The trial coverage is roughly 1991 to 2017, although not all data are complete. All cases were ultimately appealed, creating selection bias that makes the sample non-random and may have analytic implications. Nevertheless, anyone can easily replicate the analysis below using the code shown in each code block.

For each trial, the dataset records the defendant’s race, charges, judge, and prosecuting attorneys. For each of the 14,874 individual prospective jurors, it records race, gender, what happened during selection (struck for cause, struck by state, struck by defense, or seated), and a strike_eligibility field indicating which side retained peremptory power at that juror’s position in the process.

jurors_clean <- jurors |>
 mutate(
 race_clean = str_to_lower(str_trim(race)),
 struck_by_clean = str_to_lower(str_trim(struck_by)),
 race_group = case_when(
 race_clean == "black" ~ "Black",
 race_clean %in% c("white", "asian", "latino") ~ "Non-Black",
 TRUE ~ NA_character_
 ),
 state_struck = if_else(struck_by_clean == "struck by the state", 1L, 0L)
 )

jury_df <- jurors_clean |>
 left_join(
 trials |> select(id, county, defendant_race, judge,
 prosecutor_1, prosecutor_2, prosecutor_3,
 verdict, cause_number, batson_claim_by_defense),
 by = c("trial_id" = "id")
 )

article_universe <- jurors_clean |>
 filter(
 race_clean %in% c("black", "white"),
 !struck_by %in% c("Juror not struck", "Struck without notation",
 "Juror excused/absent", "Unknown")
 )

eligible_df <- jury_df |>
 filter(
 !is.na(race_group),
 strike_eligibility %in% c("Both State and Defense", "State")
 ) |>
 mutate(
 race_group = factor(race_group, levels = c("Non-Black", "Black")),
 defendant_race = as.factor(defendant_race),
 county = as.factor(county)
 )

Below, “Non-Black” combines white, Asian, and Latino jurors, who together are overwhelmingly white in these counties; I use “non-Black” when the comparison is the analytic contrast, and “white” only when the underlying group is effectively all White (for example, when describing all-White juries).

Replication of Headline Statistics

APM Reports described a pool of 6,763 prospective jurors. I identify the equivalent group as all jurors recorded as Black or White who received a definitive outcome, excluding ambiguous categories (“Juror not struck,” “Struck without notation,” and similar).

sfc <- article_universe |> filter(struck_by == "Struck for cause")
after_cause <- article_universe |> filter(struck_by != "Struck for cause")
prose_str <- article_universe |> filter(struck_by == "Struck by the state")
def_str <- article_universe |> filter(struck_by == "Struck by the defense")
served <- article_universe |>
 filter(struck_by %in% c("Juror chosen to serve on jury", "Juror chosen as alternate"))
after_pros <- article_universe |>
 filter(struck_by %in% c("Struck by the defense",
 "Juror chosen to serve on jury",
 "Juror chosen as alternate"))

tibble(
 Stage = c("Total prospective pool", "Struck for cause",
 "Pool entering peremptory", "Prosecution strikes",
 "Pool after prosecution", "Defense strikes",
 "Seated (jurors + alternates)"),
 `My N` = c(nrow(article_universe), nrow(sfc), nrow(after_cause),
 nrow(prose_str), nrow(after_pros), nrow(def_str), nrow(served)),
 `Article N` = c("6,763","1,342","~5,421","1,275","-","1,458","2,688"),
 `My % Black` = c(
 percent(mean(article_universe$race_clean == "black"), 0.1),
 percent(mean(sfc$race_clean == "black"), 0.1),
 percent(mean(after_cause$race_clean == "black"), 0.1),
 percent(mean(prose_str$race_clean == "black"), 0.1),
 percent(mean(after_pros$race_clean == "black"), 0.1),
 percent(mean(def_str$race_clean == "black"), 0.1),
 percent(mean(served$race_clean == "black"), 0.1)
 ),
 `Article % Black` = c("39%","57%","35%","71%","21%","9%","32%")
)

# A tibble: 7 × 5
 Stage `My N` `Article N` `My % Black` `Article % Black`
 <chr> <int> <chr> <chr> <chr>
1 Total prospective pool 6795 6,763 39.1% 39%
2 Struck for cause 1342 1,342 56.6% 57%
3 Pool entering peremptory 5453 ~5,421 34.8% 35%
4 Prosecution strikes 1285 1,275 70.9% 71%
5 Pool after prosecution 4168 - 23.7% 21%
6 Defense strikes 1466 1,458 8.7% 9%
7 Seated (jurors + alternates) 2702 2,688 31.9% 32%

All percentage figures replicate to within rounding error. The point of reproducing the previous analysis is to verify before building on it.

The Flow of Jurors

The prosecution strike flow shows that 70.9% of state strikes removed Black jurors. The defense flow runs the other way, 91.3% white, but only because, by that stage, the pool was already heavily white. Together, they show how Black jurors entered the process at roughly their expected share (39.1%) yet were “funneled” away from selection at specific stages of voir dire.

Statistical Analysis

Overall Strike Rates

Among jurors eligible for peremptory challenges, 49.8% of Black jurors were struck by the prosecution, compared with 11.2% of non-Black jurors. The unadjusted odds ratio is 7.84 (95% CI: 6.81–9.04). Adjusting for defendant race and county yields an OR of 8.25, indicating that the disparity persists after accounting for these factors.

# A tibble: 2 × 5
 race_group n struck_n strike_rate pct
 <fct> <int> <int> <dbl> <chr>
1 Non-Black 3320 373 0.112 11.2%
2 Black 1811 902 0.498 49.8%

County-level Variation

The disparity appears across all counties, with odds ratios ranging from 4.6 (Choctaw) to 17.3 (Carroll). In Carroll County, an eligible Black juror was struck at roughly 17 times the odds of a non-Black juror. All county-level estimates are statistically significant (p < 0.001).

# A tibble: 7 × 5
# Groups: county [7]
 county OR lower upper p.value
 <fct> <dbl> <dbl> <dbl> <dbl>
1 Carroll 17.3 8.09 40.8 3.19e-12
2 Montgomery 12.6 8.62 18.7 9.82e-38
3 Webster 9.97 3.94 26.9 2.21e- 6
4 Grenada 9.21 7.10 12.0 5.93e-61
5 Winston 8.14 5.35 12.6 7.18e-22
6 Attala 6.14 4.64 8.17 2.94e-36
7 Choctaw 4.60 2.92 7.26 4.40e-11

Does Defendant’s Race Matter?

When the defendant is Black, 52.6% of Black jurors are struck versus 9.7% of non-Black jurors. In White-defendant cases, the absolute rates differ (Black: 40.0%, non-Black: 16.2%), but the gap persists. The interaction term (OR = 0.34, p < 0.001) indicates that the disparity is statistically smaller in White-defendant cases but not eliminated.

# A tibble: 4 × 5
 term estimate conf.low conf.high p.value
 <chr> <dbl> <dbl> <dbl> <dbl>
1 (Intercept) 0.108 0.0946 0.123 4.59e-244
2 race_groupBlack 10.3 8.69 12.2 4.63e-162
3 defendant_raceWhite 1.79 1.40 2.27 2.10e- 6
4 race_groupBlack:defendant_raceWhite 0.336 0.242 0.468 9.48e- 11

Time Trend, 1992–2012

Over the span with reasonably complete data, both strike rates drifted downward together, and the gap between them barely moved. As shown by the smoothed (LOESS) trends, the Black-juror strike rate declined modestly while the non-Black rate stayed roughly flat, leaving a gap that began near 38 percentage points and ended near 32, averaging about 37 points across the period. The stability of the gap is the primary takeaway.

County-level Poverty

Six of the seven counties have poverty rates between roughly 22% and 26%, near the Mississippi median (~23%). Carroll County has the lowest poverty rate (~14%) yet the highest odds ratio (17.3). With only seven counties, this cannot support an inferential claim, but there is no visible tendency for poorer counties to show larger disparities, so local economic conditions are not an obvious explanation for the pattern.

Multilevel Model: Accounting for Clustering

Standard logistic regression treats each juror as independent. In reality, jurors in the same trial share a common local context, such as the same prosecutor, judge, defendant, and case facts, creating within-trial correlation. Ignoring this correlation produces artificially narrow confidence intervals. A mixed-effects (multilevel) logistic model accounts for this by estimating a random intercept for each trial.

The multilevel model yields an OR of 8.55 (95% CI: 7.34–9.95), compared with 7.84 from the simple logistic model. The ICC of 4.5% indicates that approximately 4.5% of the variation in strike decisions reflects between-trial differences, although the majority of the variation is within trials.

# A tibble: 3 × 4
 Model OR `95% CI` Notes
 <chr> <dbl> <chr> <chr>
1 Simple logistic (no clustering) 7.84 6.81–9.04 Assumes jurors …
2 Multilevel logistic (jurors nested in trials) 8.55 7.34–9.95 ICC = 4.5%
3 Multilevel logistic (+ defendant race) 8.54 7.33–9.95 Adjusts for def…

The Mathematics of Exclusion

How Likely are All-White Juries?

After strikes for cause, 34.8% of the eligible pool in this district was Black. The question is how often an all-White twelve-person jury should appear under race-neutral selection, and how it changes after prosecution strikes are applied.

Under a reasonable approximation, the chance of seating a twelve-person jury with no Black members from a pool that is 34.8% Black is about 0.6%, roughly 1 in 170. After applying the observed prosecution strike rates, the effective share of Black jurors available at seating falls to about 23.7%, and the probability of an all-White jury rises to about 3.9%, roughly 1 in 26.

All else equal, the observed strike behavior is associated with an all-White jury becoming about 6.6 times more likely than the race-neutral benchmark.

Across Many Trials

Small per-trial differences compound across many trials. The table below shows the probability of observing at least one all-White jury under the race-neutral benchmark and under the observed strike rates.

Number of Trials	Race-Neutral Selection	Observed Strike Rates
10	5.7%	32.7%
25	13.7%	62.8%
50	25.4%	86.1%
100	44.4%	98.1%
500	94.7%	100.0%
1,000	99.7%	100.0%

Observed Versus Expected

Across 218 trials with complete jury-composition data, 14 juries contained no Black members. Under the race-neutral benchmark, the expected count is about 1.3. Therefore, the observed number is roughly 11 times the benchmark. An exact binomial test rejects a race-neutral process at p < .001.

What These Data Can—and Cannot—Tell Us

Every trial here was appealed, and appeal is not random, so the rates (the 49.8% strike rate, the “1 in 26” probability, the per-trial figures) describe this non-random set of trials, not jury selection in the district at large or elsewhere. But selection bias cannot easily explain the disparity’s direction, consistency, and magnitude: an eightfold odds ratio that survives adjustment for county and defendant race, appears in every county at p < 0.001, and holds even when the defendant is White is not what a selection artifact looks like.

The data still cannot establish intent by any prosecutor or why any particular juror was struck; those turn on voir dire responses and case-specific reasons absent here. What it shows is an aggregate disparity: prosecutors struck Black eligible jurors at far higher rates than non-Black jurors, and 14 of 218 trials produced all-White juries, where a binomial model based on each county’s Black share predicts only about 1.3. That does not prove discrimination, but it sharpens the question the Anthony trial revived: if race-based exclusion is unconstitutional, why do disparities this large persist decades after Batson?

Transparency, Data, and the Future of Jury Selection Analysis

Debates about jury selection tend to run on speculation and competing narratives precisely because the underlying records are rarely available for independent evaluation. When data do exist, analyzing them is not straightforward, and the results must be read within an appropriate legal and statistical frame, including an unbiased account of how the data were chosen, cleaned, and analyzed.

Expanding access to court records, voir dire transcripts, and structured jury-selection datasets would let researchers, journalists, attorneys, and the public assess claims with evidence rather than assumption, and would let the kind of selection problem flagged above be measured rather than merely conceded.

Winner of the Urban Science Best Paper Award: Uncovering Spatial Patterns of Residential Settlements, Segregation, and Vulnerability of Urban Seniors

Fri, 10 Jul 2026 00:00:00 +0000

We are pleased to announce that our paper, “Uncovering Spatial Patterns of Residential Settlements, Segregation, and Vulnerability of Urban Seniors Using Geospatial Analytics and Modeling Techniques”, published in Urban Science (July 2024), has been selected as the winner of the Urban Science Best Paper Award.

The award was determined by the committee based on an overall evaluation of the originality and significance of the papers, considering citations and downloads. Congratulations to the author team!

Racial and Ethnic Disparities in Non-Fatal Self-Inflicted Firearm Injuries Among Adults in the United States, 2000-2021

Wed, 17 Jun 2026 00:00:00 +0000

Plain-Language Summary

This study analyzes nonfatal, self-inflicted firearm injuries in the United States over a 22-year period (2000-2021) using the National Electronic Injury Surveillance System (NEISS-FISS). Because structured database codes often fail to capture the context or intent behind a self-inflicted discharge, the research team used text mining and a machine learning (LASSO) classification model to analyze the unstructured clinical narratives written by emergency department staff.

The results show that while 58% of visits were likely unintentional discharges (accidents), 10% were probable suicide attempts. The narrative text mining revealed distinct, racially disparate patterns of injury: Non-Hispanic Black patients suffered the highest proportion of lower-extremity injuries (51%), while Hispanic patients had the highest rates of craniofacial injuries (26%) and a significant association with alcohol involvement (increasing from 12% in coded data to 20% in narrative data). This highlights the value of using natural language processing to analyze emergency department notes for better injury prevention and public health surveillance.

New Publication: Racial and Ethnic Disparities in Non-Fatal Self-Inflicted Firearm Injuries

Mon, 15 Jun 2026 00:00:00 +0000

Abstract

Nonfatal self-inflicted firearm injuries represent an important and understudied public health concern in the United States. We analyzed nationally representative data from the National Electronic Injury Surveillance System–Firearm Injury Surveillance Study (NEISS-FISS), 2000–2021, to characterize nonfatal self-inflicted firearm injuries among U.S. adults aged 18–64 years. Emergency department (ED) narratives were combined using text mining and an elastic net (LASSO) classification model to augment structured surveillance data. Among 7,362 unweighted cases (representing approximately 260,602 ED visits, 58% were classified as probable unintentional discharges, whereas approximately 10% were probable suicide attempts. Among cases with complete race and ethnicity data, Non-Hispanic White (NHW; 69.2%), Non-Hispanic Black (NHB; 23.2%), and Hispanic (7.6%) individuals comprised the primary analytic groups. Narrative analysis substantially increased detection of contextual factors, particularly alcohol involvement among Hispanic patients (12.0% to 20.1%). NHB patients had the highest proportion of lower-extremity injuries (51%), while Hispanic patients had the highest prevalence of craniofacial injuries (26%). The LASSO model effectively distinguished terms associated with probable suicide attempts from unintentional firearm discharges based on narrative language. Findings highlight limitations of structured intent coding within NEISS-FISS and demonstrate the value of narrative for improving characterization of firearm injury circumstances, revealing distinct racial and ethnic group patterns, to inform more tailored prevention strategies.

Plain-Language Summary

Detecting Intimate Partner Violence Injury Mechanisms Using LLMs and LDA

Wed, 29 Apr 2026 15:45:00 -0500

Dr. Gia Barboza-Salerno, Olivia McLucas, and Dr. Karla Shockley will present their research at the 2026 Symposium on Data Science & Statistics (SDSS) in Milwaukee, Wisconsin.

Presentation Details

Title: Automated Detection of Intimate Partner Violence Injury Mechanisms from Emergency Department Narratives Using Large Language Models and Latent Dirichlet Allocation

Session: AI and LLM Applications I
Date: Tuesday, April 29, 2026
Time: 3:45 - 5:15 PM CT
Format: 25-minute refereed presentation
Location: Milwaukee, Wisconsin

Registration: Register for SDSS 2026
Registration Deadline: March 25, 2026

Abstract

We developed an AI pipeline designed to overcome limitations in existing surveillance systems by distinguishing the underlying mechanisms of Traumatic Brain Injury (TBI) in intimate partner violence (IPV). The primary goal was to apply a Large Language Model (LLM) to extract and classify behavioral and contextual information from unstructured free-text narratives within the National Electronic Injury Surveillance System (NEISS).

The pipeline first cleans the narratives and uses the LLM to convert the text into six targeted structured semantic representations (e.g., action, cause, diagnosis), which are then quantified using Latent Dirichlet Allocation (LDA) topic modeling. Our results can be used to enhance injury surveillance to identify IPV in emergency department settings.

Introduction

Head, face, and neck injuries are common IPV-related injuries and frequently involve TBI, yet NEISS diagnostic codes lack the behavioral detail needed to differentiate mechanisms of harm. Although NEISS narratives contain brief descriptions of arguments, aggression, object use, and environmental context, they have been underutilized in surveillance research and rarely analyzed at scale.

As a result, diagnostic codes collapse distinct injury pathways—including accidental falls, object impacts, impulsive actions during conflict, and direct assaults—into the same categories, obscuring meaningful variation in risk and limiting clinical screening and prevention efforts.

Our method fills this gap by applying AI-based extraction and topic modeling to detect behavioral and situational information embedded in NEISS narratives, producing a more nuanced understanding of IPV-related injury mechanisms.

Methods

Analytical Framework

We developed an AI pipeline to extract and classify behavioral and contextual information from unstructured NEISS narratives related to IPV:

Text Preprocessing: Used regular expressions to flag domestic-violence–related incidents, expand abbreviations, clean shorthand, and isolate IPV cases
Custom Prompting: Generated prompts by merging cleaned narratives with structured NEISS variables
LLM Extraction: Transformed narratives into six semantic fields (action, cause, diagnosis, mechanism, object, location)
Topic Modeling: Applied LDA to identify latent themes within each semantic field
Typology Classification: Integrated regex-based typology and LDA clusters to produce quantitative injury pathways

Text Processing and Prompting

We used vectorized regular expressions to:

Flag violence-related narratives
Expand abbreviations and clean shorthand
Isolate IPV cases by identifying relationship and conflict terms
Exclude clearly non-IPV incidents

After preprocessing, we generated custom prompts by merging each cleaned narrative with structured NEISS variables to produce context-rich descriptions of the event.

LLM Extraction and Topic Modeling

The LLM transformed cleaned narratives into six structured semantic representations:

Action: What happened (e.g., verbal disputes, punching, striking objects)
Cause: Why it happened (e.g., intoxication, falls, collisions)
Diagnosis: Medical outcome (e.g., fractures, contusions, lacerations)
How: Mechanism of injury (e.g., physical struggle, impulsive acts)
What: Objects involved (e.g., walls, windows, doors, fists)
Where: Location (e.g., home, public place, bathroom)

LDA topic modeling systematically discovered and quantified overarching themes across all responses, resulting in quantifiable dimensions like interpersonal conflict or blunt impact injuries.

Data & Results

Quantification of Contextual Dimensions (LDA Topics)

Topic modeling produced interpretable themes across all semantic dimensions:

Action: verbal disputes, play fighting, punching, striking objects
Cause: intoxication, falls, collisions with walls, windows, or furniture
Diagnosis: fractures, contusions, lacerations, closed head injury
How: physical struggle, impulsive acts, heated altercations
What: walls, windows, doors, bricks, glass, fists
Where: home, public place, outside environment, bathrooms

Identified Head/Neck Injury Typologies

Narrative- and topic-derived clusters aligned with six major typologies:

Typology	%	Representative Narrative
Accidental fall or slip	38.6%	“Slipped and fell hitting head on carpet.”
Self-inflicted harm	19.9%	“Banging head against wall during argument.”
Physical assault	25.1%	“Stabbed in neck with kitchen knife.”
Blunt object impact	9.9%	“Head-butted wall; forehead hematoma.”
Environmental	2.3%	“Pepper spray during fight.”
Vehicle-related impact	1.7%	“Swerved and crashed; head/neck pain.”

Word Cloud Visualization

The figure above shows word-cloud representations of dominant latent topics extracted from LLM-generated semantic dimensions (Cause, Diagnosis, How). Each cloud visualizes high-frequency terms within a topic cluster, highlighting recurrent mechanisms of injury across IPV-related emergency department visits.

Key patterns identified:

Heavy concentration of head trauma in incidents involving walls, floors, or doors
High rates of head contact in intoxication-related falls
Frequent head or neck injury during impulsive self-directed actions

Discussion & Conclusions

Combining regex-based classification with LLM extraction and LDA provides a detailed mapping of how, where, and why brain injuries occur during IPV-related incidents.

Key Findings

Blurred Boundaries: Many injuries categorized as “accidental” or “environmental” occur in the context of relationship conflict, revealing blurred boundaries between unintentional and interpersonal harm
Enhanced Surveillance: These methods improve surveillance and detection of IPV-related injuries in emergency departments by recovering behavioral information that structured diagnostic codes cannot capture
Mechanism Identification: The narrative-based approach identifies specific mechanisms that contribute to injury, whereas structured codes often make it difficult to distinguish between distinct behavioral pathways

Future Work

Expanding prompts to capture richer contextual cues
Validating typologies with larger datasets
Examining variation across sociodemographic and clinical subgroups

About SDSS

The Symposium on Data Science & Statistics is a premier conference bringing together researchers, practitioners, and students working at the intersection of statistics, data science, and computational methods. The 2026 symposium takes place April 28 - May 1 in Milwaukee, Wisconsin.

Keywords: Traumatic brain injury, intimate partner violence, surveillance, Natural Language Processing, topic modeling, emergency medicine, LLM, LDA

Ohio Recovery Housing Conference 2026

Wed, 15 Apr 2026 13:00:00 +0000

Dr. Keith Warren, Dr. Gia Barboza-Salerno, and Taylor Harrington will present at the 2026 Ohio Recovery Housing Conference during Breakout Session D (1:00 - 2:15 PM) on Wednesday, April 15, 2026.

This presentation will share findings from the lab’s ongoing research on recovery housing, examining the role of recovery housing in supporting individuals with substance use disorder, particularly in underserved and high-risk communities.

Conference Details:

Date: April 15, 2026
Time: 1:00 - 2:15 PM
Session: Breakout Session D
Registration: Register here
Full Schedule: View conference agenda

AERA 2026: Harsher Punishments, Lesser Infractions - Disciplinary Injustice for Native American Students

Sat, 11 Apr 2026 00:00:00 +0000

Presentation at the American Educational Research Association (AERA) Annual Meeting

Date: Saturday, April 11, 2026
Time: 9:45 AM to 11:15 AM PDT (12:45 PM to 2:15 PM EDT)
Location: JW Marriott Los Angeles L.A. LIVE, 4th Floor, Diamond 2

Abstract

Exclusionary discipline disproportionately affects youth of color, but Indigenous student experiences remain underexplored. Using 2019-2020 New Mexico K-12 data, we applied random intercept multilevel modeling to examine the impact of race and infraction severity on odds of receiving harsh discipline (out-of-school suspension, justice system referral) versus in-school suspension.

Key Findings:

White students had 20% lower odds (p = .003) of receiving harsh discipline compared to Native American students
While odds of harsh discipline increased for all students as infraction severity increased, odds of receiving harsh discipline declined more sharply for White, Black, and Latine students compared to Native American students as infraction severity decreased
Native American students faced harsher consequences for less serious infractions, underscoring the need for equity-focused discipline reform

This research highlights critical disparities in school discipline practices and calls for systemic reform to ensure equitable treatment of Indigenous students in New Mexico public schools.

Authors: Allie Watson-Grace, Hannah Yang, and Dr. Gia Barboza-Salerno

This manuscript is currently under review at AERA Open.

IPR Seminar Series: Accessibility to Abortion Services, Maternal Vulnerability, and Birth Outcomes

Tue, 10 Mar 2026 12:30:00 +0000

Dr. Gia Barboza-Salerno will present at the Institute for Population Research (IPR) Seminar Series on Tuesday, March 10, 2026.

Presentation Details

Title: Accessibility to Abortion Services, Maternal Vulnerability, and Birth Outcomes

Date: Tuesday, March 10, 2026
Time: 12:30 - 1:30 PM
Location: Townshend Hall, room 038, 1885 Neil Ave Mall and Zoom

Registration: Register for Zoom access

Abstract

Does spatial accessibility to abortion care intersect with maternal vulnerability to shape adverse birth outcomes? In the post-Dobbs landscape, access to abortion services has become increasingly uneven, with substantial geographic barriers emerging across large portions of the state. This study examines whether spatial constraints on abortion access align with community-level maternal vulnerability and how these structural conditions are associated with adverse birth outcomes.

The study uses a novel Maternal Vulnerability Index (MVI), a validated, census-tract-level measure that captures multidimensional drivers of poor maternal health, including:

Healthcare access
Reproductive health resources
Physical and mental health
Environmental conditions
Socioeconomic disadvantage

Geographic information systems (GIS) and multimodal transportation network modeling are used to quantify travel burden, service deserts, and spatial accessibility to abortion. These spatial access measures are then linked to adverse birth outcomes (e.g., preterm birth), to assess whether communities facing greater abortion access constraints experience elevated perinatal risk.

By situating abortion accessibility within a comprehensive measure of maternal vulnerability, this study highlights how geographic and structural inequities in reproductive healthcare access may contribute to persistent disparities in maternal and infant health.

Living Here Predicts How You Will Die: Neighborhood Context and Mortality Patterns

Wed, 04 Feb 2026 00:00:00 +0000

Abstract

Understanding how neighborhood context shapes patterns of death is critical to identifying the spatial distribution of health inequities. Medical examiner data contain rich narrative descriptions of causes and circumstances of death that are often absent from clinical records and can offer novel insights for public health prevention.

This study integrates Geographic Information Systems (GIS), machine learning (ML), and natural language processing (NLP) frameworks to demonstrate how neighborhood clustering and predictive text analytics can jointly reveal spatially stratified mortality patterns to examine how structural neighborhood conditions predict not only whether but how people die, thereby advancing geographically tailored public health insights.

Methods

Tract-level mortgage lending and demographic data were integrated with narrative cause-of-death records from the Cook County Medical Examiner. A GIS-informed machine learning framework was used to classify neighborhood context and link it to mortality patterns in Cook County, Illinois.

Using tract geometries from the American Community Survey (ACS) 2020 and integrating:

2018 racial/ethnic composition
Median household income
Shifts in mortgage lending by borrower race/ethnicity and income from HMDA
CDC Social Vulnerability Index

A standardized feature set capturing racial/ethnic composition, economic resources, and structural change in lending markets was constructed. Unsupervised machine learning methods with gap statistics were used to derive five interpretable neighborhood clusters.

Text descriptions from cause-of-death fields were cleaned, lemmatized, and harmonized into medical concepts, then summarized by cluster using frequency and tf–idf contrasts.

Key Findings

The five neighborhood clusters showed distinct mortality profiles that mirrored structural conditions:

Disinvested Black neighborhoods: Marked by violence and street-level drug toxicity
Affluent White enclaves: Reflected prescription/synthetic drug deaths and self-harm
Latino working-class areas: Exhibited mixed patterns of chronic disease, injury, and substance involvement
White-Latino transition neighborhoods: Dominated by chronic illness and degenerative diseases
Lower-income Black neighborhoods: Burdened by chronic illness, degenerative diseases, infectious causes, and substance-related mortality

Predictive Modeling Results

Penalized logistic regression models (glmnet) were used to predict cluster membership from:

Text alone
Covariates (income, rent, poverty, age, gender)
Both combined

Key Results:

Text features exhibited high sensitivity across clusters
Socioeconomic covariates improved specificity, particularly in affluent White enclaves
Combined models (text + covariates) achieved the highest overall accuracy
Performance varied by cluster: affluent White enclaves were well distinguished by socioeconomic covariates, while text features more easily identified disinvested Black and lower-income Black neighborhoods

Research Questions

RQ1: Can neighborhood structural typologies derived from mortgage lending and demographic patterns be meaningfully linked to cause-of-death narratives?

RQ2: Does narrative text from Medical Examiner death records improve prediction of neighborhood typology?

Conclusions

These results highlight how integrating structured tract-level data with unstructured death narratives uncovers not only place-based disparities in mortality risk, but also systematic differences in how death is recorded across structurally distinct neighborhoods.

Five neighborhood clusters were identified using k-means clustering on tract-level indicators. Penalized logistic regression models demonstrated that:

Combined text and covariate models achieved the highest accuracy across most neighborhood types
Narrative language varied systematically by neighborhood type, reflecting place-based inequality in death documentation

Topics: Mortality patterns, neighborhood effects, health inequities, machine learning, natural language processing, GIS, medical examiner data, Cook County

Dr. Barboza-Salerno Addresses Grand Challenge to Prevent Gun Violence

Tue, 06 Jan 2026 00:00:00 +0000

Dr. Gia Barboza-Salerno addresses the Grand Challenge to Prevent Gun Violence in a feature by the College of Social Work. Read more about the initiatives to address gun violence on the College of Social Work blog.

A Geospatial Analysis of Built and Social Environmental Factors Associated with Student Crash Risk Near Schools in Boston, Massachusetts

Sun, 04 Jan 2026 09:00:00 +0000

Emotional and Physical Symptoms Following Intimate Partner Violence Victimization in the United States: Implications for Law and Public Health Policy

Mon, 01 Dec 2025 00:00:00 +0000

Living Here Predicts How You’ll Die: Linking Death Narratives to Neighborhood Economic Disinvestment with Unsupervised Machine Learning

Wed, 19 Nov 2025 12:00:00 +0000

Development and Spatial Validation of a Random Forest Prediction Model for Firearm-Related Injury Risk in Chicago Census Tracts

Wed, 01 Oct 2025 00:00:00 +0000

Presentation at Women in Statistics and Data Science Conference

Our presentation, Modeling Firearm Injury Risk Using Spatial Random Forests, was delivered by Dr. Gia Barboza-Salerno at the Women in Statistics and Data Science on November 12, 2025 in Cincinnati, OH.

Abstract

This research develops and validates a spatial random forest prediction model for firearm-related injury risk at the census tract level in Chicago. By incorporating spatial autocorrelation and neighborhood-level predictors, the model achieves high predictive accuracy in identifying areas at elevated risk for gun violence.

Methodology

Spatial Random Forest Approach

The spatial random forest model extends traditional random forest methods by:

Capturing Non-Linear Relationships: Modeling complex interactions between neighborhood characteristics and shooting risk
Handling Spatial Autocorrelation: Incorporating spatial structure through lag variables
Variable Importance: Identifying which neighborhood features are most predictive
Robust Predictions: Providing out-of-sample risk estimates

Data Sources

Shooting incident data from Chicago
Census tract-level socioeconomic indicators
Built environment features
Spatial lag variables to capture spillover effects

Model Validation

Cross-validation for predictive performance assessment
Spatial autocorrelation diagnostics
Comparison with traditional regression approaches

Key Findings

The predicted shooting risk map (shown above) demonstrates:

High Spatial Variation: Risk scores range from 0 to over 80 predicted incidents
Concentrated Risk: Certain neighborhoods show elevated firearm violence risk
Model Accuracy: Spatial random forest successfully captures fine-grained risk patterns
Spatial Clustering: Violence risk exhibits significant spatial autocorrelation

Applications

This spatial prediction model has important applications for:

Resource Allocation: Targeting violence prevention resources to high-risk areas
Prevention Planning: Understanding neighborhood-level drivers of gun violence
Policy Evaluation: Assessing how neighborhood changes affect violence risk
Early Warning: Detecting emerging hotspots before violence escalates

Implications

The research demonstrates that spatial random forest models can:

Improve prediction of firearm violence risk beyond traditional methods
Identify actionable neighborhood-level intervention targets
Account for complex spatial dependencies in violence patterns
Support evidence-based violence prevention strategies

Conference: Women in Statistics and Data Science Conference
Location: Cincinnati, Ohio
Date: October 1-3, 2025
Authors: Gia Barboza-Salerno, Hexin Yang, Taylor Harrington

Intimate Partner Violence & Brain Injury in U.S. Adults: Findings from the Medical Expenditure Panel Survey, 2021

Wed, 27 Aug 2025 00:00:00 +0000

Overview

National survey data show that adults who screen positive for intimate partner violence (IPV) experience more injuries and face greater medical costs than those who do not. Nine million screened IPV-positive using the HITS screening tool, and survivors were significantly more likely to utilize emergency rooms and hospitals for treating their injuries. Head and brain injuries were much more frequent among IPV survivors, and the average cost of treating these injuries was nearly twice as high compared to adults who did not screen positive for IPV.

Key Messages

Approximately 9 million adults screened positive for IPV in 2021.
IPV-positive adults had 3–4× higher ER and inpatient injury rates than IPV-negative adults.
Adults who screened positive for IPV experienced more than twice the rate of brain-injury healthcare events compared to those who screened negative.
Average per-capita spending on brain-injury care was about twice as high among IPV-positive adults.
Inpatient TBI events are less common than ER events but far more costly; counts and spending are higher among IPV-positive adults.
Poverty, social stressors, depression, and chronic conditions are significantly associated with IPV screen positivity.

Policy Relevance & Recommendations

Standardized protocols: Implement routine protocols so that IPV-positive patients with head injury are screened for brain injury, and patients with head injury are also screened for IPV.
Integrated referral pathways: Establish clear referral systems across emergency, primary, and specialty care so that when screening protocols are triggered, patients are automatically connected to appropriate follow-up services.
Follow-up for delayed symptoms: Incorporate temporal follow-up in care plans, recognizing that brain injury symptoms may appear or worsen over time.
Accessibility and appropriateness: Ensure screening and follow-up services are accessible, widely available, and tailored to the needs of IPV survivors.
Holistic screening: Extend screening beyond identifying head injury to include social and health needs that may magnify injury effects and increase vulnerability to future IPV.
Care coordination: Utilize medical social workers and interdisciplinary teams to integrate medical, psychosocial, and safety needs into a comprehensive care response.

Methodology

This brief analyzes nationally representative data from the Medical Expenditure Panel Survey—Household Component (MEPS-HC) 2021. Adults (age ≥18) were classified as IPV screen-positive if they scored ≥10 on the Hurt, Insult, Threaten, Scream (HITS) scale, a validated four-item screening tool for intimate partner violence.

Injury outcomes were identified using ICD-10-CM codes and Clinical Classifications Software Refined (CCSR) groupings from the MEPS Conditions file, with condition–event linkages used to assign emergency department (ER), inpatient, outpatient, office-based, home health, and prescription care to specific injuries. Brain injuries were defined as traumatic brain injury (ICD-10 S06) and other head injuries (S00–S09).

We estimated:

Injury and brain-injury events per 100 adults,
Per-capita spending (in 2021 USD), and
Weighted counts of injury-linked encounters.

Analyses applied MEPS survey weights (PERWT21F) and accounted for the complex design (VARSTR, VARPSU). We used survey-weighted generalized linear models (svyglm) to generate adjusted comparisons standardized over age, sex, and race/ethnicity. Risk ratios were estimated for sociodemographic factors, health-related social needs, and chronic health conditions associated with IPV screen positivity.

Suggested Citation

Barboza-Salerno, G. E., Shockley McCarthy, K., Watson-Grace, A., Harrington, T., McCarthy, M. A., Peremen, N., & McLucas, O. (2025). Intimate Partner Violence & Brain Injury in U.S. Adults: Findings from the Medical Expenditure Panel Survey, 2021. Policy brief. The Ohio State University, ISSUE Lab (Investigating Spatial Structures in Urban Environments).

Download

Download the full brief for more information

How to Analyze Caselaw Using CourtListener's API in R

Fri, 01 Aug 2025 00:00:00 +0000

I previously showed how to pull Indian Child Welfare Act (ICWA) cases using the Caselaw Access Project (Harvard Law). That API has since changed. This post updates the workflow using the Free Law Project’s CourtListener, which now partners with Harvard to make this data freely available.

Finding the Cases

Start at the CourtListener website and search for “Indian Child Welfare Act” (or "icwa"). You’ll see a page of opinion results.

Click Select Jurisdiction → State. Clear all, then select Ohio Supreme Court, Ohio Court of Appeals, and Ohio Court of Claims. This gives you a quick check on the expected volume to validate the code’s output below.

The site reported 36 cases from Ohio. Note that the exact number may change over time as new opinions are added or metadata is corrected.

Getting the Data into R

The Search API returns clean JSON results by querying for opinions that contain the exact phrase "indian child welfare act", restricting to published opinions, ordering by relevance, and (optionally) limiting results to three Ohio courts.

library(httr2)
library(purrr)
library(dplyr)
library(tibble)
library(tidyr)
library(stringr)

SEARCH_BASE <- "https://www.courtlistener.com/api/rest/v4/search/"
UA <- "R/httr2 (your-email-address) – ICWA metadata collection"

query_params <- list(
 q = '"indian child welfare act"',
 type = "o",
 order_by = "score desc",
 stat_Published = "on",
 court = "ohio ohioctapp ohioctcl" # optional: specify court
)

A few helper functions keep the parsing tidy:

pull_chr <- function(x, ...) {
 v <- purrr::pluck(x, ..., .default = NA)
 if (is.null(v) || length(v) == 0) return(NA_character_)
 as.character(v)
}

pull_num <- function(x, ...) {
 v <- purrr::pluck(x, ..., .default = NA_real_)
 suppressWarnings(as.numeric(v))
}

get_json <- function(url) {
 resp <- request(url) |> req_user_agent(UA) |> req_url_query() |> req_perform()
 resp_check_status(resp)
 resp_body_json(resp, simplifyVector = FALSE)
}

This function handles pagination automatically, following the next cursor until all pages are exhausted:

fetch_search_pages <- function() {
 req <- request(SEARCH_BASE) |>
 req_user_agent(UA) |>
 req_url_query(!!!query_params)

 dat <- req |> req_perform() |>
 (\(r){ resp_check_status(r); resp_body_json(r, simplifyVector = FALSE) })()
 pages <- list(dat)

 while (!is.null(dat$`next`)) {
 dat <- request(dat$`next`) |> req_user_agent(UA) |> req_perform() |>
 (\(r){ resp_check_status(r); resp_body_json(r, simplifyVector = FALSE) })()
 pages <- append(pages, list(dat))
 }
 pages
}

pages <- fetch_search_pages()
results <- pages |> map("results") |> flatten()

results is a list of per-case records from the API. Each element includes the case name, filing date, court identifiers, citation strings, a public URL where the caselaw text is available, and other metadata.

For Ohio, this produced ~35 unique case names versus 36 displayed on the site. Minor discrepancies happen because search hits can include multiple opinions or sibling opinions per matter, and because the index is updated over time.

Downloading the Full Text

Each opinion has an API endpoint with plain-text and HTML fields. A small loop can request the text for each case URL, with a short pause and retry between requests, and save one .txt file per opinion:

download_opinion_text <- function(url, out_dir = "opinions") {
 dir.create(out_dir, showWarnings = FALSE)
 slug <- basename(url)
 path <- file.path(out_dir, paste0(slug, ".txt"))

 resp <- request(url) |> req_user_agent(UA) |> req_perform()
 resp_check_status(resp)
 body <- resp_body_json(resp, simplifyVector = FALSE)

 text <- body$plain_text %||% body$html_with_citations %||% ""
 writeLines(text, path)
 Sys.sleep(0.5) # be considerate of rate limits
 invisible(path)
}

walk(map_chr(results, "absolute_url"), download_opinion_text)

Here are the Ohio ICWA cases downloaded to a local folder:

Visualizing the Results

You can use ggplot2 to visualize the results. Below are two examples: filings by year and volume by court.

What You Can Do Next

After downloading all data — including caselaw text and metadata — you can:

Natural Language Processing — examine sentiment or common words across opinions
Topic Modeling — explore latent themes in ICWA cases using LDA or STM
Map filings by jurisdiction — visualize geographic patterns of ICWA litigation
Integrate with other sources — link to child welfare administrative data or Census indicators

Remember to be considerate of rate limits and to include a User Agent string with your contact information in every request.

The complete workflow is available on my GitHub.

Happy Coding!

A PostgreSQL Tutorial for Analyzing Racial and Geographic Disparities in HMDA Applications

Wed, 16 Jul 2025 00:00:00 +0000

My lab recently published a paper examining the association between redlining, reinvestment, racial segregation, and firearm-related injury in Chicago, Illinois. You can read the paper here: Redlining, reinvestment, and racial segregation: a Bayesian spatial analysis of mortgage lending trajectories and firearm-related violence.

To develop reverse redlining measures, we relied on data from the Home Mortgage Disclosure Act (HMDA).

If you’re interested in patterns of lending discrimination, the HMDA dataset is an invaluable resource. It contains millions of individual loan application records submitted annually by financial institutions, making it one of the most extensive and detailed public data sources on mortgage lending in the United States. The 2024 nationwide Modified Loan/Application Register (LAR) alone contains over 12 million records and 85 fields that detail applicant and co-applicant demographics, loan terms, and property characteristics — including race, income, loan amount, interest rate, property value, and loan outcomes.

Each record includes the census tract of the loan origination, which allows researchers to situate applicants within their geographic context. This spatial identifier can be linked to other data sources (e.g., Census or ACS data), enabling robust analyses of racial disparities in lending, neighborhood-level disinvestment, and broader patterns of residential inequality.

I highly recommend The Color of Law by Richard Rothstein for anyone interested in the deep-rooted history of discriminatory housing practices in the United States. The book compellingly explains how federal, state, and local policies intentionally created and reinforced de jure segregation. Chicago, one of the most highly segregated urban areas in the country, is prominently featured.

Due to its sheer volume, the HMDA data is not well-suited for analysis using spreadsheets like Excel — particularly when analyzing multiple years or conducting joins. Instead, its relational structure lends itself to PostgreSQL, an open-source relational database management system (RDBMS) that enables efficient querying, filtering, and aggregation of large datasets. PostgreSQL also supports the PostGIS geospatial extension, enabling direct spatial querying and linking to geographic units such as census tracts or ZIP codes. Download PostgreSQL if you don’t already have it.

Step 1: Download the Dataset

The dataset is freely available to the public.

Navigate to https://ffiec.cfpb.gov/data-publication/modified-lar/2024
Choose Year: 2024 → Nationwide: Combined file
Check Include File Header, then click Download Combined Modified LAR with Header

Save or extract the file to: C:\Users\YOUR_USER_NAME\Downloads\2024_combined_mlar_header\

Step 2: Confirm the File Structure in PowerShell

The file is a pipe-delimited .txt file — it can be tricky to read in spreadsheet tools. Use Windows PowerShell to inspect the structure first:

cd "C:\Users\YOUR_USER_NAME\Downloads\2024_combined_mlar_header"

# View the header (column names)
Get-Content .\2024_combined_mlar_header.txt -TotalCount 1

# Count the number of columns (should be 85)
((Get-Content .\2024_combined_mlar_header.txt -TotalCount 1) -split '\|').Count

Step 3: Create the PostgreSQL Table

Launch your SQL interface (pgAdmin Query Tool or psql):

psql -U your_username -d your_database_name

Then define the 85-column table structure:

DROP TABLE IF EXISTS lar2024;
CREATE TABLE lar2024 (
 activity_year INTEGER,
 lei TEXT,
 loan_type INTEGER,
 loan_purpose INTEGER,
 preapproval INTEGER,
 construction_method INTEGER,
 occupancy_type INTEGER,
 loan_amount BIGINT,
 action_taken INTEGER,
 state_code TEXT,
 county_code TEXT,
 census_tract TEXT,
 applicant_ethnicity_1 INTEGER,
 applicant_ethnicity_2 TEXT,
 applicant_ethnicity_3 TEXT,
 applicant_ethnicity_4 TEXT,
 applicant_ethnicity_5 TEXT,
 co_applicant_ethnicity_1 INTEGER,
 co_applicant_ethnicity_2 TEXT,
 co_applicant_ethnicity_3 TEXT,
 co_applicant_ethnicity_4 TEXT,
 co_applicant_ethnicity_5 TEXT,
 applicant_ethnicity_observed INTEGER,
 co_applicant_ethnicity_observed INTEGER,
 applicant_race_1 INTEGER,
 applicant_race_2 TEXT,
 applicant_race_3 TEXT,
 applicant_race_4 TEXT,
 applicant_race_5 TEXT,
 co_applicant_race_1 INTEGER,
 co_applicant_race_2 TEXT,
 co_applicant_race_3 TEXT,
 co_applicant_race_4 TEXT,
 co_applicant_race_5 TEXT,
 applicant_race_observed INTEGER,
 co_applicant_race_observed INTEGER,
 applicant_sex INTEGER,
 co_applicant_sex INTEGER,
 applicant_sex_observed INTEGER,
 co_applicant_sex_observed INTEGER,
 applicant_age TEXT,
 applicant_age_above_62 TEXT,
 co_applicant_age TEXT,
 co_applicant_age_above_62 TEXT,
 income TEXT,
 purchaser_type INTEGER,
 rate_spread TEXT,
 hoepa_status INTEGER,
 lien_status INTEGER,
 applicant_credit_scoring_model INTEGER,
 co_applicant_credit_scoring_model INTEGER,
 denial_reason_1 TEXT,
 denial_reason_2 TEXT,
 denial_reason_3 TEXT,
 denial_reason_4 TEXT,
 total_loan_costs TEXT,
 total_points_and_fees TEXT,
 origination_charges TEXT,
 discount_points TEXT,
 lender_credits TEXT,
 interest_rate TEXT,
 prepayment_penalty_term TEXT,
 debt_to_income_ratio TEXT,
 combined_loan_to_value_ratio TEXT,
 loan_term TEXT,
 intro_rate_period TEXT,
 balloon_payment INTEGER,
 interest_only_payment INTEGER,
 negative_amortization INTEGER,
 other_non_amortizing_features INTEGER,
 property_value TEXT,
 manufactured_home_secured_property_type INTEGER,
 manufactured_home_land_property_interest INTEGER,
 total_units TEXT,
 multifamily_affordable_units TEXT,
 submission_of_application INTEGER,
 initially_payable_to_institution INTEGER,
 aus_1 INTEGER,
 aus_2 TEXT,
 aus_3 TEXT,
 aus_4 TEXT,
 aus_5 TEXT,
 reverse_mortgage INTEGER,
 open_end_line_of_credit INTEGER,
 business_or_commercial_purpose INTEGER
);

Step 4: Import the Data

In psql, use the following single-line import command:

\copy lar2024 FROM 'C:/Users/YOUR-USER-NAME/Downloads/2024_combined_mlar_header/2024_combined_mlar_header.txt' WITH (FORMAT csv, DELIMITER '|', HEADER true, NULL '');

This tells PostgreSQL to:

Read a pipe-delimited file
Use the header row to match columns
Treat empty strings as NULL

Note: This must be entered as a single line. You will get a confirmation message when the data import succeeds.

Step 5: Verify the Import

The official CFPB download page indicates that the 2024 nationwide file contains approximately 12 million rows. After importing, verify with:

SELECT COUNT(*) FROM lar2024;
SELECT * FROM lar2024 LIMIT 5;

If the row count is substantially less than 12 million, the import did not complete successfully.

Create the Chicago, Illinois Subset

With the full dataset loaded, you can create a focused subset with a CREATE TABLE AS SELECT statement. The query below selects conventional, owner-occupied, site-built single-family loans in Cook County (FIPS 17031), excluding missing or exempt values:

DROP TABLE IF EXISTS public.il2024f1;
CREATE TABLE public.il2024f1 AS
SELECT
 l.activity_year,
 l.state_code,
 l.county_code,
 l.census_tract,
 l.debt_to_income_ratio,
 l.combined_loan_to_value_ratio,
 l.income,
 l.property_value,
 l.loan_amount,
 l.interest_rate,
 l.applicant_ethnicity_1,
 l.applicant_race_1,
 l.denial_reason_1,
 l.rate_spread,
 l.hoepa_status,
 l.lien_status
FROM public.lar2024 l
WHERE
 l.loan_purpose = '1' -- Home purchase
 AND l.action_taken = '1' -- Loan originated
 AND l.loan_type = '1' -- Conventional
 AND l.occupancy_type = '1' -- Owner-occupied
 AND l.state_code = 'IL'
 AND l.county_code = '17031' -- Cook County
 AND l.property_value ~ '^[0-9]+$'
 AND l.income ~ '^[0-9]+$'
 AND l.loan_amount IS NOT NULL
 AND l.loan_amount <> 99999
 AND l.construction_method = '1' -- Site-built
 AND l.total_units IN ('1','2','3','4','');

Prepare for Additional Analysis

Once exported to CSV, I divided the number of Black applicants in each census tract by the total number of applicants in that tract to get the share of Black applicants. I also downloaded Median Household Income data using the tidycensus package in R (ACS 5-year estimates 2018–2023).

The maps below compare neighborhood income levels with the geographic concentration of Black mortgage applicants receiving loans in 2024. Do you observe any patterns?

The spatial overlap — or rather, the striking lack of it — between high-income areas and Black loan recipients is consistent with decades of research documenting structural barriers to homeownership for Black families in Chicago.

Conclusion

The HMDA dataset offers an unparalleled opportunity to examine lending patterns, racial disparities, and neighborhood-level investment across the United States. By importing the 2024 Modified LAR into PostgreSQL, researchers can efficiently manage, filter, and analyze millions of records while linking them to geographic and demographic data.

Next steps may include:

Join with Census shapefiles to visualize patterns by tract
Summarize by loan purpose, race, or geography
Export to PostGIS for spatial queries
Link to neighborhood typologies or redlining indices

Happy Coding!

Evaluating Child Fatality Reports using Structural Topic Models

Wed, 02 Jul 2025 00:00:00 +0000

An examination of the effect of area-level characteristics on juvenile justice and child welfare referrals using multivariate Bayesian spatial modeling