Research
Under Review
- Are Neighborhood Effects Larger for Children with Disabilities? Evidence from Texas (preprint available at SSRN)
Sole author
This study examines how neighborhood environments shape long-term educational attainment for children with and without disabilities, using administrative records on approximately 1.2 million children in Texas from six kindergarten cohorts (1994-1999). I exploit two sources of variation, the educational opportunity gap between origin and destination school districts and the duration of childhood exposure to the destination, to estimate causal neighborhood effects by disability status. Exposure to higher-opportunity districts improves both high school and college completion for all children, with effects larger for children with disabilities than for those without. These effects scale with the duration of childhood exposure, such that children who move earlier accumulate larger gains. Effects also differ by disability type, with children with physical or sensory disabilities experiencing larger gains in high school completion but attenuation at the college level, and children with cognitive or emotional disabilities showing persistent gains across both outcomes. These results are robust to alternative fixed effects specifications and displacement shock identification. The findings suggest that disability inequality is deeply localized, as children with disabilities are disproportionately shaped by the neighborhoods where they grow up.
- Growing Up in Disability-Dense School Districts: Long-Term Impacts of Childhood Exposure on Educational Attainment (preprint available at EdWorkingPapers)
Sole author
This study examines whether exposure to higher disability density during childhood affects long-term educational attainment. Using administrative data on more than 170,000 children from six Texas kindergarten cohorts (1994-1999) who move across districts during K-12, I exploit variation in disability density across school districts to estimate causal effects on high school and college completion. Moving to districts with higher disability density improves both outcomes for children with disabilities, with effects nearly twice as large as those for children without disabilities. Effects on children without disabilities are positive or at least statistically indistinguishable from zero across specifications, providing no evidence of negative spillovers and suggesting that disability-diverse school districts may benefit all children.
- A Geography of Administrative Burden: Measuring Spatial Access to Social Security Field Offices in the United States (preprint available at SSRN)
First author (with Peter Kedron)
This study measures geographic accessibility to the Social Security Administration (SSA) field offices in the United States, using the Enhanced Two-Step Floating Catchment Area method, which defines access as a ratio of supply to competing demand. Approximately 2.7 million residents live in 933 tracts where no SSA field office is reachable within the 120-minute catchment. Accessibility is positively and significantly associated with receipt of Supplemental Security Income and Social Security Disability Insurance, two programs that impose enough documentation and review burden to make an office visit likely, but not with receipt of retired-worker benefits, which impose considerably less. That contrast indicates the index captures a geographic component of administrative burden. Applying the measure to the 12 field offices whose leases were announced for termination in 2025 identifies 165,163 residents who would be left with no reachable office.
- Does Carbon Pricing Outperform Command-and-Control Regulation? Firm-Level Evidence from Korea’s Dual Regulatory Framework (preprint available at SocArXiv)
Sole author
This study exploits South Korea’s unique dual-policy, in which an emissions trading scheme (ETS) and a command-and-control program (Target Management System, TMS) operate in parallel, to compare carbon pricing with prescriptive regulation at the firm level. Using a difference-in-differences design on firm-level panel data from 2011 to 2022, I find that ETS-regulated firms reduced energy use by 5.8% to 8.8% and carbon emissions by 7.3% to 8.5% relative to TMS firms, while effects on carbon intensity were not robust. These reductions were strongest in the more market-oriented phases of the ETS, which featured allowance auctioning and benchmark-based allocation, pointing to the importance of incentive-based policy design.
- Spatial Heterogeneity in the Determinants of Antibiotic Prescribing Among Medicare Beneficiaries: A Geographically Weighted Regression Analysis (preprint available at SSRN)
Co-author
Antibiotic overuse remains a public health concern in the United States, yet how the determinants of prescribing vary across space is poorly understood. Using Medicare Part D data (2017-2023), this study examines county-level antibiotic prescribing among Medicare beneficiaries and assesses whether its determinants are spatially stable or place-dependent. Prescribing was significantly clustered, with high-prescribing counties concentrated in the Southern United States. Using spatial regression and Geographically Weighted Regression, we find that a single national estimate can obscure divergent local patterns, where provider density was positively associated with prescribing across nearly all counties, whereas rurality and poverty reversed in direction across regions. These findings suggest that uniform national stewardship strategies may be less effective than locally tailored interventions.
- mHealth Intervention Integrating Personal PM2.5 Monitoring and Deep Learning to Reduce Pediatric Asthma Exacerbations: A Pilot Study (preprint available at JMIR)
First author
Background: Fine particulate matter (PM2.5) is a major trigger of pediatric asthma exacerbations, yet individual sensitivity varies considerably. Existing interventions often adopt a uniform approach, despite this heterogeneity.
Objective: This study aimed to evaluate the feasibility and effectiveness of a pilot mobile health (mHealth) intervention that integrates personal PM2.5 monitoring, deep learning (DL)-based prediction and tailored behavioral recommendations to mitigate exacerbation risks in children.
Methods: In this 3-year pilot study, 272 pediatric patients with asthma were enrolled across nine tertiary hospitals in Korea. Using asthma symptom reports and personal PM monitoring data collected via smartphones and portable devices, a 1D CNN-LSTM model was developed to identify PM-sensitive patients and predict exacerbations. After model construction, 109 participants entered the intervention phase and were allocated to three groups (model-based intervention, forecast-based intervention, or no intervention) through initial screening and model-based grouping. The model-based group received individualized alerts with behavioral recommendations based on DL predictions, while the forecast-based group received the same recommendations based on regional air quality forecasts. The primary outcome was change in asthma exacerbation rates (measured by Intervention Effectiveness Ratio, IER).
Results: The model-based group demonstrated significant reductions in exacerbation rates (median IER decrease: 6.6%; mean IER decrease: 11.5%; P < 0.05), whereas no significant changes were observed in the other groups. Odds ratio analysis indicated that the model-based group had 5.92- and 4.22-fold lower odds of PM2.5-related exacerbations compared with the no-intervention and forecast-based groups, respectively. Stratified and adjusted analyses confirmed that the benefit of model-based alerts remained robust despite baseline differences in asthma severity and control status.
Conclusions: This pilot study demonstrates the feasibility and potential effectiveness of an mHealth intervention that integrates personal PM monitoring, DL-based prediction and tailored behavioral recommendations in pediatric asthma. This approach shows promise for reducing PM-related exacerbations and warrants validation in larger, longer-term studies.
- Geographic Accessibility to Preventive Maternal and Child Oral Health in Kenya
First author
Introduction: Preventive oral health services remain largely absent from maternal and child health (MCH) care in rural sub-Saharan Africa. This study quantified geographic accessibility to preventive maternal and child oral health (MCOH) services in Kilifi County, Kenya, and evaluated the potential impact of MCH nurse training scenarios on accessibility outcomes.
Methods: We applied the Enhanced Two-Step Floating Catchment Area (E2SFCA) method to ward-level women of reproductive age (WRA) population data and public health facility data from Kilifi County. Wards falling below the estimated Kenyan national public oral health workforce ratio of 4.3 oral health providers per 100,000 WRA were designated as priority areas and classified by disparity type. Three MCH workforce training scenarios (efficiency-centered, equity-centered, and hybrid) were evaluated against a baseline, each with a cumulative training constraint of 20 MCH nurses and midwives.
Results: Only 15 of 35 wards met the estimated national public oral health workforce ratio of 4.3 providers per 100,000 WRA at baseline (mean 3.55, SD 1.84). Below-average wards comprised 10 demand-driven and 10 provider shortage wards. All three intervention scenarios substantially improved accessibility. The equity-centered scenario extended accessibility above the national ratio to 33 of 35 wards (mean 7.19, SD 2.41), while the efficiency-centered scenario achieved the highest mean index with 28 wards above the national ratio (8.00, SD 4.50) but with greater spatial variability. The hybrid scenario yielded intermediate results with 30 wards above the national ratio (mean 7.43, SD 3.59).
Conclusions: Training MCH nurses and midwives to deliver MCOH services can substantially expand geographic accessibility without new infrastructure or specialist redistribution. The E2SFCA-based scenario framework developed here offers a spatially explicit tool for evidence-driven oral health workforce planning in resource-limited settings across Kenya and sub-Saharan Africa.
Publications
- Bayesian Spatio-Temporal Modeling for Policy Evaluation: Sensitivity of Policy Effect Estimates in the Context of COVID-19 Stay-at-Home Orders (2026, PLOS One, 21(2). e0339196)
First author (with Sunghye Choi, Dohyeong Kim, and Chang-Kil Lee)
This study applies a Bayesian spatio-temporal model to demonstrate the sensitivity of policy effect estimates to spatial and temporal structure, using COVID-19 stay-at-home orders as a case study. Unlike conventional approaches, this framework accounts for geographic spillovers, temporal dependence, and space-time interaction, all of which are central to policy effect evaluation in heterogeneous settings. Implemented via Integrated Nested Laplace Approximation (INLA), the model also accommodates missing data and supports inference in high-dimensional contexts. Using Google mobility data and policy information from the Oxford COVID-19 Tracker, we estimate four models of increasing complexity: OLS, spatial, temporal, and spatio-temporal. While simpler models suggest substantial reductions in workplace and residential mobility, these effects become statistically insignificant once spatio-temporal interactions are incorporated. This pattern indicates that earlier studies may have overstated policy effects by overlooking spatio-temporal heterogeneity. Our findings demonstrate the importance of spatio-temporal modeling for policy evaluation, particularly when working with large-scale, incomplete, and unevenly distributed data.
- Regional disparities in community reintegration after burn injury: A multi-site comparison of clinical and structural factors (2026, Burns)
First author (with Kristine J. Hahm, Dohyeong Kim, Karen Kowalske)
Purpose: This study examines regional differences in community reintegration outcomes among burn survivors treated at three major U.S. burn centers and asks whether those differences are associated with where patients live.
Methods: We used data from 531 adult burn survivors in the Burn Model System (BMS) National Database, treated at centers in Seattle, Dallas, and Boston between 2015 and 2023. We linked individual-level clinical and demographic data with county-level structural variables from the Urban Institute and compared Community Integration Questionnaire (CIQ) change scores across centers, alongside patient characteristics and residential community conditions.
Results: Dallas patients experienced a significantly greater decline in CIQ scores at six months post-injury than Seattle patients (Delta = -0.71, p =.004), while Boston patients showed outcomes comparable to Seattle (Delta = -0.27, p =.276). Clinical characteristics did not differ significantly between Dallas and Seattle. Dallas patients lived in counties with substantially worse economic conditions than Seattle patients, including lower household income at the 20th percentile, higher debt burden, lower economic connectedness, and higher crime rates. Boston patients lived in counties that did not show this pattern of disadvantage, remaining comparable to or more favorable than Seattle’s on the same indicators.
Conclusions: Community reintegration after burn injury varies systematically across burn centers in ways that correspond to differences in the structural conditions of communities patients return to. These findings highlight the need to incorporate residential community conditions into burn rehabilitation planning and discharge support.
Geospatial Analysis of Community-Level Social and Environmental Barriers for Adult Burn Injury Survivors in North Texas (2025, Burns, 51(5), 107512)
First author (with Dohyeong Kim, Richard Scotch, Dohyo Jeong, and Karen Kowalske)
This preliminary study examines geographic differences in community integration among burn injury survivors in North Texas and identifies community factors that may shape their post-injury reintegration. Drawing on data from 153 adults in the Burn Model System between 2015 and 2022, we mapped county-level changes in Community Integration Questionnaire (CIQ) scores by comparing pre-injury levels with scores at six and twelve months. We then grouped counties based on whether survivors experienced consistent declines over the 12-month period and compared these counties to all others. Preliminary results reveal clear spatial disparities: counties with persistent decreases in CIQ scores tended to have higher poverty and unemployment, more crime, and poorer access to healthy food options. These patterns suggest that rural and disadvantaged communities may provide less supportive environments for reintegration. While exploratory, these findings indicate the importance of addressing local socioeconomic and environmental barriers to improve community integration outcomes for burn injury survivors.
Have Offender Demographics Changed Since the COVID- 19 Pandemic? Evidence from Money Mules in South Korea (2024, Journal of Criminal Justice, 91, 102156)
Co-author (with Sunmin Hong and Dohyo Jeong)
This study aims to investigate how the demographic characteristics of offenders have changed after the COVID-19 pandemic. Specifically, our research focuses on shifts in the nationality, gender distribution, and age profiles of money mules during this period. We utilized arrest reports data provided by the Seoul Metropolitan Police Agency in South Korea, including all 1407 individuals arrested for money mules in Seoul from February 1, 2018, to December 31, 2021. Our findings, derived from interrupted time series analyses, show a decrease in the percentage of non-Korean money mules, an increase in the proportion of female individuals engaged in money mule activities, and a rise in the average age of money mules after the outbreak of the pandemic. These insights hold significant implications for developing targeted policy interventions to mitigate potential threats associated with money mule activities.
Do Firms Respond Differently to the Carbon Pricing by Industrial Sector? How and Why? A Comparison Between Manufacturing and Electricity Generation Sectors Using Firm-Level Panel Data in Korea (2022, Energy Policy, 162, 112773)
First author (with Hyunhoe Bae)
With firm-level panel data for seven years, this study evaluates the effect of carbon pricing policy and analyzes how firms respond to the carbon price, focusing on Korea’s Emission Trading Scheme (ETS). Assuming that firms’ responses to the carbon price may differ across industries, this study compares the manufacturing and electricity generation sectors. Our panel regression analyses show that the ETS has significant impacts on firms’ carbon reduction. However, the mechanisms through which firms reduce emissions differ by industrial sector. Firms in the manufacturing sector reduce carbon emissions by improving the energy efficiency of their facilities, whereas those in the electricity generation sector reduce emissions by phasing out fossil fuels and increasing the use of low carbon-intensive energy sources. These findings imply that carbon pricing works as designed, sending economic signals for firms to decarbonize their activities, and that its effectiveness varies according to each industry’s characteristics.