Injuries/Violence
Nowcasting County-Level Homicide Rates Based on Newspaper Articles Amulya Akkapeddi* Amulya Akkapeddi Yingjie Ma Lina Zhou Janice Williams Dongsong Zhang Ahmed Arif Rajib Paul
Introduction: The lag in homicide incident reporting by the National Death Violent Reporting System (NVDRS) due to its reliance on extensive manual data extraction limits its utility for real-time public health interventions. In this case, can we predict current firearm homicide rates without experiencing delays associated with the NVDRS data availability? To answer this, the study explores the predictive capacity of newspaper articles as a proxy for nowcasting homicide rates at the county level in North Carolina (NC).
Methods: We first retrieved newspaper articles from 30+ NC open-source publications based on keywords like “gun”, “murder, and “shot” for the years 2017-2024, followed by spatiotemporal deduplication. We then aggregated articles from 2017-2021 at the county level and analyzed their correlation with the NVDRS data. Spatiotemporal analytics were finally applied to evaluate the feasibility and effectiveness of using newspaper data to forecast homicide rates.
Results: Our previous study using the NVDRS data from 2011-2021 showed that there was an increase in statewide homicide by 65%. The county-level Social Vulnerability Index (SVI) showed a positive association with homicide rates, with an incidence risk ratio (IRR) of 5.15 (95% CrI: 4.5-5.91). 155 newspaper articles in 2019 reported firearm homicides, compared to 704 reported in NVDRS. A strong county-level correlation (Pearson’s r = 0.95) between the two data sources was observed, indicating that newspapers are robust predictors of NVDRS data. Moreover, urban counties (r= 0.97) showed higher alignment between NVDRS and newspaper-reported incidents, compared to rural counties (r=0.70), likely reflecting differences in media coverage intensity and reporting practices.
Conclusion: Newspaper data show strong potential for nowcasting firearm homicides, particularly in urban areas, with alignment to NVDRS. This approach can enhance public health responses by providing timely and actionable insights.