Media Attention and Specialty Choice: Evidence from Healthcare Violence Coverage in Türkiye

This study examines whether the intensity of healthcare violence coverage in Turkish media and online discourse is associated with shifts in medical specialty preferences among physicians, as measured by branch-level minimum scores on the Medical Specialization Examination (TUS). Drawing on Bo et al. (2020), who document that newspaper coverage of anti-doctor violence in China reduces medicine enrollment, we adapt this logic to specialty-level career decisions in Türkiye. Two text-based treatment proxies are constructed: D1, an annual count of Google News articles on healthcare violence (2013–2025), and D2, an annual violence proportion derived from a purposively constructed Ekşi Sözlük corpus annotated through a multi-stage pipeline combining human annotation and large language model extension. Branches are classified into high-, intermediate-, and low-exposure groups following Dorsey et al. (2003). Spearman correlations and year-level OLS models with exposure-group interaction terms consistently show that higher media intensity is associated with declining minimum scores in high- and intermediate-exposure branches, while low-exposure branches remain unaffected. Branch-level panel fixed effects models further confirm a global negative association between D2 and minimum scores across all branches. The D1 main effect is unstable across trend specifications, but interaction terms are robust to both year dummy and quadratic trend controls, as well as to leave-one-out sensitivity analysis. Results are reported as correlational; causal identification is not claimed.