Same hazard, different outcomes: model bias and the role of institutional capacity in earthquake fatality estimation

This study investigates why universal disaster risk models produce systematic bias across different institutional contexts, focusing on Türkiye and Chile. Although hazard–exposure-based models are widely used, their accuracy varies significantly across countries with similar seismic conditions, and empirical explanations for these discrepancies remain limited. To address this gap, a two-stage regression approach is employed using a dataset of 40 high-impact earthquakes. In the first stage, a baseline model including only physical variables is specified. In the second stage, an institutional model incorporating the civil justice indicator as a proxy for institutional capacity is introduced. Model performance is evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²). The results reveal a systematic pattern: the baseline model underestimates fatalities in Türkiye while overestimating them in Chile, indicating that prediction errors are structurally linked to contextual factors. With the inclusion of institutional variables, model performance improves substantially in both cases. For Türkiye, R² increases significantly and error metrics decline. For Chile, model fit improves and the error distribution becomes more balanced. Regression results show that the institutional weakness indicator (1 − J) is a statistically significant and robust predictor. Predicted–observed comparisons confirm that estimates generated by the institutional model closely align with actual outcomes. The findings demonstrate that earthquake fatalities cannot be explained solely by physical hazard and exposure. Instead, institutional capacity and enforcement mechanisms play a decisive role, conceptualized here as institutional implementation capacity.