Generalized Ordinal Regression Models for Seismic Fragility Analysis

Abstract

This study investigates the application of ordinal regression models in seismic fragility curve modeling, providing a flexible alternative to the traditional log-normal distribution function. A comparative analysis is conducted among various ordinal regression approaches, including the traditional Cumulative model as well as alternative methods like Sequential and Adjacent Category models, along with extensions that account for category-specific effects and heteroscedasticity. These models are applied to bridge damage data from the 2008 Wenchuan earthquake, using both frequentist and Bayesian inference methods. Model diagnostics, including surrogate residuals, are performed to assess model fit and performance. A total of eleven models are examined, from basic forms to those incorporating category-specific effects and variance heterogeneity. The Sequential model with category-specific effects, rigorously evaluated using leave-one-out cross-validation, outperforms the traditional Cumulative probit model. The findings highlight significant differences in the predicted damage probabilities, emphasizing the potential of more flexible fragility curve modeling techniques to improve seismic risk assessments. This study underscores the importance of ongoing evaluation and refinement of modeling techniques to enhance the predictive accuracy and applicability of seismic fragility models in performance-based earthquake engineering.