Performance-based grouping methods of bridge classes for regional seismic risk assessment: Application of ANOVA, ANCOVA, and non-parametric approaches

One of the key tasks to enable a regional risk assessment is to group structures with similar seismic performances and generate fragility curves representative of the grouped structures. The grouping has been traditionally performed based primarily on engineering judgment and prior experience. This paper (i) presents an overview of various statistical techniques such as analysis of variance, analysis of covariance, and Kruskal-Wallis test for grouping the bridges of similar performance; (ii) compares the groupings that emerge from the various grouping techniques; and (iii) identifies the method that has more statistical power in creating bridge sub-classes of distinct structural performance. The grouping is achieved by comparing the structural responses of bridge classes obtained from the non-linear time history analysis of bridges. The relative merits of these grouping techniques are discussed with the case study of box-girder bridges in California. Copyright (C) 2017 John Wiley & Sons, Ltd.