Revision of empirical NGA-West2 horizontal ground-motion models using fixed-effects and mixed-effects regression and Bayesian inference

This article presents a revision of empirical ground-motion models (GMMs) for the horizontal components of 5%-damped pseudo-spectral acceleration, peak ground acceleration, peak ground velocity, cumulative absolute velocity, and Arias intensity that were originally developed by the first two authors as part of the NGA-West2 Program. The GMMs were refit using fixed-effects (i.e., no random effects) and mixed-effects regression analyses that include either (1) events as a random or repeatable effect or (2) both events and sites as random or repeatable effects. True estimates of the variance components were obtained using Bayesian inference that incorporates uncertainty in the random effects and within-group errors (i.e., the residuals). As a result, the aleatory standard deviations are larger than those calculated directly from the residuals that ignore this uncertainty. Goodness-of-fit metrics show that the GMMs were improved by adding events as a random effect and improved even further by adding sites as a random effect. We found the variance components other than the between-site standard deviation to be magnitude dependent. We recommend use of the GMM that includes both events and sites as random effects because of its appropriate modeling of repeatable effects and its superior goodness-of-fit metrics.