Potential Biases in Mixed-Effects Ground-Motion Models and Variance Components Due to Uncertainty in Random Effects

ABSTRACT Empirical ground-motion models are typically estimated via mixed-effects regression, to account for correlation between records from the same event or from the same site or station. Estimated values of the random effects are often used in further analyses, for example, to develop additional submodels or to investigate physical characteristics of individual events or sites. Such analyses often do not account for uncertainty in the random effects. Using simulations, we show that neglecting these uncertainties can lead to a variety of biases, such as underestimation of variances or biased scaling with predictor variables. We demonstrate that these biases can often be overcome by estimating more complex (e.g., Bayesian) models on the total residuals of a mixed-effects regression.