A1481
Title: Non-asymptotic bounds for group sequential quasi-MLE, misspecified models, and dependence
Authors: Jay Bartroff - University of Texas at Austin (United States) [presenting]
Abstract: Results based on Stein's method are presented for the limiting multivariate normality of joint parameter estimates under group sequential sampling in a setting that allows dependence within groups. In a general parametric regression setting in which the n-th observation is the n-th subject's response regressed on their covariates, the quasi-MLE (QMLE) setup is considered in which the specified quasi-likelihood may differ from the true data generating process. Existing asymptotic normality results for the fixed-sample QMLE are extended to the group sequential setting, and non-asymptotic bounds to this limit are obtained via existing Stein's method tools. The results quantify convergence in QMLE, misspecified models, and under dependence, when the quasi-likelihood assumes independence but the true likelihood does not. As an example, these results are applied to the Poisson generalized linear mixed model.
