Abstract
Generalised linear mixed models have become a mainstream statistical tool, but many of their properties have yet to be fully understood. The first part of the talk will be a review of various attempts that have been proposed over the past four decades to extend restricted maximum likelihood estimation to generalised linear mixed models. This estimation method is a widely accepted and frequently used for fitting linear mixed models, with its principal advantage being that it produces unbiased estimates of the variance components. However, the concept of restricted maximum likelihood does not immediately generalise to the setting of non-normally distributed responses, and it is not always clear the extent to which, either asymptotically or in finite samples, such generalisations reduce the bias of variance component estimates. This part of the talk will be followed by a discussion of second-term asymptotic theory for the asymptotically harder-to-estimate parameters in independent cluster generalized linear mixed models. Improved accuracy of statistical inference, optimal design and sample size calculations are some consequences of the proposed theory.
This is joint work with Dr. Aishwarya Bhaskaran (Macquarie University), A./Prof. Francis Hui (Australian National University), Prof. Matt Wand (University of Technology Sydney) and Prof. Alan Welsh (Australian National University).
Collegamento Microsoft Teams
Organizzione
Maria Pia Feser