CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Assessing uncertainty in posterior intercepts from random effect models Authors:  Jochen Einbeck - Durham University (United Kingdom) [presenting]
Nick Sofroniou - ()
Abstract: Posterior intercepts from random effect or variance component models provide an attractive tool for the ranking or `league tabling' of cluster-level units; at the same time allowing for adjustment to covariates (on the individual or the cluster level) through the inclusion of fixed effects. For instance, one may use such intercepts for the ranking of region--wise mortality rates (where the crude, regional rates are often unreliable due to small observed counts) or for the construction of educational league tables from complex sample surveys. However, it is essential that the variability of these posterior intercepts can be assessed accurately, in order to establish whether two cluster-level units can actually be distinguished in terms of their ranking. This question is examined in the context of the `nonparametric maximum likelihood' approach to random effect modelling, in comparison to glmer and fixed effect models. The motivating application is the PIAAC literacy survey.