B1596
Title: A new distance-based framework based on half normal plots for count data
Authors: Darshana Jayakumari - Maynooth University (Ireland) [presenting]
Rafael de Andrade Moral - Maynooth University (Ireland)
Jochen Einbeck - Durham University (United Kingdom)
John Hinde - NUI Galway (Ireland)
Abstract: Model selection is one of the most crucial steps in data analysis. The different diagnostic methods may use quantitative measures such as information criteria and graphical methods based on residuals or other diagnostic quantities. The intention is to extend the graphical model selection method known as half-normal plots of residuals with a simulation envelope. A new distance-based framework that acts as an added quantitative summary to the half-normal plot with a simulated envelope is proposed. This new measure can effectively determine the most appropriate model when closely related models are included and, contrary to information criteria, can be used with marginal models. The framework is formed by calculating the sum of the distances between residuals and the median of the simulated envelope. An extensive simulation study was carried out, taking into account many different scenarios. The results show that the distance framework exhibits a robust performance in finding the true model and is comparable to BIC; in some instances, it even displays superior efficacy.
