A1248
Title: Bayesian case influence analysis for spatial autoregressive model
Authors: Cheok Hang Lei - Monash University (Australia) [presenting]
Abstract: A general methodological framework is developed for Bayesian case influence analysis for a spatial autoregressive model. An algebraic representation for the importance sampling weights associated with case-deletion is derived for use with sampled draws from a full data posterior produced via Markov chain Monte Carlo. Once these weights are obtained, a principal components analysis is used on the covariance matrix of log-case deletion weights to produce low dimensional case influence summary plots. The methodology is then applied to artificial data sets, each with a single influential observation. Influential observations are detected in the plots.
