CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Case influence diagnostics from Bayesian empirical likelihood posteriors Authors:  Catherine Forbes - Monash University (Australia) [presenting]
Abstract: The use of case influence diagnostics for moment condition models in a Bayesian empirical likelihood (EL) context is explored. Such models are common in Economics and related disciplines where theory implies a set of moment constraints, yet a full generative probability model is not specified. Case influence diagnostics provide a way to consider how well the moment condition model captures the variability in the data, and whether any of the observations exert a strong influence in the determination of the EL weights. The question of the influence of individual cases or of groups of cases is especially important in this context because the empirical moment conditions can be greatly affected by the presence of extreme observations. Following previous work, we develop case-influence diagnostic measures for Bayesian EL and consider appropriate low-dimensional summaries of case deletion which may be helpful in settings with many observations, parameters or moment conditions.