Title: Measuring parameter effects in Bayesian inference
Authors: Alastair Young - Imperial College London (United Kingdom) [presenting]
Abstract: A key objective of much of statistical theory concerns the elimination of the effects of nuisance parameters on an inference about an interest parameter. Especially important for statistical practice is quantification of the consequences of including potentially high-dimensional nuisance parameters to provide realistic modelling of a system under study. We consider easily computed measures of nuisance parameter effects in a Bayesian framework. Through decomposition of the Bayesian version of an adjusted likelihood ratio statistic, we propose a computational machinery for analysis of the effects of prior assumptions on nuisance parameters on marginal inference on an interest parameter. Extensions of the techniques allow a formal approach to general sensitivity analysis and evaluations of robustness.