Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.714957
Title: Efficient and context-dependent Bayesian model selection
Author: Underhill, Nick
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
In this thesis, we argue that the development of a number of context-dependent modifications to standard model selection approaches are warranted from an applied statistical standpoint, where we would generally accept that not only is no candidate model likely to be correct, but also that different models may be preferred for different purposes. To achieve this we propose three types of modification. First, we consider modifications to Bayes factor selection which proceed by specialising the Bayes factor to particular variables of interest, or as an alternative, by placing vague, adaptive priors on variables of less interest. We suggest that, particularly when the analyst wishes to assess models in light of a specific utility, scoring rules have an important role to play, and propose a new bias corrected score based information criterion which can be tailored to the utility at hand. Finally, we present results on a modular assessment framework for ‘big’ models whose components can be expressed in terms of exponential families. Such an approach allows components of the broader model to be assessed individually, and the assessments combined into an overall model score. We believe that this enables the analyst to allow certain judgements about data assessment periods and exchangeability of future data to be accommodated. We conclude with a discussion of areas for further research.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.714957  DOI: Not available
Keywords: QA Mathematics
Share: