Imperial College London
Alternate

ProfessorSylviaRichardson

Faculty of Medicine, School of Public Health

Visiting Professor
 
 
 
//

Contact

 

+44 (0)20 7594 3336sylvia.richardson Website

 
 
//

Assistant

 

Miss Sonia Kharbotli +44 (0)20 7594 3319

 
//

Location

 

161Norfolk PlaceSt Mary's Campus

//

Summary

 

Publications

Citation

BibTex format

@article{Ginestet:2012:10.1016/j.spl.2011.12.015,
author = {Ginestet, C and Best, N and Richardson, S},
doi = {10.1016/j.spl.2011.12.015},
journal = {Statistics and Probability Letters},
pages = {859--863},
title = {Classification loss function for parameter ensembles in Bayesian hierarchical models},
url = {http://dx.doi.org/10.1016/j.spl.2011.12.015},
volume = {82},
year = {2012}
}
Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - In this note, we consider the problem of classifying the elements of a parameter ensemble from a Bayesian hierarchical model as above or below a given threshold, C. Two threshold classification losses (TCLs)–termed balanced TCL and p-weighted TCL, respectively–are formulated. The p-weighted TCL can be used to prioritize the minimization of false positives over false negatives or the converse. We prove that, as a special case of a more general result, the p-weighted and balanced TCLs are optimized by the ensembles of unit-specific posterior (1−p)-quantiles and posterior medians, respectively. In addition, we also relate these classification loss functions on parameter ensembles to the concepts of posterior sensitivity and specificity. Finally, we discuss the potential applications of balanced and p-weighted TCLs in Bayesian hierarchical models, and how TCLs could be used to extend existing loss functions currently used for point estimation in parameter ensembles.
AU  - Ginestet,C
AU  - Best,N
AU  - Richardson,S
DO  - 10.1016/j.spl.2011.12.015
EP  - 863
PY  - 2012///
SP  - 859
TI  - Classification loss function for parameter ensembles in Bayesian hierarchical models
T2  - Statistics and Probability Letters
UR  - http://dx.doi.org/10.1016/j.spl.2011.12.015
VL  - 82
ER  -
Download