Hypothesis testing via posterior-test-based Bayes factors

Hypothesis testing via p-value has been criticized in recent years. Bayes factors (BFs) have been tipped as a possible replacement of p-value for hypothesis testing. However, the standard BFs suffer from some theoretical and practical difficulties. For example, they are not well defined under improper p...

Full description

Saved in:
Bibliographic Details
Main Authors: LI, Yong, WANG, Nianling, Jun YU, ZHANG, Yonghui
Format: text
Language:English
Published: Institutional Knowledge at Singapore Management University 2023
Subjects:
Online Access:https://ink.library.smu.edu.sg/soe_research/2671
https://ink.library.smu.edu.sg/context/soe_research/article/3670/viewcontent/PLR36.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Singapore Management University
Language: English
Description
Summary:Hypothesis testing via p-value has been criticized in recent years. Bayes factors (BFs) have been tipped as a possible replacement of p-value for hypothesis testing. However, the standard BFs suffer from some theoretical and practical difficulties. For example, they are not well defined under improper priors and are subject to Jeffreys-Lindley-Bartlett’s paradox under vague priors. Moreover, they are difficult to compute for many models. In this paper, we propose to compare sampling distributions of the posterior-test-based statistics for hypothesis testing. Two posterior-test-based BFs are constructed from the posterior version of the likelihood ratio test and the Wald test, respectively. Under regularity conditions, we show that the new methods can avoid the p-hacking problem and the problems in the standard BFs. The advantages of the proposed methods are investigated using several simulation and empirical studies.