Reconciling the Bayes Factor and Likelihood Ratio for Two Non-Nested Model Selection Problems

TL;DR

This paper establishes a theoretical link between the Bayes Factor and likelihood ratio for non-nested model selection, providing insights valuable for forensic science and clarifying methodological differences.

Contribution

It derives an expression showing the Bayes Factor as an expected value of the likelihood ratio under a specific posterior distribution, bridging two key statistical methods.

Findings

Bayes Factor equals the expected likelihood ratio under a posterior distribution.

The relationship clarifies the connection between Bayesian and likelihood-based approaches.

Practical implications for forensic science model comparison.

Abstract

In statistics, there are a variety of methods for performing model selection that all stem from slightly different paradigms of statistical inference. The reasons for choosing one particular method over another seem to be based entirely on philosophical preferences. In the case of non-nested model selection, two of the prevailing techniques are the Bayes Factor and the likelihood ratio. This article focuses on reconciling the likelihood ratio and the Bayes Factor for comparing a pair of non-nested models under two different problem frameworks typical in forensic science, the common-source and the specific-source problem. We show that the Bayes Factor can be expressed as the expected value of the corresponding likelihood ratio function with respect to the posterior distribution for the parameters given the entire set of data where the set(s) of unknown-source observations has been generated according to the second model. This expression leads to a number of useful theoretic and practical results relating the two statistical approaches. This relationship is quite meaningful in many scientific applications where there is a general confusion between the various statistical methods, and particularly in the case of forensic science.