Bayesian variable selection
Definition
Bayesian variable selection refers to the use of Bayes factors and Bayesian model averaging techniques to test for conditional independence between pairs of variables in graphical models, requiring specification of prior distributions on both network parameters and network structure. In the context of Markov Random Field models for ordinal and binary data, this approach uses the inclusion Bayes factor to determine whether edges should be present or absent in a network. The sensitivity of the inclusion Bayes factor to prior specification—particularly the scale of priors on edge weights and choices regarding network structure configuration—is critical for making informed, evidence-based inferences about conditional independence relationships in psychological data.
Sources: Sekulovski et al. (2024)
Related Terms
Applications
Bayesian Variable Selection and Conditional Independence Testing
Bayesian variable selection, operationalized through the inclusion Bayes factor, is designed to test for conditional independence between pairs of variables in network psychometric models. The method allows researchers to quantify evidence for or against the presence of edges while accounting for uncertainty about both network structure and edge weight parameters.
Sources: Sekulovski et al. (2024)
Bayesian Variable Selection and Prior Specification
The effectiveness of Bayesian variable selection depends critically on informed choices about prior distributions for both network structure and edge weight parameters. The scale of these priors substantially affects the inclusion Bayes factor's sensitivity and ability to distinguish between the presence and absence of edges, making prior specification a fundamental concern for applied researchers.
Sources: Sekulovski et al. (2024)
