Latent variable selection based on grouped Bayes factors
Over the last few years, the use of latent variables has increased in popularity due to the urge of dealing with unobserved or hypothetical constructs. As these can only be approached indirectly through measurable indicators, a variable selection technique which could handle such aspects have become more and more important. From the Bayesian model uncertainty framework, we propose the grouped Bayes factor to measure the importance of the latent variable while explaining the variability of the response of interest. As the prior plays an essential role in the performing, we compare the behaviour of different ones. In particular, we focus on priors which take into account the relation between the measurable indicators that make up the latent variable.
Palabras clave: Latent variables Bayes factor Prior distributions Variable selection