Methodological simulation reveals improved recovery of sparse factor loadings in multivariate models, highlighting enhanced Bayesian parameter estimation.
Key Points
To develop a novel Bayesian penalized-complexity prior based on expected Kullback–Leibler divergence that effectively encourages sparse loading structures in exploratory factor analysis.
Formulated a prior distribution on a function of the expected Kullback–Leibler divergence between a sparse base model and a dense full model.
Evaluated performance using simulation benchmarks against existing Bayesian procedures and demonstrated factor selection on empirical mental ability test scores.
The expected penalized-complexity prior recovered sparse factor loading structures with greater accuracy across diverse simulation conditions compared to standard methods.
Demonstrated an effective practical strategy for selecting the true number of latent factors in cognitive test score applications.