On the truncation criteria in infinite factor models. Issue 1 (28th July 2020)

• Record Type: Journal Article
• Authors: Schiavon, Lorenzo; Canale, Antonio
• Is Part Of: Stat. Volume 9:Issue 1(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-07-28
• Subjects: adaptive Gibbs sampling -- cumulative stick‐breaking process -- factor analysis -- multiplicative gamma process -- principal component analysis; Statistics -- Periodicals; 519.2
• DOI: 10.1002/sta4.298
• Languages: English
• ISSNs: 2049-1573

Abstract

Abstract : The number of latent factors, in factor analysis, is typically unknown and motivated by a rich literature on priors distributions, which progressively penalize the number of factors in infinite factor models. Adaptive Gibbs samplers that truncate the infinite factor models are typically used for posterior inference. In this paper, we introduce a novel strategy to adaptively truncate the number of factors that is more interpretable, stable and consistent, with respect to standard approaches.