2012/10/16 by Ricardo Silva, Silva, Ricardo
Computer Science · Mathematics · #Bayesian Modeling and Causal Inference #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Statistical Methods and Bayesian Inference #Statistical Methods and Inference #Statistical Methods in Epidemiology
paper · pdf · doi:10.48550/arxiv.1210.4905
openalex publication_date 2012/10/16 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
Latent variable models are used to estimate variables of interest quantities\nwhich are observable only up to some measurement error. In many studies, such\nvariables are known but not precisely quantifiable (such as "job satisfaction"\nin social sciences and marketing, "analytical ability" in educational testing,\nor "inflation" in economics). This leads to the development of measurement\ninstruments to record noisy indirect evidence for such unobserved variables\nsuch as surveys, tests and price indexes. In such problems, there are\npostulated latent variables and a given measurement model. At the same time,\nother unantecipated latent variables can add further unmeasured confounding to\nthe observed variables. The problem is how to deal with unantecipated latents\nvariables. In this paper, we provide a method loosely inspired by canonical\ncorrelation that makes use of background information concerning the "known"\nlatent variables. Given a partially specified structure, it provides a\nstructure learning approach to detect "unknown unknowns," the confounding\neffect of potentially infinitely many other latent variables. This is done\nwithout explicitly modeling such extra latent factors. Because of the special\nstructure of the problem, we are able to exploit a new variation of composite\nlikelihood fitting to efficiently learn this structure. Validation is provided\nwith experiments in synthetic data and the analysis of a large survey done with\na sample of over 100,000 staff members of the National Health Service of the\nUnited Kingdom.\n