Estimating latent linear correlations from fuzzy frequency tables

TL;DR

This paper introduces a new EM-based method for estimating latent linear correlations from fuzzy frequency tables, which are common in social sciences and other fields dealing with imprecise categorical data.

Contribution

It develops a computational approach using generalized natural numbers and reformulates the estimation problem within an EM framework for fuzzy counts.

Findings

Fuzzy EM-based estimator outperforms standard methods with imprecise data.

Simulation results demonstrate improved efficiency of the proposed method.

Applications show practical utility in social science data analysis.

Abstract

This research concerns the estimation of latent linear or polychoric correlations from fuzzy frequency tables. Fuzzy counts are of particular interest to many disciplines including social and behavioral sciences, and are especially relevant when observed data are classified using fuzzy categories - as for socio-economic studies, clinical evaluations, content analysis, inter-rater reliability analysis - or when imprecise observations are classified into either precise or imprecise categories - as for the analysis of ratings data or fuzzy coded variables. In these cases, the space of count matrices is no longer defined over naturals and, consequently, the polychoric estimator cannot be used to accurately estimate latent linear correlations. The aim of this contribution is twofold. First, we illustrate a computational procedure based on generalized natural numbers for computing fuzzy frequencies. Second, we reformulate the problem of estimating latent linear correlations from fuzzy counts in the context of Expectation-Maximization based maximum likelihood estimation. A simulation study and two applications are used to investigate the characteristics of the proposed method. Overall, the results show that the fuzzy EM-based polychoric estimator is more efficient to deal with imprecise count data as opposed to standard polychoric estimators that may be used in this context.