Main content
Comparing Factor Analysis and Item Response Theory with multimodal latent distributions
Date created: | Last Updated:
: DOI | ARK
Creating DOI. Please wait...
Category: Project
Description: Factor Analysis (FA) and Item Response Theory (IRT) are latent variable modelling techniques with wide usage and acceptance in psychometrics, among other social sciences. Nevertheless, while their performance is optimal with normally-distributed data and latent variables and counting with corrections for non-normal data, little is known about their behavior among non-normal latent distributions. Within the variety of non-normal distributions, multimodal latent distributions are examined manipulating (1) the number of modes they present, and (2) the distance between them. By two simulation studies, results shown an incremental tendency of overestimation in parameters and latent scores while multimodality is more intense (this is, incrementing the number of modes and their distance between them). Further, fit measures were not influenced by latent variable’s multimodality in FA techniques, while IRT techniques were. Therefore, evidence shows researcher’s exposition to an apparent blindness to be modelling multimodal distributions. Even more so, parameter overestimation may lead to aberrant modelling decisions due to its desirability and publishability. Alternative models and assessment guidelines are proposed, alongside with future studies recommendations.