When can categorical variables be treated as continuous? A comparison of robust continuous and categorical SEM estimation methods under suboptimal conditions.

Mijke Rhemtulla, Patricia É. Brosseau-Liard, Victoria Savalei

Psychological Methods · 2012 · 2.5K citations · 58 references

Concepts

TL;DR

The study compares robust normal theory maximum likelihood and robust categorical least squares methods for estimating confirmatory factor analysis models with ordinal variables. The authors simulated data from two CFA models varying category numbers (2–7), sample sizes, latent distributions, and threshold patterns to evaluate the methods. Cat‑LS produced more accurate factor loadings and standard errors, especially with fewer than five categories, while ML performed comparably for factor correlations and fit; cat‑LS was more affected by sample size and non‑normality, whereas ML was more biased by asymmetric thresholds and large loadings, leading the authors to recommend cat‑LS for variables with fewer than five categories and ML when there are five or more categories, small samples, and symmetric thresholds, with either method acceptable for six to seven categories.

Abstract

A simulation study compared the performance of robust normal theory maximum likelihood (ML) and robust categorical least squares (cat-LS) methodology for estimating confirmatory factor analysis models with ordinal variables. Data were generated from 2 models with 2-7 categories, 4 sample sizes, 2 latent distributions, and 5 patterns of category thresholds. Results revealed that factor loadings and robust standard errors were generally most accurately estimated using cat-LS, especially with fewer than 5 categories; however, factor correlations and model fit were assessed equally well with ML. Cat-LS was found to be more sensitive to sample size and to violations of the assumption of normality of the underlying continuous variables. Normal theory ML was found to be more sensitive to asymmetric category thresholds and was especially biased when estimating large factor loadings. Accordingly, we recommend cat-LS for data sets containing variables with fewer than 5 categories and ML when there are 5 or more categories, sample size is small, and category thresholds are approximately symmetric. With 6-7 categories, results were similar across methods for many conditions; in these cases, either method is acceptable.

References

58