Choosing the right correlation for ordinal data: practical recommendations from a simulation study
Frontiers in Applied Mathematics and Statistics · 30.07.2026 · 12
DOI: 10.3389/fams.2026.1823819 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Applied Mathematics and Statistics, online erschienen am 30.07.2026. Verfasst von Nicolai D. Ayasse und 1 weiteren Autor:innen. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Objectives
Polychoric and polyserial correlation coefficients were developed to quantify the relationship between two ordinal items (polychoric) or an ordinal and continuous item (polyserial) where all latent variables underlying the ordinal items and the continuous item are assumed to be normal. The nonparametric Spearman correlation is widely known and utilized for many data distributions, including ordinal and non-normally-distributed variables, thus its use overlaps with conditions under which polychoric and polyserial correlations are appropriate. The current study aimed to compare the accuracy and bias of these correlation types.
Methods
Simulated data were used to identify the most accurate correlation type for different conditions that either met or violated assumptions of polychoric and polyserial correlations.
Results
When their assumptions were met, polychoric or polyserial correlations demonstrated good accuracy and consistently outperformed Spearman correlations. When the polychoric correlation's assumptions were violated, it maintained its accuracy advantage over Spearman correlations. When the polyserial correlation's assumptions were violated, it generally maintained its accuracy advantage over Spearman correlations, with some exceptions; however, the polyserial correlation never demonstrated more than 5 percentage points more absolute percent error than Spearman, while Spearman at times demonstrated as much as 50 percentage points more absolute percent error than polyserial correlations. While two-step and maximum likelihood polychoric correlations performed similarly, as did two-step and maximum likelihood polyserial correlations when their assumptions were met, the estimates of the two polyserial methods increasingly diverged when their assumptions were violated as the degree of skewness increased or the population correlation magnitude increased.
Conclusions
This simulation study demonstrates that under most conditions where items are observed to be skewed, use of polychoric or polyserial correlations is a defensible choice.
Zitieren
APA
Ayasse, N. D., & Coon, C. D. (2026). Choosing the right correlation for ordinal data: practical recommendations from a simulation study. Frontiers in Applied Mathematics and Statistics, 12. https://doi.org/10.3389/fams.2026.1823819
Vancouver
Ayasse ND, Coon CD. Choosing the right correlation for ordinal data: practical recommendations from a simulation study. Frontiers in Applied Mathematics and Statistics. 2026;12. doi:10.3389/fams.2026.1823819
BibTeX
@article{Ayasse2026,
author = {Ayasse, Nicolai D. and Coon, Cheryl D.},
title = {Choosing the right correlation for ordinal data: practical recommendations from a simulation study},
journal = {Frontiers in Applied Mathematics and Statistics},
year = {2026},
volume = {12},
doi = {10.3389/fams.2026.1823819}
}
EUVIMED