Introduction

Items are developed to be indicators of a construct, and preferably, every item represents only one idea or behaviour. When a researcher has items that are influenced by more than one dimension, they are classified as dimension-complex. This is the case of constructs such as core affect that have circumplex structures. According to Guttman (1954), a circumplex is a factorial structure with two axis, and every point represents an item. There are 2 packages in R that can analyse the circumplex structure of a dataset: one is the package “psych†and the other is the package “CircEâ€. While their performance on continuous data is documented, it is unclear how well these packages fare with categorical data.

Objectives

The aim of the study was to verify the performance and results of both packages when polychoric and Pearson correlation matrices are used.

Methodology

The first dataset consisted of N=422 participants that answered an online questionnaire of core affect, and the second dataset was simulated to have the same characteristics (e.g. sample size, number of items, number of categories). We based our interpretations of circumplex structures on:  the correlations heat map from “psych†package and model fit results from “CircE†package.

Results

When the empirical and simulated dataset were analysed based in polychoric correlations, the correlation matrix did not have a “wave†pattern as assumed by a circumplex structure, and the fit indices from “CircE†were not satisfactory, even though the second dataset was simulated to have a circumplex pattern. When both datasets were analysed based in Pearson correlations, the results were acceptable for the empirical dataset with exception of the CFI, and the simulated dataset had a good fit.

Conclusions

We concluded that circumplex analyses performed by these R packages have to be interpreted with caution when polychoric correlations are used.