| This study assessed the finite sample performance of using polychoric correlations together with a robust WLS estimator (polychoric EFA) for item-level exploratory factor analyses. Results for polychoric EFA were compared to those for item-level EFA using PM-correlations and ML estimation. Based on RMSEA and CFI fit statistics, polychoric EFA was very effective in determining the correct number of factors to extract, particularly when items were asymmetrically distributed and produced very accurate communality estimates when population communalities were strong. However, when communalities were weak, the accuracy of these parameter estimates was strongly influenced by sample size. Despite instances in which PM EFA was as effective as polychoric EFA in identifying the correct number of factors, the communality estimates it produced tended to be negatively biased (relative bias > 10%). Applied researchers are strongly recommended to use polychoric EFA for item level EFAs. |
