Model-Selection-Based Approaches to Identifying the Optimal Number of Factors in Multilevel Exploratory Factor Analysis

ABSTRACT

This study examined the accuracy of commonly used model fit indexes in identifying number of factors in multilevel exploratory factor analysis using Monte Carlo simulations. Multilevel data were generated according to different scenarios of factor structures: cluster numbers, cluster sizes, and intraclass correlation coefficient (ICC) conditions. The results showed that when using the model-based approach, most of the commonly used fit indexes could identify the correct number of factors at the lower level, except for the within-level SRMR, AIC, CFI, and the within-level CFI. However, most of the fit indexes extracted fewer factors at the higher level when ICC and sample size were small. When using the design-based approach (assuming the same factor structure across levels), most of the fit indexes were able to identify the correct number of factors, except for SRMR, AIC, and ∆AIC, and CFI.

KEYWORDS:

• Clustering
• model fit indexes
• information criteria
• model selection
• multilevel exploratory factor analysis

Disclosure statement

We have no known conflict of interest to disclose.

Notes

¹ We also examined factor loading estimates based on the geomin rotation at each level when the correct models were identified. In general, the estimated factor loadings were consistent with the population model when the correct model was identified.