Investigating Parallel Analysis in the Context of Missing Data: A Simulation Study Comparing Six Missing Data...

Area of Science:

• Psychology
• Statistics
• Data Science

Background:

• Exploratory factor analysis (EFA) is vital in psychology for identifying latent variables and developing questionnaires.
• Self-report questionnaires frequently contain missing values, complicating EFA, particularly factor retention.
• Limited research addresses handling missing data within the EFA factor retention process.

Purpose of the Study:

• To evaluate the performance of six distinct missing data methods in the context of EFA factor retention.
• To compare the accuracy of parallel analysis as a factor retention criterion across different imputation techniques.
• To identify optimal strategies for managing missing data in EFA.

Main Methods:

• A simulation study was conducted to compare expectation-maximization, predictive mean matching, Bayesian regression, random forest imputation, complete case analysis, and pairwise complete observations.
• Data were simulated with varying correlated/uncorrelated factor structures, variable counts, observation numbers, and missing data mechanisms.
• Two procedures for combining multiply imputed datasets were examined.

Main Results:

• No single missing data method proved universally superior across all simulated conditions.
• Random forest imputation demonstrated the best performance in most scenarios for factor retention accuracy.
• Complete case analysis and pairwise complete observations were frequently outperformed by multiple imputation methods.

Conclusions:

• Random forest imputation offers a robust approach for handling missing data in EFA factor retention.
• Applying parallel analysis to an averaged correlation matrix enhances the effectiveness of imputation methods.
• Further research is needed to refine missing data handling techniques in complex EFA scenarios.