Incorporating Uncertainty Into Parallel Analysis for Choosing the Number of Factors via Bayesian Methods.

Area of Science:

• Psychometrics
• Statistical modeling
• Data analysis

Background:

• Parallel analysis (PA) is widely used for factor determination but can be inaccurate with small sample sizes or low factor loadings.
• Existing methods often fail to account for the inherent uncertainty in factor number selection.
• Researchers recommend using multiple methods due to PA's limitations.

Purpose of the Study:

• To introduce Bayesian parallel analysis (B-PA) as a novel method to address uncertainty in factor determination.
• To compare the performance of B-PA against frequentist revised parallel analysis (R-PA).

Main Methods:

• Developed and implemented a Bayesian parallel analysis (B-PA) approach.
• Compared B-PA with revised parallel analysis (R-PA) using real and simulated data.
• Analyzed performance under conditions of small sample sizes, low factor loadings, and less distinguishable factors.

Main Results:

• B-PA quantifies uncertainty in factor number selection by providing a probability distribution.
• B-PA offers valuable insights into uncertainty, especially in statistically unfavorable conditions.
• Even when the most probable factor number is incorrect, B-PA can indicate a high probability for the correct number.
• B-PA demonstrated slightly higher accuracy than R-PA when the mode of the probability distribution was used to select the number of factors.

Conclusions:

• Bayesian parallel analysis (B-PA) effectively incorporates and communicates uncertainty in determining the number of factors.
• B-PA offers a more robust approach than traditional methods, particularly in challenging data scenarios.
• The B-PA method provides a probabilistic framework for factor analysis decisions, enhancing reliability.