Exploratory factor analysis is an analytic technique used to determine the number of factors in a set of data (usually items on a questionnaire) for which the factor structure has not been previously analyzed. Parallel analysis (PA) is a technique used to determine the number of factors in a factor analysis. There are a number of factors that affect the results of a PA: the choice of the eigenvalue percentile, the strength of the factor loadings, the number of variables, and the sample size of the study. Although PA is the most accurate method to date to determine which factors are valid, there is still room for improvement. One area of PA not yet examined concerns the degree to which the actual eigenvalue exceeds the random data-based eigenvalue. All methods to date accept that any amount of difference, regardless of how trivial, between the two eigenvalues is enough to satisfy the criterion for a meaningful eigenvalue. However, a more prudent course may be the use of a greater margin than a simple absolute difference between the two eigenvalues. This research examines the accuracy of the simple difference versus the 10% standard. This standard will be the difference of 10% or more between the actual eigenvalue and the parallel analysis eigenvalue. Also examined are the efficacy of this standard in conjunction with the 50th, 90th, 95th and 99th percentiles. Results indicated that the 99th percentile, in conjunction with the 10% standard, is a more accurate method for determining accuracy when using parallel analysis to determine valid factors in a common factor analysis.
Advisor(s) or Committee Chair
Dr. Reagan Brown
Applied Statistics | Industrial and Organizational Psychology | Multivariate Analysis | Numerical Analysis and Computation
Matsumoto, Mandy, "An Investigation of the Accuracy of Parallel Analysis for Determining the Number of Factors in a Factor Analysis" (2017). Honors College Capstone Experience/Thesis Projects. Paper 696.