I usually perform exploratory factor analysis in Lorenzo-Seva and Ferrando's software FACTOR. However, I have felt recently attracted to the powerful psych package in R. The problem is that when I run exploratory factor analyses in psych, Horn's PA suggests many more factors than expected. I don't know what the underlying calculations in FACTOR are, but I would expect to obtain the same number of suggested factors if the same analysis is being carried out in both programs. Has anybody experienced this problem? Is it a problem at all? Maybe I should restate my question as follows: Is parallel analysis in Lorenzo-Seva and Ferrando's FACTOR program underfactoring?
As a reseraher working in exploratory factor analysis,
Factor software uses three different types of factor analysis including 95% percent, Horn parelel analysis and optimal parelel analysis. Obtained random eigenvalues may be calculated different methods such as permutation of raw data or randomly created virtual samples using same number of factors and variables. Often, obtained eigenvalues differ slightly across different methods. 95% percent eigenvalue calculate Glofferd (1995) approach, he suggested that Horn (1965) may be overfactoring in most cases. Horn (1965) uses mean eingenvalues across virtual samples.
Lastly, above mentioned parelel analysis uses pearson correlation matrix as input optimal paralel analysis uses either pearson or polycoric correlation matrix based on user specification. Different results may be related to two reasons.
Firstly, authors uses different formulas to calculate virtual eigenvalues and some authors use principal components others use principal axis factoring as a result you will obtain different results.
Secondly, ıf authors use the same method, default calculated mean eigenvalues may be different. Factor create 500 virtual samples to obtain mean eigenvalues as a default. I dont know
Revelle's psych package. Different samples may lead different random eigenvalues.
Best wishes.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Software