| Message/Author |
|
|
|
|
It isn't clear to me whether Mplus bases its parallel analysis on eigenvalues from EFA or PCA. My reading of the literature is that it is best to use PCA eigenvalues when using parallel analysis to make decisions about the number of factors to extract (even when one plans to use EFA when extracting and interpreting factors).
I'd be grateful for any clarification that folks can provide.
Thanks!
Skyler |
|
|
|
|
| We use eigenvalues from the observed correlation matrix. This is the same as PCA would use if the variables are standardized. |
|
|
|
|
| Thanks much for your quick response! |
|
|
| Margarita posted on Sunday, February 15, 2015 - 4:02 pm
|
|
|
Dear Dr. Muthen,
Unfortunately I do not have the new version of MPlus, and so I have to conduct a parallel analysis using the SPSS syntax proposed by O'Connor.
I was wondering if you know whether it is preferable to use Principal Components Eigenvalues or FA eigenvalues to make decisions as to how many factors to extract when my plan is to use EFA afterwards for factor interpretation?
Thank you in advance,
Best,
Margarita |
|
|
|
|
| In the Mplus parallel analysis the eigenvalues are for the sample correlation matrix, not the estimated correlation matrix so it's using principal component eigenvalues. |
|
|
| Margarita posted on Monday, February 16, 2015 - 7:37 am
|
|
|
| Thank you for your prompt reply! |
|
|
|
|
| A couple questions. I've read that parallel analysis programs produce data sets with normally distributed random numbers. (1) I wonder if eigenvalues generated by these data sets can be meaningfully compared to observed data with high levels of skewness or kurtosis? If not, is there a way to produce random data sets that might approximate the structure of the observed data? (2) With nonnormal data, I wonder if the MAP test may be more appropriate? Is it appropriate to use the MAP test in an EFA context? Does Mplus provide a way of doing this? |
|
|
|
|
(1). Q1: I am not sure. Q2. Not that I know.
(2) MAP is a procedure for getting factor scores. I don't know what a "MAP test" is. |
|
|
|
|
| Thanks for the quick reply. I am referring to Velicer's (1976) Minimum Average Partial (MAP) Test. I found a paper by O'Connor (2000) that provides SPSS and SAS syntax for both a parallel analysis and Velicer's MAP test. I was hoping there might also be a means for doing this in Mplus. |
|
|
|
|
| I read in another post that parallel analysis does not work well with categorical data. Is it safe to assume that the results might also be misleading for items measured on 5-point rating scales? |
|
|
|
|
No, Mplus doesn't do that MAP test.
The problem with parallel analysis for categorical variables is due to using poly choric correlations. With 5-point Likert scales you can typically treat the variables as continuous and the problem isn't there. |
|
|
| Paula Vagos posted on Tuesday, April 14, 2015 - 3:30 am
|
|
|
Hello, I am trying to do a parallel analysis in mplus and keep getting this error message:
*** ERROR in ANALYSIS command
Unknown option:
PARALLEL
My syntax is:
VARIABLE: NAMES ARE U1-U6;
CATEGORICAL ARE U1-U6;
ANALYSIS: TYPE = EFA 1 3;
ROTATION IS CF-VARIMAX;
PARALLEL = 50;
PLOT: TYPE = PLOT3;
Could you please help me? |
|
|
| Paula Vagos posted on Tuesday, April 14, 2015 - 5:20 am
|
|
|
| on a follow-up note, taking out the "categorical are u1-u6" part of the syntax did not fix the problem... |
|
|
|
|
| Perhaps you are using a version before that option was added. |
|
|
|
|
Hello, and thank you for your reply. I was using version 7.0 and realized that a search within mplus did not returned that command. So I am now working on the demo 7.3 version and I got the file.
The only way to read this file is through R, right?
Thank you |
|
|
|
|
sorry, just answered my own question  |
|
|
| Back to top |
