Parallel analysis and bifactor EFA
Message/Author
 Philippe Golay posted on Friday, September 28, 2012 - 5:52 am
I wonder whether the ideal number of factor to retain with a bifactor efa is N or N+1, given N is the number of factor suggested by parallel analysis. Because a perfect 5 factor oblique structure would be best recovered by 6 (and not 5) factors with a bi-factor efa (1 general & 5 specific). What would you recommend ?

Best regards

Philippe
 Bengt O. Muthen posted on Monday, October 01, 2012 - 9:15 am
Bi-factor EFA with 1 general and m-1 specific factors has the same model fit (same ML loglikelihood and number of parameters) as an m-factor regular EFA. So bi-factor EFA with a total number of m factors is just a different rotation than the m-factor EFA.

In your case, if you have perfect fit for 5 EFA factors, you would also have that same perfect fit for a bi-factor model with 1 general and 4 specific factors. So, if parallel analysis suggests 5 factors, you can go with bi-factor using 1 general and 4 specifics; i.e. asking for 5 factors when doing bi-factor EFA.
 Philippe Golay posted on Tuesday, October 02, 2012 - 12:28 am
Thank you very much. That makes a lot of sense.
 Mplus User posted on Thursday, February 22, 2018 - 10:54 am
I know this is available somewhere and I can't seem to find it. What is the syntax for specifying a bifactor EFA?

VARIABLE: NAMES ARE x1-x16;
USEVARIABLES x1-x16;
ANALYSIS: TYPE = EFA 1 6;
 Bengt O. Muthen posted on Thursday, February 22, 2018 - 3:27 pm
UG Chapter 4 covers EFA. UGex 4.7 shows bi-factor EFA.
 Mplus User posted on Monday, February 26, 2018 - 7:45 am
Thank you!

For a bifactor EFA, do you recommend the BI-GEOMIN rotation or the BI-CF-QUARTIMAX rotation?

I have a model with 2 specific correlated factors and 1 general factor, and I have interval data.
 Bengt O. Muthen posted on Monday, February 26, 2018 - 7:59 am
No recommendation - they shouldn't be too different. Interpretability should be the primary guide.