EFA interpretation
Message/Author
 Liz Woodruff posted on Tuesday, April 26, 2011 - 8:55 am
I'm a bit lost as to how to interpret the output for my EFA. Can you point me in a direction?
 Linda K. Muthen posted on Tuesday, April 26, 2011 - 9:44 am
EFA is discussed in the Topic 1 course video and handout.
 Liz Woodruff posted on Tuesday, April 26, 2011 - 9:52 am
OK, could I send you my output to get your help. It seems that the best fitting model is the one with 9 factors. However, I have one variable that has a negative residual variance (i.e., LogY1). Does this mean my model is inadmissible? If so, am I at a loss with my data?

Thank you.
 Linda K. Muthen posted on Tuesday, April 26, 2011 - 11:00 am
A negative residual variance means the model is inadmissible and that you are extracting too many factors. You should have some idea of how many factors to expect based on how the items were created. EFA should be used with items that were carefully created to measure certain dimensions.
 Alexandra Alipan posted on Thursday, November 26, 2015 - 5:38 pm
Hi Bengt and Linda,

I am using mplus 7.4 and trying to conduct an EFA with 24 ordinal items using WLSMV estimation technique. I have read the manual, handout and listened to the video you provide on this topic, which has really helped. However, I am a beginner when it comes to Mplus and still have a few questions on what output is available for EFA:

1) Can you obtain Standardized Residuals in EFA, like you can in CFA (so I can check if any problematic items / values above 2.00)?

2) I requested MOD(ALL 0) so I can see if any changes need to be made to the model – I am used to looking at these when performing a CFA. However, the mod indices output in EFA looks different to CFA. In the EFA I got a Theta Matrix. Should I be looking at high values in the modification indices/ Theta matrix to help decide on model modifications such as separating items into another factor or removal of items? (I will make sure that any modifications are based on content of items/ justified by theory).

3) Should I mostly be inspecting the correlation matrix, eigenvalues (greater than one and scree plot), chi-square, RMSEA, SRMR, CFI, TLI, checking for negative residual variances, geomin rotated loadings, and interpretability to decide on whether modifications need to be made?

4) Is there anything else I should be requesting in the output command of input file? Or looking at in the output?
 Bengt O. Muthen posted on Thursday, November 26, 2015 - 6:32 pm
1) Check if you get it when requesting RESIDUALS

2) Unlike CFA, for EFA there are no restrictions on the loadings or factor covariance matrix that can be freed up. The only restrictions are for Theta elements - and looking for that can be helpful.

3)-4) Yes and you can also check the Parallel = 50 option.
 Alexandra Alipan posted on Monday, November 30, 2015 - 1:55 pm

1) I requested RESIDUAL, however, I only get Model Estimated Correlations and Residuals for Correlations? There were no values above 2.00 but when I request STDYX in a CFA there are many in the standardised residual matrix. (If I type in STDYX to the output section for EFA a warning appears saying that standardised is not available for EFA). So should I just look at the theta matrix, correlation matrix, and factor loadings to check for problematic items?

4) I tried to request Parallel = 50 but a warning appeared: The PARALLEL option is not available for TYPE=EFA with categorical outcomes.

Many thanks,
Alexandra
 Bengt O. Muthen posted on Monday, November 30, 2015 - 5:30 pm
1) Modindices for Theta are more useful than residuals.

4) True.