EFA with covariate PreviousNext
Mplus Discussion > Exploratory Factor Analysis >
Message/Author
 Raheem Paxton posted on Wednesday, April 07, 2010 - 2:18 pm
Hi,

I'm interested in running an EFA consisting of categorical variables with a continuous covariate. I'm not sure whats going on, but for some strange reason its not working out as I expected. I'm hoping that you can guide me in the analysis. The coding that I am using is below. Thanks in advance

ANALYSIS: TYPE = BASIC;
ROTATION=GEOMIN;

F1-F4 BY BEER - CALORIE;

F1-F4 ON AGE_COHE
OUTPUT: MOD(10) STANDARDIZED;
PLOT: TYPE IS PLOT2;
 Linda K. Muthen posted on Wednesday, April 07, 2010 - 4:26 pm
Please see the Version 5.1 Examples and Language Addendum on the website with the user's guide. You should not be using TYPE=BASIC. This is for sample statistics.
 luk bruyneel posted on Wednesday, April 10, 2013 - 6:14 am
Hi,
My question is whether chiefs perceive factors better/worse than their staff. To compare factors, I check whether the factor model holds for the groups. Since the instrument has previously only been used among staff, I do this exploratory.
I ran EFA using ESEM i.e. MODEL: F1-F6 by A_1_1-A_1_32(*1). This resulted in RMSEA=0.048, CFI=0.963, TLI=0.945 and chi-square 1076(351). The factor structure made theoretical sense.
Next I added my covariate ‘f1-f6 ON function’ to the model statement. This resulted in RMSEA=0.045, CFI=0.966, TLI=0.948 and chi-square 993(345). The factor loadings were stable. Function was significant for 4 of 6 factors.
Lastly I added direct effects ‘A_1_1-A_1_32 ON function@0’. This resulted in identically the same model fit information. M.I.s were low (around 10).
Q:
-Can I conclude from this at the same time that it appears that my model is close enough to measurement invariance to make comparisons, but that it can already be seen that there is population heterogeneity (so explore this further)?
-Is it plausible that after adding the direct effects my model fit stays identical?
-I want to make these comparisons in a multilevel approach (I have 87 clusters and within each cluster a chief). Can I take factor means (or do you suggest another measure) and proceed in a multilevel regression?
Kind regards
 Linda K. Muthen posted on Wednesday, April 10, 2013 - 11:29 am
I think 10 is rather high given that the cutoff for signficance is 3.4 for one degree of freedom.

No, adding direct effect may change model fit.

I would do a TWOLEVEL CFA. I would not use factor scores.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: