Testing for multicollinearity PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Sophia Fikuart posted on Thursday, June 15, 2017 - 10:29 am
Hello Dr. Muthen,

I am running a multilevel model (I compare the Intercept-only-Modell with a Model 1 (Control Variables), a Model 2 (Direct Effects of Predictor Variables) and a Model 3 (Adding Interaction Terms). Some results of Model 3 seem to be weird (a standardized estimate for an interaction term in Model 3 over 1.0 (1.3) and the Chi-Square-Difference Test (comparing Model 3 to Model 2) seems not to be possible because the
-2*LogLikelihood Value of Model 3 is HIGHER than the -2*LogLikelihood Value of Model 2.

Question 1:
Is it possible that the -2*LogLikelihood Value of Model 3 is higher than the one of Model 2, even if the models are nested
(Model 3 includes additionally the interaction terms)?

When I look at the correlation between my predictor variables I subsume that there ist multicollinearity between the interaction term (which has this weird estimate)and a predictor.
A there - like in SPSS - any opportunities in Mplus to test for multicollinearity through a Test like VIF or Tolerance?

...As I run the model in Mplus, I am not sure how to transfer the results to SPSS to check for multicollinearity...

Thank you very much.

Kind regards,

 Bengt O. Muthen posted on Thursday, June 15, 2017 - 6:00 pm
Q1: No - send to Support along with your license number.

Q2: You can have standardized values greater than 1 if you have more than one predictor. But as you say, this may point to high correlations among the predictors. Perhaps you didn't center the 2 variables going into the interaction. No VIF in Mplus.
 Sophia Fikuart posted on Sunday, June 18, 2017 - 1:48 am
Thank you very much!

I send it to support. How long does it approximately take to get an answer from your support team?


 Carolina Sinning posted on Thursday, November 07, 2019 - 9:08 am

I have the case of having two highly correlated moderator variables. Actually, I run the moderating effects in two different models due to model complexity issues (showing different explained variances). But is there anything like VIF or tolerance values for Moderators to Show the importance of Accounting for both Moderators? How would you suggest to proceed?

Thank you very much in Advance!
 Bengt O. Muthen posted on Thursday, November 07, 2019 - 3:55 pm
This question is suitable for SEMNET.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message