Correlation between intercept and slope PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Seung Bin Cho posted on Tuesday, February 27, 2007 - 8:58 pm
When there is high (almost perfect) correlation between intercept and slope in growith curve model, the model fails to run or gives improper solution.
When I ran the model with random options and analysis it failed to run with saying "nonpositive defninite estimated correlation matrix".
I guess this is the natural reaction of mplus because intercept or slope factor has negative or 0 residual variance if the correlation between them is close to 1(or -1).
Is there any remedy for this problem?
 Bengt O. Muthen posted on Tuesday, February 27, 2007 - 9:11 pm
Assuming that the random slope has a significant variance, you could try centering at another time point.
 Kätlin Peets posted on Friday, February 22, 2008 - 9:43 am
I have a similar question. My slope and intercept are highly correlated (more than .80). So far I have done it so, that the loading of the variable at Time 1 is set to 0). However, when I center so that intercept is the last time point, the correlation is even higher. It would not be such a problem, but I am trying to examine whether individual differences in the slope and intercept predict differences in other variables. So...what would be the solution? Should I try to do growth mixture analysis?
Thank you.
 Kätlin Peets posted on Friday, February 22, 2008 - 10:02 am
I forgot to add...the problem is that when I include both the intercept and slope in the model, all the paths are nonsignificant. However, the paths from the intercept become significant when I fix the paths from the slope to be zero (and the other way around).
 Linda K. Muthen posted on Friday, February 22, 2008 - 2:29 pm
Try centering at the average. If that does not help, please send your input, data, output, and license number to support@statmodel.com.
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: