Two-level non-linear growth model PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
 Jinxin ZHU posted on Tuesday, September 27, 2016 - 6:27 pm
Dear Dr. Muthens,

I want to ran a two-level growth model, and have non-linear growth: three time points, free the time score at third time points at both student and school level.

I found that the starting value of the time score affect the final estimates. Do you have any suggetion?

Is there any example using Mplus to ran a two-level non-linear growth model?

Thank you so much!
 Bengt O. Muthen posted on Wednesday, September 28, 2016 - 5:41 pm
Be sure to hold the free time score equal across the two levels.

To avoid starting value effects you can try Starts = 10.
 Jinxin ZHU posted on Wednesday, September 28, 2016 - 6:43 pm
Thank you so much.

Would you please suggestion how to constrain the free time score equal across the two levels?

I tried to label the time score but was told that "Equality/parameter labels are not allowed in a growth statement."

Also, what will happen if they are not set to be equal across the two levels?

Thank you so much!
 Jinxin ZHU posted on Thursday, September 29, 2016 - 12:59 am
Dear Dr. Muthen,

I have solved the problem of labeling using the 'By' option, but still got the warning£º


I changed the stating value and it helped.

However, I found that either the starting value of time score or staring value set using "Starts =" affect the estimates and do not know which result I should use.

Would you please kindly suggest what I should do and why different starting value result in different estimation?

Thanks again!
 Bengt O. Muthen posted on Thursday, September 29, 2016 - 8:41 am
Choose the solution with the highest loglikelihood value.
 Jinxin ZHU posted on Thursday, September 29, 2016 - 6:20 pm
Thank you so much!

Is there any example specifying the syntax of multilevel non-linear growth model, as well as multiple process multilevel non-linear growth model?
 Bengt O. Muthen posted on Friday, September 30, 2016 - 3:52 pm
No, but you can do most anything using BY using estimated time scores to capture non-linearity.

See also articles by Grimm on our website under Growth Modeling - this takes a more general approach to non-linear growth.
 Jinxin ZHU posted on Thursday, February 23, 2017 - 7:47 pm
When running a Two-level parallel process latent growth model, the default estimator is MLR. However, I got a warning that THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE.

1. When I chose Estimator = ULSMV, the problem solved. Will the result be affected by using different estimator?

2. I found the model data fit did not agree with each other. The RMSEA less than 0.05, while CFI and TLI around 0.85. Are there any guiding suggestions regarding model-data fit for multilevel models?

Thank you very much!
 Linda K. Muthen posted on Friday, February 24, 2017 - 4:11 pm
1. Please send the two outputs and your license number to
2. Model fit statistics do not always agree with each other.
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