Time points for Cubic GMM
Message/Author
 Jonathan Codell posted on Thursday, February 07, 2013 - 2:24 pm
I have 12 data points (T1-T12) and would like to model for potential cubic growth in GMM. I managed to fit a decent linear and quadratic growth models but am unclear on to determine the appropriate times in the syntax for a cubic model:

i s | T1@0 T2@1 T3@2 T4@3.........T12@11

i s q | T1@0 T2@.01 T3@.04 T4@.09......T12@1.21

What would the appropriate syntax time be for a cubic model?

i s q cub | T1@? T2@?..........T12?

Thank you for your assistance.
 Linda K. Muthen posted on Thursday, February 07, 2013 - 2:31 pm
Yes, just add a fourth growth factor.
 Jonathan Codell posted on Thursday, February 07, 2013 - 2:46 pm
Thanks for the quick response. Can you clarify exactly what you mean though. As shown in my initial post I already have four growth factors listed (i s q cub) before the "|" symbol, but I am unclear on what I should list after the "@" symbol for each of the 12 time points. Thank you for your explanation.
 Linda K. Muthen posted on Thursday, February 07, 2013 - 3:48 pm
You list the linear time scores after the | symbol. I just noticed you did this incorrectly for the quadratic model. The time scores are always the linear and Mplus computes the others. Keep the on a small scale like 0 .1 .2 etc.
 Jonathan Codell posted on Thursday, February 07, 2013 - 11:33 pm
Thank you for the clarification. It looks like I was making this more complicated than needed. I really appreciate your help.
 Kiki van Broekhoven posted on Monday, July 03, 2017 - 11:16 pm
I have 7 data points and would like to investigate various forms of growth (linear, quadratic, cubic). I wondered what would be the best way to do this, that is: is there a specific order to introduce the various growth factors? So first try linear, than linear and quadratic, and so on? (and then check whether mean and variance of the growth factors are different from zero) Or should I be guided by other things?
 Kiki van Broekhoven posted on Friday, July 07, 2017 - 7:38 am
Could someone help me with this question? Thank you in advance.
 Bengt O. Muthen posted on Friday, July 07, 2017 - 4:35 pm
If theory can't guide you, yes, first try linear, than linear and quadratic, and so on and then check whether mean and variance of the growth factors are different from zero.
 Kiki van Broekhoven posted on Tuesday, July 25, 2017 - 3:25 am
Thank you for your reply. If I understand correctly, this could imply that e.g. a linear, quadratic AND cubic growth factor are needed when all of means and/or variances are different from zero?
 Bengt O. Muthen posted on Tuesday, July 25, 2017 - 5:19 pm
Right.
 Kiki van Broekhoven posted on Monday, August 28, 2017 - 5:52 am
Are 7 time points enough to model potential cubic growth? Or should such a model preferably include more measurement points?/isn't there a minimum number of measurement points to consider cubic growth
 Linda K. Muthen posted on Monday, August 28, 2017 - 11:33 am
Four time points is enough for quadratic so I think a cubic model can be identified for seven time points. You will get a message if not.
 Kiki van Broekhoven posted on Monday, August 28, 2017 - 3:28 pm
Okay thank you. Would this also be true for a model with only 5 data points? I think that might really be too few data points to model cubic growth but I'm not sure.
 Kiki van Broekhoven posted on Monday, August 28, 2017 - 3:32 pm
(I believe I remember Bengt Muthen mentioning in another thread that he wouldn't bother using cubic growth for a model with only 4 data points so maybe this also applies to 5 data points)
 Bengt O. Muthen posted on Monday, August 28, 2017 - 4:38 pm
5 data points isn't really enough for a cubic.
 Mollie Marr posted on Thursday, November 09, 2017 - 2:33 pm
Could you provide a bit more information on why 5 data points would not be enough to include a cubic term?

We have 5 data points and get a good model fit when we include a cubic term (the shape appears to be cubic). Would this mean we have over fitting?
 Bengt O. Muthen posted on Friday, November 10, 2017 - 4:45 pm
It is certainly doable but perhaps not very convincing as a growth form - perhaps not much power to reject the model as you would have with more time points.