Multiple Cohort Growth Model (by grad... PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Elissa Hamlat posted on Sunday, March 04, 2018 - 11:14 am
I would like to estimate a Multiple Cohort Growth Model similar to the example in ch 6.18 of the user guide. Instead of cohorts based on birth year, my three cohorts are grouped according to grade at baseline 3rd, 6th, and 9th grade.

So ages within each cohort do range at each assessment (each cohort was assessed at three timepoints 18 months apart). Can I use the syntax from 6.18 as-is or are there any adjustments I need to make to how my data is structured or to the syntax?
 Elissa Hamlat posted on Sunday, March 04, 2018 - 11:15 am
Below is a draft of my syntax, thanks!
Note: APD is the outcome whose trajectory I want to model.

GROUPING = g (1 = 3RD 2 = 6TH 3 = 9TH );
MODEL:
i s |APD1@0 APD2@.15 APD3@.3;
[i] (1); [s] (2);
i (3); s (4);
i WITH s (5);
APD1 ON age1;
APD2 ON age2;
APD3 ON age3 (g6);
APD1-APD2;
APD3 (v6);
MODEL SIXTH:
i s |y1@.3 y2@.45 y3@.6;
APD1 ON age1 (g6);
APD2 ON age2;
APD3 ON age3 (g9);
APD1 (v6);
APD2;
APD3 (v9);
MODEL NINTH:
i s |y1@.6 y2@.75 y3@.9;
APD1 ON age1 (g9);
APD2 ON age2;
APD3 ON age3;
APD1 (v9);
APD2-APD3;
 Elissa Hamlat posted on Sunday, March 04, 2018 - 11:28 am
I also want to clarify that my primary goal is to model the trajectory of APD over 7 timepoints from grade 3rd to 12th (ages 8 to 18).
 Bengt O. Muthen posted on Sunday, March 04, 2018 - 2:17 pm
You don't need to include age as a predictor unless such small differences within grade are substantively important. And isn't it the case that

age2 = age1 + 1.5
etc for age3?

If so, these age variables are correlated 1 which is a problem. You need only the first one.


Using 6.18 is fine - your input looks ok.

Note that we request that postings be limited to one window only. For longer mesages, send to Support.
 Elissa Hamlat posted on Sunday, March 04, 2018 - 4:40 pm
Thank you, this is helpful.

You are correct, age2 = age1 + 1.5 and so on.

Would you then suggest dropping the time varying covariates (ages) from the model? Or just using one age variable (e.g., Age1) since they are all correlated as you mention?

If the latter, would I just include the one regression (APD1 ON age1;) in my models, and drop the other two regressions?

I am surprised that this was not an issue in Ex 6.18 as I thought the ages there would be correlated as well.
 Bengt O. Muthen posted on Monday, March 05, 2018 - 11:11 am
You can use Age1. Ex 6.18 has time-varying covariates labeled "a" but they are not intended to reflect age.
 Elissa Hamlat posted on Monday, March 05, 2018 - 11:36 am
Ok, thank you. What do the time varying covariates "a" in Ex 6.18 represent?

Would it just be another outcome variable (e.g., if my main outcome is depression, a tvc of interest may be anxiety) ?
 Bengt O. Muthen posted on Monday, March 05, 2018 - 1:23 pm
In an education application you might have

y = math achievement in a certain grade

a = math course taking in a certain grade
 Elissa Hamlat posted on Monday, March 19, 2018 - 11:21 am
Thank you, I think I understand now.

using the same basic model design above, I have moved on to using multiple indicators for the outcome at each timepoint.

So APD1, APD2, and APD3 are now latent factors made up of two manifest indicators each (e.g., APD1 BY apd_c apd_p).

The model runs ok except that the mean for the intercept growth factor of my first group (Grade 3) is zero. This does not fit well with my data. I would like to free it to vary but nothing I have tried works - using an *, setting a starting value, labeling it. Can you help?
 Bengt O. Muthen posted on Monday, March 19, 2018 - 3:34 pm
Send your output to Support along with your license number.
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: