Predicting Future Observations PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Mark Sweatman posted on Wednesday, April 03, 2013 - 9:50 am
It's my understanding that you can utilize growth curve modeling to predict future observations. I would like to do that for a clinical trial that I am working on. I would like to predict what our 12 month follow-up should be based on five previous time points of data collection. However, I’m at a loss of how I should write the syntax in order to accomplish this. Here is my syntax thus far (which works fine for predicting the intercept, slope, and significance levels for the two groups I’m looking at – walkers vs. non-walkers):

Title: 10M Walk Group Linear Growth Curve Analysis

Data: File is C:\IGC 10M Analysis.csv;

Variable: Names are StudyID Group Name LEMS Walk Pre Twelve Eighteen Post SixMonth;
Usevariables are Walk Pre Twelve Eighteen Post SixMonth;
Grouping is Walk (0=Non-Walkers 1=Walkers);
Missing is all (999);

Model: i s | pre@0 twelve@1 Eighteen@2 Post@3 SixMonth@4;

Plot: Type is Plot3;
Series = Pre Twelve Eighteen Post SixMonth (*);

What is the syntax I should plug in so I can predict a 12Month follow up at time point 5?

Thanks so much for your help with this!
 Linda K. Muthen posted on Wednesday, April 03, 2013 - 5:04 pm
You can use MODEL CONSTRAINT to do this. Use the formulas for y shown on Slide 45 of the Topic 3 course handout on the website using the time score for the 12 month follow up. I think you time scores are incorrect in the statements above. The Topic 3 course handout also shows how the determine time scores. If twelve and eighteen represent weeks, the time scores should be 0 1 1.5 not 0 1 2.
 Mark Sweatman posted on Thursday, April 04, 2013 - 10:57 am
Thank you! I now understand the concept behind calculating a new time point variable better, but I’m a little stumped on how to translate that into syntax. My updated syntax is:

Data: File is H:\IGC 10M Analysis.csv;

Variable: Names are StudyID Group Name LEMS Walk Pre Twelve Eighteen Post SixMonth;
Usevariables are Walk Pre Twelve Eighteen Post SixMonth;
Grouping is Walk (0=Non-Walkers 1=Walkers);
Missing is all (999);

Model: i s | pre@0 twelve@1 Eighteen@1.5 Post@2 SixMonth@4;

Model Constraint: new (twelvemo);
twelvemo = img + smg*8;

Plot: Type is Plot3;
Series = Pre Twelve Eighteen Post SixMonth (*);

Under Model Constraint, I was trying to specify IMG as the intercept growth factor, SMG as the slope growth factor, and 8 as the time score for the observation I want to predict at point 8 in the time series. I fixed the time scores on your advice, and corrected them to be 0 1 1.5 2 4 & 8. Can you please advise me to the correct syntax to get a predicted value for twelve month follow up?

Thank you so much for your assistance!
 Linda K. Muthen posted on Friday, April 05, 2013 - 9:30 am
Model: i s | pre@0 twelve@1 Eighteen@1.5 Post@2 SixMonth@4;
[i] (p1);
[s] (p2);

Model Constraint: new (twelvemo);
twelvemo = p1 + p2*8;

To be sure the formulas are correct, run it also for time score 4 and check the RESIDUAL output to be sure you get the same number.
 Mark Sweatman posted on Tuesday, April 09, 2013 - 6:14 am
Thank you so much Dr. Muthen. I really appreciate it. I will run this and then double check it with time score 4. :-)
 Mark Sweatman posted on Tuesday, April 09, 2013 - 1:26 pm
Dr. Muthen,
Thank you for helping me out with my growth curve model and for correcting my syntax. I ran the model and everything seemed to converge just fine; although, now I need a little help in interrupting the results. My model is grouped by two populations – walkers and non-walkers. But when I look at the results for my new calculated variable, I only see values (i.e. estimate, p-value, etc.) for one calculation of my new time series variable towards the bottom of the output under “New/Additional Parameters.” I don’t see separate calculations for the walkers group the non-walkers group. Additionally, I don’t really understand the estimate that is produced. That’s not supposed to be the mean for the new time series that was calculated, is it? Also, can a new time series be graphed?
 Linda K. Muthen posted on Tuesday, April 09, 2013 - 1:52 pm
Please send the output and your 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: