Missing occasions in growth model PreviousNext
Mplus Discussion > Missing Data Modeling >
 Shige Song posted on Friday, June 16, 2006 - 6:14 am
I am trying to do a SEM-based growth model on data set that is highly unstructured (I like the SEM-based approach because it is very flexible with the functional form of the growth grajectory). I managed to convert the 5-wave (long) data into wide data with 19 (0-18) age groups. The model I want to estimate is this:

i s | i s | ht0@0 ht1@1 ht2*2 ht3*3 ht4*4 ht5*5 ht6*6 ht7*7 ht8*8 ht9*9 ht10*10 ht11*11 ht12*12 ht13*13 ht14*14 ht15*15 ht16*16 ht17*17 ht18*18;

Now the problem is that some (6) of the 18 age categories do not of a single case in it. Mplus refuses to continue with this data structure. Then I tried to impute the data but the imputation program (ICE with Stata) refused to proceed.

Here are My questions:
1. Is there a way to "trick" Mplus to continue?
2. Are there better ways to handle this situation in Mplus (besides adopting an individually-varying time point approach)?


 Bengt O. Muthen posted on Friday, June 16, 2006 - 2:09 pm
No information is lost if you delete the variables (age categories) that don't have a single case in them. This then essentially gives non-equidistant time scores, but that's fine.

Note also the new Mplus data handling command LONGTOWIDE.
 Shige Song posted on Friday, June 16, 2006 - 2:18 pm
Hi Bengt,

I can just do

i s | i s | ht0@0 ht1@1 ht2*2 ht3*3 ht10*10 ht11*11 ht12*12 ht13*13 ht14*14 ht15*15 ht16*16 ht17*17 ht18*18;

and disregard the fact that ht4 - ht9 do not exist? Will non-equidistant time scores create any problems? Where can I read more about this issues? Thanks!

 Bengt O. Muthen posted on Friday, June 16, 2006 - 2:49 pm
Right, no problem. Take the example of 4 time points with fixed time scores 0, 1, 2, 3. If you don't have anyone observed at wave three, you can simply work with 3 outcomes and the non-equidistant time scores 0, 1, 3. I don't know if this is written about.
 Shige Song posted on Saturday, June 24, 2006 - 1:45 pm
I combined two data sets and want conduct a set of growth models. Group A has 19 observations from age 0 to age 18, while Group B has a subset of the age groups (0-2, 9-18, for example).

If I pool them together and estimat one big model with a dummy variable assessing between data set difference, no further data manipulation is required and I can get some good results.

If I pool them together and conduct a multi-group analysis, however, I got error message saying that group B has no non-missing values for Y3-Y8, and refuses to continue.

I can get around this by excluding Y3-Y8 from the model statement, but this will be a big loss information for group A. Is there a better way to proceed?


 Bengt O. Muthen posted on Saturday, June 24, 2006 - 5:25 pm
I dont think a multi-group analysis can handle this, but perhaps type = mixture using Knownclass for the groups works.
 Shige Song posted on Sunday, June 25, 2006 - 12:55 am
Hi Bengt,

You are right. With type=mixture and knownclass, there is no need to exclude any time points. Thanks!

 Shige Song posted on Tuesday, June 27, 2006 - 1:48 pm
In a growth model, if I have missing values in covariates (x1 and x2) and I don't want to lose them, is it accetable to just put a statement like:

i s | ...;
i s by x1 x2;
x1 x2;
x1 with x2@0;

with type = missing? Are there other alternatives? Thanks!

 Shige Song posted on Wednesday, June 28, 2006 - 4:40 am
Found my answer here: http://www.statmodel.com/discussion/messages/22/984.html?1148321076

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