Age as time varying covariate?
Message/Author
 Dan posted on Wednesday, June 17, 2015 - 12:31 pm
Hi,

Is it possible & appropriate to include age as a time varying covariate in LGCA?

I have a dataset with 4 ~"equally" spaced waves, although there is about 1 year range in ages within a given wave (e.g., in Wave 1, age ranges from 18-19; in Wave 2, age ranges from 19-20).

1) Modeling a linear LGCA: When I include AGE as a time varying covariate in the linear model, the fit is very good as suggested by CFI, TLI, SRMR, RMSEA, and lower BIC. [As a sidenote, AGE is not a significant covariate at any wave in this model]

However, when I remove AGE from the time varying covariate list, the model fit becomes abysmal as indicated by the various indicators.

2) So, I tried LGCA with individually-varying times of observation, using AGE as a the TSCORES. However, I get a giant residual variance for ICEPT (200+, when outcome Y is only scaled 0-16).

I'm trying to understand if keeping age as a TVC is OK, and further, trying to understand what's happening when I attempt individually-varying times of observation.

Thanks for any direction you can provide!
 Bengt O. Muthen posted on Wednesday, June 17, 2015 - 6:13 pm
I would use TSCORES, but make sure you have age transformed to have a zero value at the start, and perhaps divide it by 10 to get an easier scale for the iterations.
 Dan posted on Tuesday, June 23, 2015 - 1:13 pm
Thank you. I have a follow-up question about interpretation of time-varying covariates.

A current growth curve model I'm working with is similar to Example 6.10:

--------------------

i s | y0@0 y1@1 y2@2;
i s ON x1 x2 x3;

y0 ON a0 b0 c0;
y1 ON a1 b1 c1;
y2 ON a2 b2 c2;

--------------------

What would be the proper interpretation of a significant time-varying covariate (e.g., a1) in the y1 regression?
Is it just a standard regression where a1 influences the y1? Or is "a1" exerting an effect on the slope?
[everything is continuous]

Thanks, and sorry if I missed this elsewhere on the board.
 Bengt O. Muthen posted on Wednesday, June 24, 2015 - 6:53 pm
It's a standard regression with y1 as DV.

Indirectly this changes the slope because the mean of y1 is affected.
 Dan posted on Thursday, June 25, 2015 - 6:45 am
Thank you!