|
|
 |
| Best approach to LCGA with time-varyi... |
 
|
| Message/Author |
|
|
| Michael Lap posted on Tuesday, April 17, 2018 - 7:35 am
|
|
|
Dear Linda and Bengt:
Variables X and Y are measured at 4 time points. These variables are curvilinearly related to each other at each time point.
1) We would like to test an LCGA model identifying class trajectories in X (including a quadratic term) and use Y at Time 1 as a covariate.
is this approach appropriate or we have to include Y as a time-varying covariate, as it correlates with X at each time point?
2) is there a way to include a time-invariant covariate Z (dichotomous) that is expected to moderate the above-mentioned class trajectories?
Thank you. |
|
|
|
|
1) It is statistically ok, but how good it is depends on your substantive setting and your research question.
2) Use c on z. |
|
|
| Michael Lap posted on Wednesday, April 18, 2018 - 6:33 am
|
|
|
Thank you very much, Bengt.
I've just read in another thread about including a variable as both time-varying and time-invariant covariate.
1) Can I include:
a) Y (managers) at T1 as a predictor of employee trajectory membership
b) and Y at T1-T4 as time-varying covariate:
USEVARIABLES ARE empT1 empT2 empT3 empT4
mgrT1 MgrT2 MgrT3 MgrT4;
Missing are all (-999);
CLASSES = c(3);
analysis: type = MIXTURE random;
starts = 100 50;
model:
%overall%
I1 S1 Q1| empT1@0 empT2@1 empT3@2 empT4@3;
I1-Q1@0;
c#1-c#2 on mgrT1;
empT1 ON mgrT1;
empT2 ON mgrT2;
empT3 ON mgrT3;
empT4 ON mgrT4;
Will it mean that both intercept AND slope of employee class trajectories depend on the management at T1, controlling for correlations between management and employees at each time point?
2) if I include variable Z as a moderator, it will be another predictor:
c#1-c#2 on mgrT1 Z;
Will it mean that both intercept And slope of employee class trajectories depend on the management at T1 and variable Z, controlling for correlations between management and employees at each time point?
Thank you! |
|
|
|
|
| These general analysis questions are suitable for SEMNET. |
|
|
| Back to top |
|
|
