Erkan Er posted on Saturday, October 10, 2015 - 9:17 pm
I have collected data with the same instrument three times (1 month after each other) from the same population. When I test model with each data set I obtain a good model fit, with some minor differences across data sets (for example, some relationships are not significant in one data set).
Then, I tried doing clustered/multilevel SEM analysisÁ Mplus runs the code (below) successfully. I obtain three result sets for three separate models (actually they are the same model), and all the relationships among the latent variables in each model are NS. I tried clustered/multilevel model because the parameters are all estimated using all of the data, so should have smaller standard errors and I should get more power to detect your effects.
Does my logic make sense? Do you have any idea why I receive bad results when I do clustered/multilevel SEM? Any help is appreciated. I can also send my data + code if needed.
NAMES = X1-X41; USEVARIABLES ARE X1-X41; GROUPING IS X3 (1 = 1 2 = 2 3 = 3); cluster is X1; ANALYSIS: TYPE IS complex; ESTIMATOR=ML;
model: !shortened REL BY X4 X5 X7 X10; TS BY X12-X15;
COSTS BY X19-X22; BENFS BY X23-X25;
model 1: !shortened BENFS ON TS REL (a1); COSTS ON TS REL (a2);
So when you say "three times from the same population", do you mean that it is the same people or different people at the different time points?
If X1 is the student id, what is clustered within student?
Erkan Er posted on Monday, October 12, 2015 - 6:06 pm
It is the same population. Actually, nothing is clustered within student. I just meant to group students according to time points: first data collection (1), second data collection (2), and third one (3).
If you have largely the same individuals at each of the 3 time points you should not treat time point as group in a multiple-group analysis. Multiple-group analysis needs to have independent observations from the different groups, that is, no same people.
Instead, those data should be analyzed as a longitudinal model, that is, with p variables measured at T time points you analyze a model with p*T variables.