Anonymous posted on Tuesday, December 13, 2005 - 1:41 am
Dear Bengt and Linda,
Could you tell me what is the difference between the following syntax commands in Mplus (version 3)? I thought that these would be equivalent, but only the option 1. fits well to the data. Also, when I look at the mean values of the growth factors (with the TECH4 command in the first option), I obtain different mean values with these two options (too extreme with the 2. option, where B for the i = 1.00, and for the s = -.90, whereas with the TECH4 in the first option they both are zero?!).
I'm letting the program to estimate the T2 score loading, because this gives me significant variation in the linear slope (short-term longitudinal data, with highly stable timescores). Also, because the measurement interval between T2 and T3 is twice as long as the interval between T1 and T2, the linear trend loadings would be set as 1,2,3 - letting the middle timepoint to be estimated I have set the loading of T3 on 2, instead of three - does this make sense? Also, is it so that following the rules of (SEM) model idenfication, I have to constrain something (e.g., fix the variance of one time score to zero)when there is one unfixed timescore loading? The program does not run this kind of syntax otherwise, at least the one in described in "Option 1".
Also, how do I know when to report the standardized coefficient (i.e., Beta) for the variance estimates in the Mplus output, and when to report in fact the b (when the standardized coefficients indicate value .999, this is pretty clear though)?
See Growth Models under The | Symbol in Chapter 16. There is a table that shows the comparison of the BY and | languages.
If you have outcomes that are collected at unequal distances, this should be reflected in the time scores. For example if you collected data at weeks 1, 2, and 4, the times scores would be 0, 1, and 3.
We prefer working with unstandardized coefficients. If you get 999 for a coefficent, there is probably a negative variance somewhere,