Tao Xin posted on Monday, January 06, 2003 - 12:12 pm
Dear Brendt, I met a research situation in which the data structure is as same as Coleman's data you used for mix14. However I have some questions about the example you posted. In the syntax of mix14, you defined the latent structure as following: [membr1$1*-1] (1); [attd1$1*-2] (2); [membr2$1*-3] (3); [attd2$1*-2] (4);
... I am wondering what kind of criteria you used to decide thresholds within in each class. And also, I don't understand the meaning of the numbers in parentheses (eg., (6)). I checked the Mplus manual, but I didn't find the explanation.
For Coleman's data, it's logical to assume that there are two types of correlation. The first is the correlation between indicators within each measure occasion, and the second is correlation between different measurement occasions (or raters). We may think this model as a probit-linear mixed model with correlated errors. Would you please tell me if Mplus can deal with the probit-linear model with correlated errors? Thank you very much.
Have a good holiday.
bmuthen posted on Monday, January 13, 2003 - 10:25 am
This is an example of having 2 binary latent class variables. One refers to membership which is measured at 2 time points (2 indicators). The other refers to attitude which is measured at 2 time points (2 indicators). So each latent class variable describes the correlation of the same measure across time. The 2 latent class variables are also correlated.
This is handled in Mplus using equality constraints on the item parameters. You will find in the Mplus user's guide that equalities are specified using parentheses. For example, equality (1) says that the membership item parameter does not change across the first 2 classes. This implies that Mplus considers 4 classes ordered as: m-class 1, a-class 1; m-class 1, a-class 2; m-class 2, a-class 1; and m-class 2 a-class 2. Here m stands for membership and a stands for attitude.