Anonymous posted on Thursday, November 27, 2003 - 6:18 pm
Hi, I have two questions: FRIST QS: I am doing a CFA using 3 latent variables, X, Y, and Z. I have 3 categorical predictors (smoking status(a), gender(b), and grade(c)). What I would like to do is include all possible interaction terms involving these predictors. Could you please let me know what the syntax is?
SECOND QS: If I would like to do the same CFA model (XYZ) at two time points and see whether they vary / invary, how do i do it? What are the things I set equal if I want to say the two models at two points are exactly the same including latent indicators. When i do the following, the program automatically puts WITH statement connecting the two models (y2 with y1; z2 with z1) . Shall I fix them to zeros? Could you please let me know of some reference in testing repeated measure models?
model z1 = x1 y1; y1 = x1;
z2 = x2 y2; y2 = x2;
Thanks in advance for your help
bmuthen posted on Friday, November 28, 2003 - 4:34 pm
Regarding your first question, you simply do what you do in regular regression, namely create products of your predictors and then include those additional predictors in the model. So no special Mplus syntax is called for.
Regarding your second question, you probably DO want the WITH parameters included in your model to capture correlations across time. For example, if you have 3 indicators of one factor at two time points and you want to say that the model is the same at the two time points, you would specify
f1 by y11 !f1 is the factor at time 1 y12 (1) y13 (2); f2 by y21 !f2 is the factor at time 2 y22 (1) y23 (2); [y11 y21] (3); [y12 y22] (4); [y13 y23] (5); f1 f2 (6); [f1@0f2@0];
perhaps also adding time-invariant residual variances. Here, Mplus would automatically correlate f1 and f2. This makes sense because a person's factor values, and therefore outcomes, are most likely correlated across time.