|
|
 |
| Mixture model with all missing in som... |
 
|
| Message/Author |
|
|
|
|
I have data with 3 variables, where, by definition, some cases have data on only 1 or 2 of the variables. Still, several classes are expected. Ideally, I would like to estimate something like the model below, but the syntax is clearly incorrect as Mplus tries to estimate all parameters in all classes. Any thoughts?
%c1#1%
[X1];
X1;
%c1#2%
[X1];
X1;
%c1#3%
[X1];
[X2];
X1;
X2;
X1 WITH X2;
%c1#4%
[X1];
[X2];
X1;
X2;
X1 WITH X2;
%c1#5%
[X1];
[X2];
[X3];
X1;
X2;
X3;
X1 WITH X2;
X1 WITH X3;
X2 WITH X3;
%c1#6%
[X1];
[X2];
[X3];
X1;
X2;
X3;
X1 WITH X2;
X1 WITH X3;
X2 WITH X3; |
|
|
|
|
In classes 1 and 2 where you mention only X1, do you know that these classes have no observations on X2, X3? If so, you can add "training data" to say that classes with no X2, X3 observations belong to either class 1 or class 2.
For classes where a variable is not observed, you can hold its parameters equal to classes where they are observed - this will not affect the estimates because you have no observations and it will eliminate the problem of having parameters in classes for which there are no corresponding data. |
|
|
|
|
@Bengt thank you; regarding your first point: Yes, I have dummies which I can use to identify observations as belonging to class 1/2, or 3/4, or 5/6. What syntax would I use to incorporate this "training data"?
Regarding your second point: Thank you, I think I can do this. |
|
|
|
|
| See UG ex7.23. |
|
|
| Back to top |
|
|
