3-step manual calculation PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 SF Wang posted on Friday, April 18, 2014 - 12:36 pm
Dear professors,

I need to do the 3-step manual calculation (to take uncertainty of latent class assignment into account) comparing some variables among the latent classes (to see whether they differ in different latent classes).

Could you check whether I am doing it correctly? Thanks a lot!

C_all2 is most likely assignment from GMM.

...

CLASSES = c(3);
!3 classes identified in GMM

Analysis:
estimator = mlr;
TYPE = MIXTURE;
starts 50 20;

model:
%overall%

%c#1%
[faq] (m1);
[C_all2#1@3.290];
[C_all2#2@1.041];


%c#3%
[faq] (m3);
[C_all2#1@-3.363];
[C_all2#2@-9.301];

%c#2%
[faq] (m2);
[C_all2#1@7.708];
[C_all2#2@11.832];

!Logits for the Classification...
! 1 2 3
! 1 3.290 1.041 0.000
! 2 7.708 11.832 0.000
! 3 -3.363 -9.301 0.000

model test: m3=m1; m3=m2;
!test all means are the same
 SF Wang posted on Friday, April 18, 2014 - 12:45 pm
Another question: what model is assumed in comparing the means? Thanks.
 Linda K. Muthen posted on Saturday, April 19, 2014 - 10:47 am
As long as you have NOMINAL = c_all2;

MODEL TEST should be specified as:

MODEL TEST:
0 = m3 - m1;
0 = m3 - m2;
 SF Wang posted on Tuesday, April 22, 2014 - 6:57 am
Dear Linda, thank you for your answer! Is it equivalent to ANOVA (uncertainty incorporated)?

Another questions: For categorical variables, the manual calculation could be the following, correct?

nominal is C_all2;
missing are all (-999);
idvariable is rid;
CLASSES = c(3);

Analysis:
TYPE = MIXTURE;
starts 50 20;

Model:
%overall%

c#1 on gender (m1);
c#2 on gender (m2);

%c#1%
[C_all2#1@3.290];
[C_all2#2@1.041];

%c#2%
[C_all2#1@7.708];
[C_all2#2@11.832];

%c#3%
[C_all2#1@-3.363];
[C_all2#2@-9.301];

model test: m1=0; m2=0;


Thanks a lot!
 Bengt O. Muthen posted on Wednesday, April 23, 2014 - 10:53 am
This looks correct. But you don't need Starts. Note that you findm appendices with Mplus scripts for 3-step on our website.
 SF Wang posted on Tuesday, April 29, 2014 - 6:17 am
for categorical variables with multiple categories, would it be right to create dummy variables, then use the dummy variables in the same way as above code?

e.g. 3-category nominal variable:

c#1 on dummy1 dummy2 (m1);
c#2 on dummy1 dummy2 (m2);
...
model test: m1=0; m2=0

?

Thank you!
 Linda K. Muthen posted on Wednesday, April 30, 2014 - 6:11 am
If you have a nominal variable as a covariate, you can create a set of dummy variables and use them as covariates. For an ordered categorical covariate, you can treat it as a continuous variable or create a set of dummy variables.

By putting a label at the end of

c#1 on dummy1 dummy2 (m1);

you are holding the regression coefficients of dummy 1 and dummy 2 equal. The z-test in the output tests if this coefficient is zero. I'm not sure what you are trying to do in MODEL TEST.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: