Two-level LPA PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Jan Olav Christensen posted on Wednesday, February 03, 2016 - 11:49 am
Dear drs. Muthen,

I am running a twolevel latent profile analysis in which the BIC seems to suggest relatively high numbers of classes on both levels. I am wondering if the existence of zero class counts for some combinations of within and between classes (e.g. zero persons assigned to the between 2/within 8 class) is unacceptable and implies misspecification - or does it simply mean that some level 2 classes have fewer level 1 classes occurring in them than others. All between- and within-level classes have persons assigned to them.
 Bengt O. Muthen posted on Thursday, February 04, 2016 - 6:58 pm
Having those empty cells is not a problem and should not hurt BIC.
 Jan Olav Christensen posted on Tuesday, February 16, 2016 - 1:10 am
Hi again,

I want to regress distal outcomes on latent profiles at both levels. Is the procedure for doing this a straightforward extension of regressing a distal outcome on a latent categorical in a single-level model? More specifically; can I simply compare thresholds across within-level profiles and compare means across between-level profiles, as in the following? (distal_between is the aggregated counterpart of distal_within)

...
VARIABLE:
CATEGORICAL = distal_within ;
WITHIN = ....... distal_within ;
BETWEEN = ........cb distal_between ;
...
MODEL:
%WITHIN%
%OVERALL%
cw distal_within ON covariates ;
%BETWEEN%
%OVERALL%
cw ON cb ;

MODEL cw:
%WITHIN%
%cw#1%
[distal_within$1] (t1);
%cw#2%
[distal_within$1] (t2);

MODEL cb:
%BETWEEN%
%cw#1%
[distal_between] (t3);
%cw#2%
[distal_between] (t4);

MODEL CONSTRAINT:
new(logOR_2_1) ; logOR_2_1 = t1-t2 ;
new(OR2vs1) ; OR2vs1 = exp(t1-t2) ;

MODEL TEST:
t3=t4 ;
 Bengt O. Muthen posted on Tuesday, February 16, 2016 - 6:44 pm
This looks correct.
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: