Multilevel mixture modeling PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 xue li posted on Friday, June 16, 2006 - 12:32 pm
Hi, every one:

There are 3000 clusters, within each cluster there are 10 subjects . Each subject has a depent variable measured 0 or 1, and each subject has no covariate.

I want to use two level mixture model to analysis this data. The reason is the observed data are grouped into 2 classes. Class one has mean -3 and variance approximately 0; and the second class has mean 0 and variance approcimately 4.Approximately 37.7% subjects are in class 1. The problem is the model never converge. Could someone tells me whether there is mistakes in my program. Thanks.
my email xli28@uic.edu

I use the following code.
Title: Mixture model for breast cancer project-race by age specific poverty
two-level logistic regression


Data:
!file is C:\xli28\project1\EB\mplus for EB\EB6ageg.dat;
file is C:\xli28\project1\EB\sim_age3.dat;
!file is F:\xue new\sim_age3.dat;
variable: names are ncluster povind int;
usevariables are povind;
missing are . ;
classes = c (2);
categorical=povind;
cluster=ncluster;
analysis: type =twolevel random mixture;
algorithm=integration;
estimator=ML;
integration=GAUSSHERMITE(15);
adaptive=on;

model:


%within%
%overall%
i|povind@0;
[i];
i;


%c#1%
[i*-3];
i@0;


%c#2%
[i*0];
i*4;


%between%
%overall%
[povind$1@0];
output: tech1; tech8;
 xue li posted on Friday, June 16, 2006 - 12:37 pm
Hi, the error message is as follows.
Since the data set is created by simulation. The starting value will be plausible. I can't figure out the reason.


Error message:

Unperturbed starting value run did not converge.


THE ESTIMATED WITHIN COVARIANCE MATRIX IN CLASS 1 COULD NOT
BE INVERTED. COMPUTATION COULD NOT BE COMPLETED IN ITERATION 1.
CHANGE YOUR MODEL AND/OR STARTING VALUES.

THE ESTIMATED WITHIN COVARIANCE MATRIX IN CLASS 0 COULD NOT
BE INVERTED. COMPUTATION COULD NOT BE COMPLETED IN ITERATION 1.
CHANGE YOUR MODEL AND/OR STARTING VALUES.

SERIOUS PROBLEM IN THE OPTIMIZATION WHEN COMPUTING THE POSTERIOR
DISTRIBUTION. CHANGE YOUR MODEL AND/OR STARTING VALUES.
THE LOGLIKELIHOOD DECREASED IN THE LAST EM ITERATION. CHANGE YOUR MODEL,
STARTING VALUES AND/OR THE NUMBER OF INTEGRATION POINTS.



THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ERROR IN THE
COMPUTATION. CHANGE YOUR MODEL AND/OR STARTING VALUES.






MODEL RESULTS

Estimates

Within Level

Latent Class 1

I |
POVIND 1.000

Means
I -3.000


Variances
I 0.000

Latent Class 2

I |
POVIND 1.000

Means
I 0.000


Variances
I 4.000

Between Level

Latent Class 1

Thresholds
POVIND$1 0.000

Latent Class 2

Thresholds
POVIND$1 0.000

Categorical Latent Variables

Within Level

Means
C#1 0.000
 Bert Weijters posted on Friday, October 17, 2008 - 8:35 am
I'm running a mixture twolevel model in mplus. However, I would like to define the latent class variable as a between level variable. In other words, I would like to classify 'clusters' (in the mplus multilevel language). How can I do this?

I tried including the statement
BETWEEN = c;
but it looks as if this statement is being ignored.

The reason why I want to classify at the between level is that my respondents are situated at this level; the within level has 18 ratings of different experimental scenario's by each respondent. I'm not interested in segmenting at this level (because I would probably end up with 18 classes corresponding to the 18 ratings), but at the respondent level.
In the previous discussion I found the following statement - has this been implemented?
"In future Mplus versions, we will also have a latent class variable varying on the between level. That would be a school-level variable, classifying different schools."
 Linda K. Muthen posted on Friday, October 17, 2008 - 8:43 am
See Example 10.2. If this is being ignored, perhaps you are using a version of Mplus where this was not yet implemented.
 Bert Weijters posted on Friday, October 17, 2008 - 11:50 am
I was wrong. "between=c" is not being ignored, but I misinterpreted the output: it lists the class counts in terms of level 1 (the within level), but I now see that observations from the same cluster are indeed assigned to the same class. Thanks.
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: