Estimating classes at Respondent level PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Mahima Hada posted on Monday, July 05, 2010 - 10:21 am
Hi,

I ran an experiment with 5 manipulations (4-within, 1 between), and I have two continuous dependent variables. Each Respondent gave me four responses.

I want to estimate latent classes at the RESPONDENT level, not the response level. At the current analysis, I do get instances where one respondent's response 1 is in class 1 and response 3 is in class 2.

I used the cluster variable to indicate the respondents. And specified a between and within level model.

thanks,
Mahima

Code:
VARIABLE: NAMES ARE Rep Needs_sim Homoph R_inten SF_fam Lcons_1 Lcons_2 Lcons_conf MAN manid;
USEVARIABLES ARE Lcons_1 MAN Rep SF_fam;
CLASSES = c(2);
WITHIN = Rep;
BETWEEN = SF_fam;
CLUSTER = MAN;
IDVARIABLE IS manid;
ANALYSIS: TYPE=TWOLEVEL MIXTURE RANDOM;
MODEL: %WITHIN%
%OVERALL%
Lcons_1 on Rep;
%BETWEEN%
%OVERALL%
Lcons_1 on SF_fam;
OUTPUT: TECH1 TECH11 TECH7;
PLOT: TYPE=PLOT3;
SAVEDATA: FILE IS 2class_bet.dat;
SAVE=CPROB;
 Linda K. Muthen posted on Tuesday, July 06, 2010 - 8:11 am
What are the names of the four response variables in your data set?
 Mahima Hada posted on Thursday, July 08, 2010 - 5:56 am
Hi,

The response variable I am interested in is only 1 Lcons_1. (The other two are Lcons_2, Lcons_conf.)

The within subject manipulations are REP, Needs_sim Homoph R_inten. Between subject is SF_FAM.

And "MAN" identifies one manager - who has 3 or 4 responses
 Linda K. Muthen posted on Thursday, July 08, 2010 - 6:14 am
I would treat this as a multivariate rather than a multilevel analysis. I would arrange the data so that each each respondent is one record in the data set with four variables, one for each response. Multivariate analysis takes care of the clustering due to respondent.
 Mahima Hada posted on Thursday, July 08, 2010 - 1:24 pm
Hi Linda,

Thanks for your response. So to be clear, my equations would be something like:

Lcons_1A on RepA R_intenA Needs_simA HomophA SF_FAMA;
Lcons_1B on RepB R_intenB Needs_simB HomophB SF_FAMB;
Lcons_1C on RepC R_intenC Needs_simC HomophC SF_FAMC;

where each set - A,B,C - corresponds to one response from a manager.

Mahima
 Linda K. Muthen posted on Thursday, July 08, 2010 - 1:58 pm
Yes.
 Mahima Hada posted on Friday, July 16, 2010 - 8:36 am
Thanks. I tried your method - but as soon as I include two sets of equations together (Response 1 and 2 from Manager 1), the model is not able to get a solution.

I get the error:

" -1268.235 unperturbed 0

THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY. THE COMPUTATION OF
THE FISHER INFORMATION MATRIX COULD NOT BE COMPLETED.

THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE
COMPUTED. THIS IS OFTEN DUE TO THE STARTING VALUES BUT MAY ALSO BE
AN INDICATION OF MODEL NONIDENTIFICATION. CHANGE YOUR MODEL AND/OR
STARTING VALUES. THE SAMPLE COVARIANCE MATRIX COULD NOT BE INVERTED FOR CLASS 2. PROBLEM INVOLVING VARIABLE REP2."

If I use starting values, I still get the same error.REP2 does not give me any problems if I estimate only the model for Response 2.

My feeling is that there is correlation between the two responses - which is creating problems in the estimation. But that is part of the data - and what I want to capture in clusters.
 Mahima Hada posted on Friday, July 16, 2010 - 8:37 am
My Code:

ANALYSIS: TYPE=MIXTURE;
STARTS 1500 50;
!ALGORITHM=INTEGRATION;
MODEL: %OVERALL%
!%c#1%
![bias_av*1 Lcons_av*1 Lcons_av2*1];
!%c#2%
![bias_av2*1 Lcons_av*1 Lcons_av2*1];
Bias_av on Rep R_inten;
Bias_av2 on Rep2 R_inten2;
Lcons_av on Rep R_inten Needs_sim Homoph
bias_av;
Lcons_av2 on Rep2 R_inten2 Needs_sim2 Homoph2
bias_av2;M
 Linda K. Muthen posted on Friday, July 16, 2010 - 9:29 am
Please send your full output and license number to support@statmodel.com.
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: