Hi- I am interested in estimating a joint GMM and discrete time survival model in MPLUS (V6). My main interest is in examining the relationship between the hazard and the growth trajectories controlling for some invariant covariates. However, I am not sure if it is more appropriate to link the hazard to the growth parameters in the three hypothesized classes that I have, or to predict the classes using the hazard. In other words, do I code as:
usevariables are y0-y5 u0-u5 x1 x2; cluster is clust; strata is strat; categorical are u1-u5; classes = c(3); missing are all (-9999); Analysis: Type= Mixture complex missing; Algorithm= integration; model: %overall% i s | y0@0 y1@1 y2@2 y3@3 y4@4 y5@5; f on u1-u5@1; c#1 on f u0 x1 x2; c#2 on f u0 x1 x2;
or as: usevariables are y0-y5 u0-u5 x1 x2; cluster is clust; strata is strat; categorical are u1-u5; classes = c(3); missing are all (-9999); Analysis: Type= Mixture complex missing; Algorithm= integration; model: %overall% i s | y0@0 y1@1 y2@2 y3@3 y4@4 y5@5; f on u1-u5@1; i on u0 x1 x2; s on f u0 x1 x2; %c#1% i on u0 x1 x2; s on f u0 x1 x2; %c#2% i on u0 x1 x2; s on f u0 x1 x2; %c#3% i on u0 x1 x2; s on f u0 x1 x2;
or using a scheme other than what is specified above? your guidance on this is appreciated. Thanks
I think the major association between survival and GMM would be via the latent classes, so that hazards are different in different latent classes. The i, s variation is within-class variation and there is less of that given the latent classes already dividing things up, so relating i, s to hazards is less fruitful.
This then by default lets the i, s, f means change across classes.
See also the Muthen-Masyn (2005) article and the report:
Muthén, B., Asparouhov, T., Boye, M., Hackshaw, M. & Naegeli, A. (2009). Applications of continuous-time survival in latent variable models for the analysis of oncology randomized clinical trial data using Mplus. Technical Report.
Carl Lamote posted on Tuesday, January 17, 2012 - 2:26 am
Dear Dr. Muthen, I am estimating a DTSM (predicting school dropout, based on engagement trajectories). I read the paper on DTSMA and the different chapters on survival and mixture in the manual. However, I have some problems with my analyses. I am using a dataset of 6411 students in secondary school. For each student, we have a measure of emotional engagement (y) in Grade 7, 8, 10, 12 (no information on grade 9 and 11), and we know the year of dropout (also for grade 9 and 11). I came up with this model: usevariables are x1 x2 et1-et6 y1 y2 y4 y6; cluster = sch_long; categorical are et1-et6; missing are all(999999); classes = c(2); ANALYSIS: Type = Mixture Missing Complex; MODEL: %OVERALL% i BY y1y6@1; s BY y1@0 y2@1 y4@3 y6@5; f BY et1-et6@1; [i* s*]; [y1-y6@0]; i s ON x1 x2; f ON x1 x2; c#1 ON x1 x2; %c#1% [i* s*]; y1 y2 y4 y6 (1); i s ON x1 x2; %c#2% [i* s*]; y1 y2 y4 y6 (2); i s ON x1 x2; This model runs fine, but I want to know whether the hazards are different within each class. I tried including the "algorithm-integration" and "i s f ON x1 x2", as a read in the previous reply. However, including this, the model did not converge and gave some errors. Can you point out the mistake? Thanks in advance! (I'm using Mplus 4)
I would suggest getting Version 6.12 and then if you are having problems to send the output and your license number to firstname.lastname@example.org. There have been many changes between Verson 4 and Version 6.12.