Message/Author |
|
|
I am trying to estimate a parallel process latent growth model with five waves of data from adolescence to early adulthood. The outcomes are depression (continuous) and number of conduct disorder symptoms (count, maximum 4). The sample size is 662. My input is: usevar are DEP1 DEP2 DEP3 DEP4 DEP5 cdc41 cdc42 cdc43 cdc44 cdc45 ; count are cdc41 cdc42 cdc43 cdc44 cdc45 ; Analysis: estimator = mlr ; Algorithm = integration ; Integration =5 ; starts 40 2 ; stiterations = 20 ; MODEL: icond scond qcond | cdc41@0 cdc42@1 cdc43@2 cdc44@3 cdc45@4 ; idep sdep qdep | dep1@0 dep2@1 dep3@2 dep4@3 dep5@4 ; However, I receive the following error messages: THE ESTIMATED COVARIANCE MATRIX COULD NOT BE INVERTED. COMPUTATION COULD NOT BE COMPLETED IN ITERATION 227. CHANGE YOUR MODEL AND/OR STARTING VALUES. THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY. ESTIMATES CANNOT BE TRUSTED. Is there a special way to model continuous and non-continuous growth curves together? Any advice would be much appreciated. |
|
|
As a first step, I would try each growth model separately. I would not use INTEGRATION = 5. I would instead use INTEGRATION = MONTECARLO (5000). |
|
|
Thanks - I have already tried each growth model separately, and they run fine and make sense. I tried your montecarlo suggestion, but still fail to get estimates, and the following error message: THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-ZERO DERIVATIVE OF THE OBSERVED-DATA LOGLIKELIHOOD. THE MCONVERGENCE CRITERION OF THE EM ALGORITHM IS NOT FULFILLED. CHECK YOUR STARTING VALUES OR INCREASE THE NUMBER OF MITERATIONS. ESTIMATES CANNOT BE TRUSTED. THE LOGLIKELIHOOD DERIVATIVE FOR PARAMETER 17 IS -0.66093711D+01. Is it possible that the data are simply not suitable for this type of model? |
|
|
Please send the output and your license number to support@statmodel.com. |
|
|
Hello, In a similar model with dual trajectories, one normally distributed and one with a Poisson distribution (COUNT), how would one interpret a regression path from the latent intercept of the count trajectory to the intercept or slope of the continuous process? In other words, how is the regression estimate from a continuous variable regressed on a latent Poisson variable interpreted? Oppositely, how would the estimate from a Poisson latent variable regressed on a continuous variable be interpreted? In this case I assume that the exponentiated regression estimate could be interpreted as "a one unit increase in ...". Thanks, |
|
|
The latent variables for both growth processes are continuous. So all relationships are linear regression relationships. |
|
Back to top |