Message/Author 

Isaac posted on Thursday, April 14, 2011  7:22 am



Hi, I'm building an LGMM with 4 points. The initial time point has some variability in when the data was collected so I'm trying to decide how to deal with this. The variablility is normally distributed so initially I thought of putting the loading for this time point at the mean. I then thought of using TSCORES but I see that can only be done with TYPE=RANDOM which doesn't allow for classes. Am I right about this. I don't actually think the variability is very meaningful but I'm concerned about doing this correctly. Any advice? 


Type=Random Mixture is available with TSCORES and AT. But it can lead to slow computations. Instead, I would investigate the sensitivity to timing by using a multiple cohort, multiplegroup approach where you divide the initial time point variability into a few meaningful groups. Then build your model from UG ex6.18. 

Isaac posted on Thursday, April 14, 2011  9:16 am



Thanks Bengt, That's EXTREMELY helpful 

Isaac posted on Thursday, April 14, 2011  11:53 am



Hi Bengt, I;m getting an error saying that Type=Mixture is not allowed for multiplegroup. I'm following ex 6.18. Am I doing something wrong here? 


In mixture multiplegroup is handled via KNOWNCLASS (see index). 

Xu, Man posted on Thursday, September 14, 2017  12:46 am



Dear Dr. Muthen, I have a question about TSCORE as well. I have a model with 2 growth curves. The model did not run. Would you pls take a look: WEIGHT IS cweights; USEVARIABLES ARE !zf02m1 dmrraw emrraw fmrraw !variable for first growth curve ccbmi dcbmi ecbmi fcbmi gcbmi !variables for second curve cbmiage dbmiage ebmiage fbmiage gbmiage;!measurement time in months (age) TSCORES = cbmiage dbmiage ebmiage fbmiage gbmiage; ! this was the age in months at the data collection ANALYSIS: estimator is mlr; TYPE = RANDOM; MODEL: i_matrix s_matrix  dmrraw@0 emrraw@2 fmrraw@4 AT dbmiage ebmiage fbmiage; i_bmi s_bmi  ccbmi@0 dcbmi@2 ecbmi@4 fcbmi@6 gcbmi@8 AT cbmiage dbmiage ebmiage fbmiage gbmiage; i_matrix s_matrix ON i_bmi; s_bmi with s_matrix; OUTPUT: STDY STDYX SAMPSTAT; *** ERROR in MODEL command The following statement involving  is not correct. The syntax requires a variable on the lefthand side of the AT keyword. Problem with: I_BMI S_BMI  AT CBMIAGE DBMIAGE EBMIAGE FBMIAGE GBMIAGE 

Jon Heron posted on Thursday, September 14, 2017  7:43 am



Hi Kate when using tscores and AT the whole command has to be on one line. I think that's the reason for your error. 

Xu, Man posted on Thursday, September 14, 2017  8:16 am



Hi Jon, Thanks! I shortened the variable names but it still seem to have a problem: USEVARIABLES ARE !zf02m1 dmrraw emrraw fmrraw ccbmi dcbmi ecbmi fcbmi gcbmi c d e f g; TSCORES = c d e f g; ANALYSIS: estimator is mlr; TYPE = RANDOM; MODEL: i_matrix s_matrix  dmrraw@0 emrraw@2 fmrraw@4 AT d e f; i_bmi s_bmi  ccbmi@0 dcbmi@2 ecbmi@4 fcbmi@6 gcbmi@8 AT c d e f g; i_matrix s_matrix ON i_bmi; s_bmi with s_matrix; OUTPUT: STDY STDYX SAMPSTAT; *** WARNING in MODEL command Duplicate AT relationship in definition of random effect: Y9 X AT D 

Jon Heron posted on Thursday, September 14, 2017  8:17 am



Ahh, can you use two AT commands? 

Jon Heron posted on Thursday, September 14, 2017  8:23 am



yes you can as it's done here http://www.statmodel.com/discussion/messages/14/478.html?1496824847 Not sure what is wrong there, perhaps it's related to the imbalance of the two processes  3 and 5 time points. Depending on how varying time is you could just widen your dataset and not use tscores. 

Jon Heron posted on Thursday, September 14, 2017  8:47 am



I think I've sussed it  you need a different set of time variables for each line. If you read the link above (the bit about multiple zeros) that may help 

Back to top 