topic proc traj convergence tcov problem in Statistical Procedures

proc traj convergence tcov problem

yvon — Tue, 23 Jun 2020 13:39:18 GMT

Hi everyone ,

I have a database (n=3000 obs). C30Score01 is a my outcome variable (summary score of quality of lifeC30 going from 0 to 100) PA01=continuous Physical Activity and imc01=continuous bmi.

I would like to modelised trajectories of my « summary score of quality of lifeC30 » with some independant variable (RISK) and also dependant one (TCOV) over 4 time point.

1- When my « VAR » is continuous, without tcov parameter i can easily found some trajectories by changing the order according the chosen group. But once i add « TCOV » to my model, i have « ERROR: False convergence » no matter the order used. Do you have any advice to solve this issues??

When i use a binary version of the summary score of quality of lifeC30 i can found some trajectories by using TCOV and RISK parameter together.

proc traj data=base out=out outstat=os outplot=op outest=oe ;
var C30Score01-C30Score04; indep D01-D04;
model cnorm; max 100; ngroups 3 ; order 2 2 2;
risk age tabac marital income surgery axilla ;
tcov PA01-PA04 imc01-imc04;
id id_group ;
run;

2- Do you have any idea ? when i put max 100.1 than max 100 (Qol going from 0 to 100) this convergence trouble disapear and i can found trajectories.

Thanks for you help !!

Re: proc traj convergence tcov problem

tbnassau1 — Thu, 19 May 2022 16:19:20 GMT

Did you ever figure this out? I'm having the same issue but a smaller sample size. Wondering if there is too much missingness in the tcov variable or if it's because of how tcov is distributed in the given trajectory groups??

Re: proc traj convergence tcov problem

greatdabon11 — Fri, 08 Jul 2022 13:17:37 GMT

@yvon

This must have been a long time, but I came across your post and wondered if you found a solution to your challenge.

from what I know, the RISK variables need to be categorical for it to work. Have you tried creating dummy variables for the categories of age or simply create a binary variable out of your RISK variables?

Bon