Randomized Clinical Trial
Message/Author
 Michelle Finney posted on Sunday, September 30, 2012 - 11:49 pm
Dear Drs Muthen,
I am analyzing data from a large clinical trial with three arms: one Placebo (pl) and two treatments (t1; t2). I am applying the approach explained in Muthen & Curran (1997).
In addition to "treatment effects" (pl vs tr1 and pl vs tr2) I am interested in examining the effect of a continuous covariate (X) on the <continuous> outcome within each grp (pl, tr1, & tr2).
The model without the covariate does not converge. The ssize in each grp varies from 200 to 290. (below is my script.)
My questions are:
1) Does the approach explained in Muthen & Curran can be extended to two treatment arms vs placebo? If so, would you please check my syntax?
2) How would you add the effect of the covariate into this modeling approach?
 Michelle Finney posted on Monday, October 01, 2012 - 12:04 am
here is the code:
i s | v0@0 v18@1.5 v36@3 v48@4;
i tr1 | v0@0 v18@1.5 v36@3 v48@4;
i tr2 | v0@0 v18@1.5 v36@3 v48@4;
[v0-v48] (1);
[i@0];
i (2);
s (3);
i WITH s (4);
[s] (5);
tr1@0;
tr2@0;
q WITH i@0
s@0;
tr1@0;
tr2@;
v1-y36 WITH v2-v48;
tr1 ON i;
tr2 on i;
tr2 with tr1@0;
MODEL control:
tr1 ON i@0;
tr2 on i@0;
[tr1@0];
[tr2@0];
 Bengt O. Muthen posted on Monday, October 01, 2012 - 8:31 am
1) Yes. You have to use 3 groups instead of 2.

2) The covariate effect could be constant over the groups, so specified as i s on x in the overall part of the model.

 Michelle Finney posted on Monday, October 01, 2012 - 6:40 pm
Dear Dr Muthen,
The q is a typo, I don't have quadratic growth.
I ran the following model with and without the covariate "PL727" and I get a "nonconvergent - possibly non-identified model." I tried fixing a negative residual variance to a low number and it didn't work. Is there something obviously wrong with my code?
i s | f120@0 f1218@1.5 f1236@3 f1248@4;
i e2 | f120@0 f1218@1.5 f1236@3 f1248@4;
i cee | f120@0 f1218@1.5 f1236@3 f1248@4;
i s on PL727;
[f120-f1248] (1);
[i@0];
i(2);
s(3);
i WITH s (4);
[s] (5);
e2@0;
cee@0;
f120-f1236 WITH f1218-f1248;
e2 on i;
cee on i;
cee with e2@0;
MODEL Placebo:
e2 ON i@0;
cee on i@0;
e2 with cee@0;
[e2@0];
[cee@0];
 Linda K. Muthen posted on Tuesday, October 02, 2012 - 10:32 am
You should have three groups not two if you have two treatements and one placebo. Please send the output and your license number to support@statmodel.com.