- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Posted 08-16-2008 08:38 AM
(1221 views)
I am dealing with a problem to model longitudinal proportional data using both GEE (M1) and GLMM (M2). I am not sure if my syntax is correct especially for GEE since most SAS books only cover longitudinal "binary" data. I am even not sure if PROC GENMOD can handle GEE modeling for "repeated proportions". Both proportional and binary data belong to binomial distribution but the former one is in group level while the latter one individual level. Hope any one can share experience or opinion with me.
Thanks
Jacob
===================================
M1: GEE
proc genmod data=dsn;
class ID group;
model num/den = time group / link=logit dist=binomial type3;
repeated subject=ID/type=ar(1);
run;
M2: GLMM
proc glimmix data=dsn;
class ID group;
model num/den = time group / link=logit dist=binomial solution;
random int time/subject=ID type=un;
run;
ID time num den rate group
1 1 4 5 0.8 A
1 2 3 4 0.75 A
1 3 3 3 1 A
2 1 6 10 0.6 B
2 2 5 5 1 B
2 3 8 10 0.8 B
3 1 1 2 0.5 A
3 2 2 5 0.4 A
3 3 3 5 0.6 A
Note that rate = num / den
Thanks
Jacob
===================================
M1: GEE
proc genmod data=dsn;
class ID group;
model num/den = time group / link=logit dist=binomial type3;
repeated subject=ID/type=ar(1);
run;
M2: GLMM
proc glimmix data=dsn;
class ID group;
model num/den = time group / link=logit dist=binomial solution;
random int time/subject=ID type=un;
run;
ID time num den rate group
1 1 4 5 0.8 A
1 2 3 4 0.75 A
1 3 3 3 1 A
2 1 6 10 0.6 B
2 2 5 5 1 B
2 3 8 10 0.8 B
3 1 1 2 0.5 A
3 2 2 5 0.4 A
3 3 3 5 0.6 A
Note that rate = num / den
2 REPLIES 2
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I do not know where is the binary data, it is not 0/1 coded, it multinormal data I think
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I don't think this is multinomial (you wrote multinormal - I don't know what that is), but he has clearly specified the model correctly. For each class he has the number of trials and the number of observed events, and he is assuming that the number of events is a binomial distribution with probability p_{class}.