topic Re: error when used proc genmod （GEE modelling） in Statistical Procedures

error when used proc genmod （GEE modelling）

KKAY — Tue, 19 Jul 2022 18:23:51 GMT

I read a paper which introduced the the proc genmod to build the partial proportional odds model. I copy the code and run. It is OK .

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
data ppom;
input id y x1 x2@@;
cards;
1 2 0 1 17 2 1 0 33 1 1 0
2 1 0 1 18 0 0 0 34 0 1 0
3 2 0 0 19 0 0 1 35 2 0 1
4 1 1 1 20 0 0 1 36 1 0 0
5 2 0 1 21 3 1 1 37 3 1 1
6 2 1 1 22 3 1 1 38 2 0 0
7 0 0 0 23 1 1 0 39 2 0 0
8 2 0 1 24 1 1 1 40 2 0 1
9 0 0 1 25 0 0 1 41 2 0 1
10 1 0 1 26 3 0 0 42 2 0 1
11 1 1 1 27 2 0 1 43 0 0 1
12 2 0 1 28 2 0 0 44 2 1 1
13 2 0 1 29 2 0 1 45 2 1 1
14 2 0 0 30 1 1 1 46 1 0 1
15 2 0 1 31 3 1 0 47 1 1 1
16 3 1 0 32 2 0 1 48 3 0 0
;
data ppom2;
set ppom;
yy=(y=3); logit=3;output;
yy=(y ge 2);logit=2; output;
yy=(y ge 1);logit=1;output;
run;

proc genmod desc;
class id logit;
model yy=x1 x2 logit logit*x1/link=logit dist=bin type3;
repeated subject=id/type=un;
run;

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

But if I changed the few numbers, it showed there are errors.

ERROR: Error in computing the variance function.
ERROR: Error in parameter estimate covariance computation.
ERROR: Error in estimation routine.

Why? The errors did not always come. But it did. For example I changed the y=0 when id=33 as following

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

data ppom;
set ppom;
if id=33 then y=0;
run;
data ppom2;
set ppom;
yy=(y=3); logit=3;output;
yy=(y ge 2);logit=2; output;
yy=(y ge 1);logit=1;output;
run;

proc genmod desc;
class id logit;
model yy=x1 x2 logit logit*x1/link=logit dist=bin type3;
repeated subject=id/type=un;
run;

Re: error when used proc genmod （GEE modelling）

KKAY — Tue, 19 Jul 2022 22:37:37 GMT

Is anyone could help me?

Re: error when used proc genmod （GEE modelling）

Ksharp — Wed, 20 Jul 2022 12:30:36 GMT

Maybe @Rick_SAS or @StatDave could give you a hand ?

Re: error when used proc genmod （GEE modelling）

StatDave — Wed, 20 Jul 2022 13:40:52 GMT

That method is not necessary. It is now easy to fit a partial proportional odds model or a fully nonproportional odds model using the UNEQUALSLOPES and/or EQUALSLOPES options. The following fits the fully nonproportional odds model.

proc logistic data=ppom;
model y=x1 x2 / unequalslopes;
run;

See this note that provides much more detail and an example of fitting both the partial proportional odds model and fully nonproportional odds model.

Re: error when used proc genmod （GEE modelling）

KKAY — Wed, 20 Jul 2022 16:09:19 GMT

Thank you very much! I will learn about what you suggest and try.

Re: error when used proc genmod （GEE modelling）

KKAY — Wed, 20 Jul 2022 16:10:01 GMT

Thank you for your reply! I will try! Thank you!