proc genmod estimate statement question

[senorkoki]

hi,

i run the following code to transform a categorical variable.

DATA nb;
SET nb;
IF (numchron =0) THEN chronic = 0;
IF (numchron =1) THEN chronic = 1;
IF (numchron =2) or (numchron =3) or (numchron =4) or (numchron =5) or (numchron =6) or (numchron =7) or (numchron = 8) THEN chronic = 2;
RUN;


Then I run the following poisson regression:


proc genmod data= nb;
class chronic (ref = "0" param=ref );
class education (ref = "0" param=ref);

model ofp = education privins chronic /dist = poisson
link=log;

OUTPUT OUT=resid predicted=pred RESCHI=reschi RESDEV=resdev RESLIK=reslik
RESRAW=resraw STDRESCHI=stdresc STDRESDEV=stdresd;

estimate "log estimate education 1-11" education 0 /exp; /* reference */
estimate "log estimate education grad HS" education 1 /exp; /* this line gives exp of beta */
estimate "log estimate education BeyondHS" education 2 /exp; /* this line gives exp of beta */

estimate "log estimate" privins 1 /exp; /* this line gives exp of beta */

estimate "log estimate 0 illness" chronic 0 /exp; /* */
estimate "log estimate 1 illness" chronic 1 /exp; /* */
estimate "log estimate 2+ illnesses" chronic 2 /exp; /* */
run;





I'm concern with my estimate statement
The manner which I wrote it:

estimate "log estimate 0 illness" chronic 0 /exp; /* reference */
estimate "log estimate 1 illness" chronic 1 /exp; /* 1 chronic illness */
estimate "log estimate 2+ illnesses" chronic 2 /exp; /* 2+ chronic illnes */

Does chronic '1' in the estimate statement "estimate "log estimate 1 illness" chronic 1 /exp; /" actually obtain the log estimate of the chronic '1' category ? Likewise, does "estimate "log estimate 2+ illnesses" chronic 2 /exp;" obtain the log estimate for the chronic '2' category using '0' as reference?

Very confused. The results that are returned make sense, code works, just have a lingering doubt.

Thanks for all the help

[jrai]

Here is what I think:

1) Using exp option, you're not getting the log estimate of the illness but estimate of the illness itself. You should not use the exp option if you want log estimate of the illness.
2) Your statements estimate the mean of your DV assuming that intercept=0, education=0 and privins=0. If this is what you intend to have as an output then yes it uses 0 as reference.