class statements & odds ratios - how to differentiate

Reply
Occasional Contributor
Posts: 13

class statements & odds ratios - how to differentiate

I have the following statement:

proc logistic data=joe.extra2 order=formatted;

class sex (ref='female')/param=ref;

class tt (ref='<=2')/param=ref;

class clinic (ref='austin')/param=ref;

model a1ctri=tt sex race age clinic;

contrast 'female 2-3.4' tt 0 0 1 sex 0/estimate=exp;

contrast 'female 3.5-5.7' tt 0 1 0 sex 0/estimate=exp;

contrast 'female >5.7' tt 1 0 0 sex 0/estimate=exp;

contrast 'austin 2-3.4' tt 0 0 1 clinic 0 0 /estimate=exp;

contrast 'austin 3.5-5.7' tt 0 1 0 clinic 0 0 /estimate=exp;

contrast 'austin >5.7' tt 1 0 0 clinic 0 0 /estimate=exp;

run;

My contrast odds ratios are the same for each Female and each Austin contrast statement with the same tt level.

Which makes sense, since Female and Austin are both reference categories.

My question is, will changing the param=ref to param=effect, giving my referent categories -1 values, still measure the odds of of a tt class stated against the tt reference class?

For instance, if I changed it so that my code was now,

proc logistic data=joe.extra2 order=formatted;

class sex (ref='female')/param=effect;

class tt (ref='<=2')/param=effect;

class clinic (ref='austin')/param=effect;

model a1ctri=tt sex race age clinic depression htn dyslipidemia;

contrast 'female 2-3.4' tt 0 0 1 sex -1/estimate=exp;

contrast 'female 3.5-5.7' tt 0 1 0 sex -1/estimate=exp;

contrast 'female >5.7' tt 1 0 0 sex -1/estimate=exp;

contrast 'austin 2-3.4' tt 0 0 1 clinic -1 -1/estimate=exp;

contrast 'austin 3.5-5.7' tt 0 1 0 clinic -1 -1 /estimate=exp;

contrast 'austin >5.7' tt 1 0 0 clinic -1 -1/estimate=exp;

run;

would the odds ratio of the first class statement truly be, "Controlling for sex=female, the odds of a1c outcome for tt=2-3.4 vs. ref(<=2) is ____"?

I'm not sure if changing the param=ref to param=effect still tells sas that i am looking for odds for tt=2-3.4 vs. ref(<=2)

Thanks, kinda in a quick crunch.,help

Super User
Posts: 19,860

Re: class statements & odds ratios - how to differentiate

Posted in reply to joebaker7

I think the odds ratio shouldn't change, but the P-Values from the model will.

You could always test it Smiley Happy

Ask a Question
Discussion stats
  • 1 reply
  • 153 views
  • 0 likes
  • 2 in conversation