topic Re: PROC PHREG Contrast statement for hypothesis testing in Statistical Procedures

PROC PHREG Contrast statement for hypothesis testing

SASnewbiee — Wed, 08 Jul 2020 20:45:49 GMT

Hello,

I am trying to understand how we can test a level that is also a reference group??

I'm focusing on race here for hypothesis testing. Race is categorized as:

1 White

2 Other

3 Black

4 Hispanic

Code I used:

Proc PHREG data=x;

class race (ref=first);

model time*status(0)= age race gender;

contrast 'white+other Vs. Black+hispanic' race (1 1 -1 -1);

run;

Will this contrast statement test the hypothesis of category what/other vs black/hispanic? or do I need to remove the class statement to do so, since the class statement would make the ref category to 0..

So confused.

Re: PROC PHREG Contrast statement for hypothesis testing

ChrisNZ — Fri, 10 Jul 2020 02:04:10 GMT

Moved to STAT forum in the hope you'll get an answer there.

Re: PROC PHREG Contrast statement for hypothesis testing

SteveDenham — Fri, 10 Jul 2020 12:41:06 GMT

Rather than writing a contrast statement (which requires knowing coefficients for all the effects in a model, even though most would be zero in your case) try using an lsmestiimate statement which compares the average of white and other to the average of black and hispanic;

lsmestimate race 'white+other Vs. Black+hispanic' 1 1 -1 -1/divisor=2;

Reference levels don't enter into things this way. Instead least squares means for each level are constructed and then compared using the coefficients. You can check this by also adding

lsmeans race;

and hand computing the difference. It should match the lsmestimate value.

SteveDenham