03-18-2017 08:55 PM
Does anyone know if there is a way to perform a logistical regression but not have the results use a reference variable to determine the results. For example I am modeling the likelihood of requiring an intervention in patients with a disease and I want to control for race gender and location:
proc surveylogistic data=kid.prs03;
class race female hosp_division / param=ref;
model intervention (event='1') = race female hosp_division ;
title 'Likelihood of Bronch by insurance ctrl for gender, location';
where age < 19 and disease=1;
The results are given as such:
But instead of getting a result for one race compared to another (rate of intervention in african-americans = 1.911 (1.545-2.362) compared to caucasians) I am hoping to get results that compare a race to all other races (instead of a reference race). That way, my results would be, 'compared to other races, African-Americans are more likely to undergo the intervention with an OR of ***'.
Anyone know how to do this?
07-03-2017 02:19 PM
Effects parameterization (or coding) compares each level to the average of all of the levels. Reference parameterization, which you used, compares each level to the reference level. To use effects coding, specify PARAM=EFFECT.