I'm doing multiple logistic regression in SAS. One of the independent variables has 10 different events (0 to 9) and 0 is directly related to the dependent variable being positive. Now, if I code so that ref = '0', none of the other events under this independent variable are statistically significant; whereas if I code so that my ref is something other than 0, almost all the events are significant. Needless to say, the model is changing significantly in both these cases, but the sensitivity, specificity, false POS, and false NEG are staying the same. Any help as to which way is correct is much appreciated.
1
votes
Friendly reminder, make sure to specify that PARAM=REF otherwise you don't get referential coding.
– Reeza
I have specified it. Thanks for the reminder though!
– Uttara
1 Answers
1
votes
It doesn't really matter what you're reference groups is, the overall model will be the same.
The only thing that will change is how your word your explanation of an effect relative to your reference group.
Here are some articles that might help:
http://www.theanalysisfactor.com/strategies-dummy-coding/ http://www.theanalysisfactor.com/when-dummy-codes-are-backwards/
