|
|
I'd like advice on how to deal with differing Exp(B) and odds ratio estimates in CSLOGISTIC.
I'm using the complex samples menu (CSLOGISTIC procedure) to calculate the effect of various categorical (A,B) and continuous independent variables (C,D) on a binary outcome (E).
The model includes an interaction term between the variables (A*B*C). A is a binary predictor variable, dummy coded as 0 or 1.
Model effects for A are significant, p=0.023.
CSLOGISTIC automatically uses the highest category as the reference category, giving in the parameter estimates for A:
Exp(beta)=1.555, 95% CI 1.09 - 2.218 when A=0
In the same model, I specify an odds ratio calculation for the variable A, producing.
OR = 1.22, 95% CI 0.825-1.791.
Is there a clear explanation for this difference, where the 95% CI for Exp(beta) does not cross 1 but the OR does cross 1?
My understanding is that Exp(beta) assesses the average effect across both levels of the predictor, whereas the odds ratio specifically contrasts one level of the predictor (A) against the other. Is this correct? Is it the only difference in the approaches? If so, is there a reference detailing how one approach has been chosen over the other?
|
Locked
