LR model identifies a predictor as significant, but OR=1

pcc25 — Mon, 10 Oct 2016 15:03:52 GMT

I performed a logistic regression with stepwise selection, and a variable that was retained has on OR point estimate of 1.000, with 95% Wald CLs of 1.000-1.001. Is this likely a glitch, or are the "true" point estimate and lower CL just a titch above 1 and victims of rounding (e.g., lower CL 1.00001, point estimate 1.00008)?

The code:

PROC LOGISTIC DATA=train;

CLASS landcover (REF='11') / PARAM=REF;

MODEL case (EVENT='1')=landcover water_pct temperature precipitation / SELECTION=STEPWISE;

RUN;

Re: LR model identifies a predictor as significant, but OR=1

Rick_SAS — Mon, 10 Oct 2016 15:18:08 GMT

Because the variable was selected, I would guess the CI is [1 + epsilon, 1.001]. You can use the ODS OUTPUT statement to see the full precision of the estimates:

ods output OddsRatios=OR;

PROC LOGISTIC ...;
RUN;

proc print data=OR; 
format LowerCL UpperCL 16.14;
run;

topic LR model identifies a predictor as significant, but OR=1 in Statistical Procedures

LR model identifies a predictor as significant, but OR=1

Re: LR model identifies a predictor as significant, but OR=1