I am running a logistic regression in R and for some reason, some of my outcome variables o a ridiculously high log odds (ie -14), which corresponds to a p value of 0.9. But this does not make sense to me as e^-14 is an odds close to 0, which is well below no difference in odds. Can someone explain what is going on and why this is happening?
I do notice that the ones that this occur on seem to be those with a very low in one group (ie: 22/22000), but still does not help me with interpreting this.
Call:
glm(formula = as.factor(Y) ~ BMI_NUM + SEX + AGE + RACE,
family = "binomial", data = df)
Deviance Residuals:
Min 1Q Median 3Q Max
-0.6620 -0.1430 -0.1188 -0.0987 3.4692
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -18.58551 549.85748 -0.034 0.97304
BMI_NUM 0.02014 0.01187 1.697 0.08964 .
SEXmale -0.15877 0.16449 -0.965 0.33443
AGE_90ormore -0.43949 0.21364 -2.057 0.03967 *
AGE30_39 -0.03423 1.02629 -0.033 0.97340
AGE40_49 0.14823 0.61841 0.240 0.81056
AGE50_59 -0.37233 0.41475 -0.898 0.36934
AGE60_69 -0.30675 0.25620 -1.197 0.23119
AGE70_79 -0.45086 0.20601 -2.189 0.02863 *
AGElessthan30 -13.84406 783.82263 -0.018 0.98591 ######This One
RACEAsian -0.64351 0.71603 -0.899 0.36880
RACEBlack 0.67466 0.27934 2.415 0.01573 *
RACEOther -13.74345 540.78513 -0.025 0.97972 ######And this one
