I have a binary dependent variable, and some numeric, binary, and ordinal independent variables. The whole idea is to create a predictive model based on all these data, which can be reported to others, i.e. an equation.
So let's say I clean my data, and use the glm function, I get something like this from coef(summary(my.model)):
Estimate Std. Error z value Pr(>|z|)
(Intercept) -4.327404e+01 4.900553e+02 -0.088304404 9.296347e-01
var1 6.789180e-02 2.593766e-02 2.617499440 8.857663e-03
var21 1.826271e+00 4.428294e-01 4.124096884 3.721922e-05
var3.L -3.871398e+00 1.179214e+03 -0.003283033 9.973805e-01
var3.Q -1.228205e+01 8.789344e+02 -0.013973801 9.888509e-01
var3.C -5.066052e-01 3.930721e+02 -0.001288835 9.989717e-01
var4.L -7.128841e+00 3.728134e+02 -0.019121740 9.847440e-01
var4.Q -2.954424e+00 2.152443e+02 -0.013725909 9.890487e-01
In this case the first var1 is just numeric values, var21 is binary (hence the 1 at the end?), and then the others are ordinal predictors with 4 and 3 factors respectively.
So let's just say this is the final model (it is most definitely not). How would I interpret this, and what should I report ?
I'm guessing that I will have some kind of term, which is inputted into the regular logistic regression equation, i.e:
prob = 1 / (1 + exp(-P))
where:
P = -43.2 + 0.067*var1 + 1.8*var2(1 or 0) + ...
And then what ? Do I use/keep both the linear term, quadratic term, and cubic term, or only select the one who has the lowest p-value (although in this case they are pretty similar and very high), or some other metric ?
