continuous and categorical variables in logit regression model

Question

Can any one guide me on my model in this form:

1 or 0 = (X1 + X2 + X3 + X4 + X5 + K1 + K2 + K3 + K4) + (X1*K4 + X2*K4 + X3*K4 + X4*K4 + X5*K4) + (C1 + C2 + C3).

X1 to X4 and K1 to K3 are independent and continuous variables,

K4 is a dummy variable and a moderator variable,

C1 to C3 are control and continuuos variables.

I have more than 580 observations. When I run the logit regression I discovered that the moderator in this case K4 is not statistically significant but the interaction terms (X1*K4 + X2*K4 + X3*K4 + X4*K4 + X5*K4) were statistically significant.

Moreover, when I checked the odds ratios, I found that X2 and K4 have odds ratios running in 2000+.

What is the problem of my model? I am using Stata 14.

Answer 1

There are two issues here and I will take your second one first.

Large odds ratios of the size you describe are usually the result of what is called separation. That is to say that some combination of your predictors (almost) perfectly predicts your outcome. You need to investigate your data in more detail to see where this happens and then decide what to do about it. If your goal is prediction then it is helpful but if your goal is explanation of the science it can be more tricky.

You should not try to interpret main effects, in your case k4 in the presence of interactions. This is because the interaction is telling you that the effect of k4 varies with respect to values of another so there may not be anything which you can meaningfully call the effect of k4. But in this case you really need to deal with the separation issue first.