long time listener, first time caller. Unfortunately I can't show code as the computer the analysis is done on has rather tight security.
I need to test for joint significance in a logit model that includes an interaction term. Normally I would use a likelihood ratio test, but every time I do it results in: Pr(>Chisq) < 2.2e-16. This is even if I try using random numbers as an IV. If I decrease the sample size, I can control the likelihood ratio. However, choosing a sample based on the test statistic seems like a faux-pas.
Can anybody give me some insight into what is going on / suggest a different way to test my model with a large sample (>700,000 obs)