evaluating statistical difference between model coefficients: 2 different models


I have two logistic regression models with the same dependent and independent variables. The only difference between the 2 models is the way the data are aggregated: one model is based on census tracts, and the other on parcels.

I want to determine whether there is a statistically significant difference between the models' coefficients. What do you think is the most appropriate test for paired differences in this case? Would it be Student's t-test?


the same data aggregated in a different way gives the same conclusions, I'm not sure what you mean, then, by "whether there is a statistically significant difference between the models' coefficients"?


No cake for spunky
People continue to confuse what a statistical difference is. They believe it means a substantive, aka large or not large, difference in the effect size, coefficients whatever. When in fact it does not mean that generally. Formally all these statistical test do is test the null; substantively I think they tell you whether the differences you find are likely to exist in the population (although I have never seen that stated by an authority, it's purely my understanding).

What it doesn't do usually is tell if you if two numbers are signficantly different from each other - so you know this is enough of a difference to matter substantively. There are some cases where this is not true apparently. Nested chi square difference test apparently do tell you if two values are signficantly different. F change test do tell you if R square, and predictiability has improved - although even in these cases I am not sure you can interpret this substantively as a "large" versus "small" change.