Controlling for something that does not vary :)

noetsi

Fortran must die
#1
We are supposed to negotiate with a federal agency. Part of the process is finding variables that explains the results that are not known (or used anyhow) by the agency in question. The problem is that they control for a variety of variables some of which are economic (these are tied to the state). They have the data for customers that are in all states, but we only have our customers in one state. I have calculated how our state varies on these measures from the national average but this is an aggregate measure. Every customer, who closed in our state, will have the same level on this variable the level of the state economic variable.

I need to control for these variables, but I can't figure out how to do that since our customers don't vary on them. The issue is that, for sake of argument, say gender is not in their model and we show its important. They could argue that its not important when controlling for economic factors, the factors I can't figure out how to add to our regression model.

One possibility is to use the level of the state when they closed (this varies by month). The problem is that this variance is likely not that different in the short run and I am not sure how you would code it. 24 distinct levels say for two years of closures would not be interval, but would be hard to code as a dummy variable.