Control variables

We are using a set of control variables used by a federal agency to control for certain effects across all states. The DV is income. I suspect that the control variables work differently in fact in different states. So for example Hispanic in one state might earn the same as non-Hispanics, and in another state they might earn significantly less because what a hispanic means in practice may vary and the job markets and their relationship to it as well. In one state education might lead to higher income and in another state not (because you don't get employed if you have higher income in the second state).

In this case do variables that show you are in or not in one of these variables actually control for their impact?


Less is more. Stay pure. Stay poor.
This would likely be a case of effect heterogeneity if the variable is different across different states. So without knowing more about the models, just putting the variable in the model would likely only control for its overall effect, so it would blend the variables effect across the states.
I am not sure what you mean by overall effect. You mean this would be an average effect across state, but its actual effect would be different between states. Sort of like a MLM model.

The trick here is to use this to compare the states while ignoring those variables. I am not sure that is valid if there is effect heterogeneity, that is if it really controls for them. No one cares about these variables at all, they want to control for them to compare states on what they actually do control.

I don't think this achieves that which is really the issue.


Less is more. Stay pure. Stay poor.
I may be wrong, but if you think there could be state specific effects for a control variable across states, not including them via interactions just controls for its overall effect.