Aside from level-of-organization differences, how could you possibly get enough N and the correlation matrices to control for things like population and cultural differences, demographics, income differences, employment sector differences, macroeconomics, interaction effects, and so on?
This is a multi-thousand variable problem with no control sets, because none of the other polities share the same base variables.
Aside from level-of-organization differences, how could you possibly get enough N and the correlation matrices to control for things like population and cultural differences, demographics, income differences, employment sector differences, macroeconomics, interaction effects, and so on?
This is a multi-thousand variable problem with no control sets, because none of the other polities share the same base variables.
This is a great point, and there’s a recent paper which argues that this fact dooms state-level difference-in-difference designs.
Because there are literally only 50 states, you have terribly low power and little ability to isolate these other confounders.
How Much Should We Trust Modern Difference-in-Differences Estimates? - Amanda Weiss:
https://osf.io/preprints/osf/bqmws_v1