Jared Bernstein calls employment in the DC area compared to the nation a "natural experiment" demonstrating the effectiveness of fiscal stimulus. He writes:
"It’s not like our area is totally insulated from recession, but the countercyclical expansion of federal spending does give the region a notable edge compared to much of the rest of the country. (Though the point of my presentation was that with the fed gov’t poised for contraction, our region needs to diversify.)
Note how the jobs line for the DC region is flat while that of the nation falls steeply. It’s a bit of a natural experiment: if you actually apply some serious Keynesian stimulus, you can minimize job losses."
It was wrong when the Conley-Dupor paper did it, and I said it, and it was wrong when the Feyrer-Sacerdote paper did it, and I said it (although I did think they had a nice instrument - it just wasn't an instrument that could identify the parameter they were looking to identify).
You can't look at differences between sub-regions and get macroeconomic impacts. This should bother people for and against stimulus. First, if stimulus has a positive impact, then the fact that the states are open economies on the same currency will create spillover effects that bias the impact estimates downward. If stimulus doesn't have a positive impact and there's crowding out, then any gains in one state are going to come at the expense of other states, and inter-state differentials will be biased upwards. As one of my mentors at Urban in all things econometric likes to put it, "the estimates are mush".
Identification of impacts in macroeconomics is hard. You can't just look at raw data like John Taylor does. You can't do inter-state comparisons like these people do. You have to do hard work identifying your model. Role models in empirical macro should be economists like Robert Barro and Christian Romer.