The most successful of these studies focus on the variation in Recovery Act spending across states. Some of this variation resulted from differences in the recession’s severity. For example, there was much more spending on unemployment insurance in Michigan than in Wyoming, because unemployment rose much more in Michigan. We wouldn’t want to look at that variation and say Recovery Act spending caused unemployment to be higher, because causation clearly ran in the other direction.
But some Recovery Act spending was allocated via formulas unrelated to economic conditions in particular states. For example, some road-repair expenditures were based on the miles of highways in each state, and some aid to state governments was based on pastMedicaid funding. This kind of spending provides a sort of natural experiment: some states received more treatment from the Recovery Act than others for relatively random reasons.
TWO careful studies have looked at the relationship between this formulaic spending and employment. Both find that states that received more money fared substantially better. This is the strongest direct evidence that the Recovery Act contributed to employment growth. Based on the estimated size of the effect, the studies suggest that the act created more than three million jobs.
See the rest of the article here