Tyler Cowen asks where he can get ahold of the Associated Press  study showing that unemployment rates were not affected by stimulus spending on transportation and infrastructure projects and is told there is no study to see, only statistical results. I’m no defender of infrastructure spending as stimulus, and watching a perfectly good sidewalk outside my office get a completely unnecessary and inexplicably slow repaving hasn’t helped my optimism either.  But I think the authors of this study have some questions to answer before anyone should take the results seriously.

  • Endogeneity –  Do you have any reason to assume that getting more stimulus money wouldn’t be correlated with needing more stimulus money? If areas that got the most stimulus money did so because they looked to policymakers like they were going to have higher unemployment, and your analysis shows that the unemployment these areas is the same it is in areas that got less stimulus, then that is evidence that it worked; they were expected to do worse, and they didn’t.
  • Spatial Relationships – Was it tested whether spending in neighboring counties impacted employment in other nearby counties? Were the effects tested at different levels of aggregation? As Brad DeLong points out, if spending in one county creates jobs in nearby counties that had no spending, then the results will be biased downwards.
  • Controlling for Other Stimulus – Did they test whether infrastructure stimulus negatively correlated with other stimulus spending? If counties that received more infrastructure stimulus spending received systematically less of other kinds of stimulus, then you might not pick up any positive impacts.

If I read an academic paper making the same claim as this paper, I would be very curious to know how they could have any certainty about their results unless they controlled for these things. The endogeneity problem by itself is significant enough to warrant skepticism.