Associated Press ran a "stimulus watch" piece in which it discusses a test that seemed designed to show that the stimulus failed. According to the article, the test involved comparing the changes in unemployment rates in counties that received the most amount of stimulus spending for road construction with those that received the least. The study found no differences in the movements in unemployment in these counties. In fact, it would be surprising if the tests found anything else. The road spending in the stimulus in 2009 amounted to $20 billion, about one fifteenth of the total stimulus. Assuming that this spending had a multiplier effect of 1.5, this means that it would have generated $30 billion of additional growth. That is equal to 0.2 percent of GDP. If its effect on employment was comparable to its impact on GDP, we would expect that a county with the average spending would see its employment rate increase by 0.2 percent or approximately 0.12 percentage points of the civilian population. There would be a comparable decline in its unemployment rate. This would be undetectable in any measure of county level unemployment, since random sources of fluctuation would dwarf this change, making it undetectable. In principle, counties that had two or three times this amount of stimulus would have a proportionately larger effect. However, the timing of the appropriation would not coincide with the spending. Some counties that received large appropriations in 2009 may not actually begin much spending on projects until 2010. Unless the study effectively controlled for spending rates, rather than just reporting appropriations, it would not be testing the relationship between stimulus spending and unemployment. (The multiplier effect would also take time to work through.) The other major problem with this approach is that many counties across the country are quite small. If they received large amounts of stimulus spending relative to their size, much of the employment would spill over to neighboring counties. In such cases, even if a substantial amount of employment was generated relative to the county's size, the impact on unemployment could be felt as much in neighboring counties as in the county that received the stimulus. Examining unemployment rates by county virtually guaranteed that the study would find no effect of the stimulus. It is almost inconceivable that it would have found a measurable decline in the unemployment rate. A more serious study would have looked at the changes in payroll employment by county and, in particular, employment in construction.
--Dean Baker
