The Federal Reserve Bank of San Francisco has the technical paper Highway Grants: Roads to Prosperity? on its web site. The abstract for the paper is:
Federal highway grants to states appear to boost economic activity in the short and medium term. The short-term effects appear to be due largely to increases in aggregate demand. Medium-term effects apparently reflect the increased productive capacity brought by improved roads. Overall, each dollar of federal highway grants received by a state raises that state’s annual economic output by at least two dollars, a relatively large multiplier.
Near the end, the paper addresses some differences between this result and previous published results.
Our estimated multipliers are noticeably larger than those typically found in the literature on the effects of government spending. For instance, in a recent survey, Valerie Ramey reports multipliers between 0.5 and 1.5 (see Ramey 2011b). One possible reason for the wide differences is that we consider a very different form of government spending. Most of the literature concentrates on the multiplier effect of military spending. But such spending is arguably nonproductive in an economic sense. By contrast, government investment in infrastructure, such as roads, can raise the economy’s productive capacity. In that respect, it can have a higher fiscal multiplier. Another difference is that we concentrate on the multiplier effect on GSP, while the literature typically studies the effect on U.S. GDP as a whole.
This last part goes to the irony of Republican claims that governments don’t create jobs, but cuts in defense spending will cut jobs. In fact, government defense spending is less efficient at creating economic output than infrastructure spending. If the defense spending multiplier is really 0.5, then this would mean that annual economic output is raised 50 cents for each government dollar spent on defense. Depending on where that government money came from, defense spending could actually lessen economic output.
People not familiar with the fancy statistical methods used by economists these days may have qualms at the way the result was derived. It is probably worthwhile to maintain a modicum of skepticism at such deep data mining.