Tyler Cowen and Alex Tabarrok have written a splendid article, "A Skeptical View of the National Science Foundation’s Role in Economic Research" in the summer Journal of Economic Perspectives. Many of their points apply to research support in general.
The article starts with classic Chicago-style microeconomics: What are the opportunity costs -- money may be helpful here, but what else could you do with it? What are the unexpected offsetting forces -- if the government subsidizes more, who subsidizes less? What is the whole picture -- how much public and private subsidy is there to economics research without the NSF? Too many good economists just say "economic research is a public good, the government should subsidize it."
They go on to ask deeper questions, "Are NSF Grants the Best Method of Government Support for Economic Science?" The NSF largely supports mainstream research by established economists at high-prestige universities. Are there better "public goods," undersupported by other means, for it to support?
Yes. Among others, replication and data. There are few current rewards for replication, and much economics research is not replicable. We live in the age of big data, but it's expensive and hard to access. The NSF has done commendable work here -- and other government agencies including the Census, Bureau of Labor Statistics, Federal Reserve, etc. provide huge public goods by collecting and disseminating good data. Without data we would not exist. That strikes me as the single most underfunded public good in the economics sphere.
I'm less a fan of their proposal to support "far out" research, naming "post-Keynesians, econo-physicists, or the Austrians." While they cite popular authors and a "gadfly'" sensational claims for the end of macroeconomics in 2009, in fact Macroeconomics is not all that much changed since the crisis and recession, and none of these claims -- nor the wackier approaches -- have in fact borne any fruit. Yes, it's easy to support mediocre incremental research, but government agency that must appear impartial can too quickly end up subsidizing crank research, of which there is plenty in economics (see my inbox!)
They ask a great question. If the government wants to subsidize economic research, why hand out grants, rather than hire people directly?
I think there are good answers here. Another big subsidy to economics research which they do not mention are the legions of government employees already doing it. The Federal Reserve, Treasury, OFR, CEA, SEC, CFTC, HHS, EPA and hundreds of other agencies employ thousands of PhD economists who spend considerable if not full time on "research," and are expected to write academic journal articles. Make up your own mind about the value of this effort. The success of the research university I think points to an important externality between doing research, teaching it, and evaluating it through service to the profession. Also, research coming out of government agencies always seems to find just how wonderful those agencies' policies are. However, replication and data production, or other more easily guided research seems a good fit.
Also not mentioned is the danger that government subsidized research ends up being politicized, or at least ends up calling for more government.
One of the main methods of NSF support is "summer support." Universities pay academics on a 9 month basis. If you get an NSF grant it pays for 2 months of "summer support." This is, of course, a fiction. In fact, most universities chop up the "9 month" salary into 12 pieces anyway. And most academics are not about to go work elsewhere in the summer -- it's the only time to really focus on research, and as Alex and Tyler point out the rewards to publishing are huge. By and large the NSF does not (or did not when I last looked in to it) buy off teaching or other duties, the one thing that might free up some marginal research time. Alex and Tyler mention low labor supply elasticities as a reason to be cautious about the effectiveness of support. They don't mention this system, practically guaranteed to be a pure transfer rather than induce more research.
On the other hand, NSF grants are typically awarded based on a working paper. They already are a "prize" as Tyler and Alex recommend. So perhaps the lump-sum nature of the reward is not such a bad idea, and ends up subsidizing good research rather than more effort.
I stopped applying for NSF grants some time ago. Sometime in the mid-1990s, I was driving through Indiana, and I saw a guy hooking a shiny new boat up to his pickup truck. It occurred to me, my NSF check for that summer was worth about 5 boats. I didn't think I could get out of the car and say with a straight face that he and four neighbors should forego their boats so I could work on unit roots for the summer. I'm not pure either; I still benefit from many government subsidies, not least of which the tax-deductibility of charitable contributions.