Size Matters


Not long ago I served on an NSF panel: a collection of researchers who look at some grant proposals and provide recommendations to NSF program officers who then make decisions about which, if any, of the proposals to fund. After the panel finished, the program manager asked us for feedback about various issues including the question: How large should NSF awards be? To put the question in context, in the programs that fund my kind of research, NSF offers “small” awards of up to $500,000 over up to three years, “medium” awards of up to $1.2M over up to four years, and “large” awards of up to $3M over up to five years. Few large awards are given and the bar for getting these is very high. Most grant proposals in a given size category ask for close to the maximum amount of money. For a given amount of grant money, should NSF (and other agencies) give more small awards or few large ones? I’ll present various arguments and give my opinion at the end.

  • Most grant money in computer science in the US pays PhD students around $20k per year to be “research assistants” working not only towards their degrees, but also towards larger research goals (however, in terms of grant money, the cost can be upwards of $50k per year due to university overhead and other factors). A problem with short-duration grants is that they cannot support PhD students for their entire lifetime in the system, requiring multiple grants to be pieced together.
  • Small grants cause significant administrative overhead at the NSF (running panels, negotiating budgets, etc.). Large grants reduce NSF overheads but push this burden out into the universities who now have to deal with more complicated accounting and reporting.
  • In Europe a lot of the money has moved towards very large grants. From what I hear, this necessitates a lot of jockeying for position, forming coalitions with the correct geographical distribution, etc. Nobody from Europe who I’ve talked to loves this or thinks it is very effective.
  • It can be difficult for small institutions to compete for large grants.
  • Small grants (unless accompanied by higher hit rates) force researchers to spend more time writing proposals, reducing the amount of real work they get done.
  • Large grants favor the status quo: they most often go to groups who have previously received large grants. Often, these groups belong to research centers and institutes that have evolved very sophisticated machinery for acquiring large grants and could not survive without them.
  • Some work is almost impossible to do using small grants. For example, expensive equipment or large software implementation efforts may be required to make a project succeed.
  • Large grants may engender collaboration, which can be beneficial. On the other hand, collaboration has overhead. Furthermore, some fraction of large grants support what I call “false collaboration” which happens when researchers get together and acquire a large grant, but then the individuals just go off and work on whatever they want, separately.

Clearly some sort of balance is needed. My guess is that most funding, maybe 80%, should go towards a large number of small awards, with the rest supporting carefully-chosen larger efforts.

,

4 responses to “Size Matters”

  1. In Germany politicians want to copy the US model and install an Ivy League. There is a special program for large grants, the “Excellence Initiative”, to strengthen the top universities.

  2. Great blog post. You’ve done a great job of assessing the tradeoffs.

    I too favor making a very large fraction of the funding dollars go to small grants. Here is my reasoning. I think the NSF panel review process is awesome: it does a great job of capturing merit, and ignoring other factors that should be irrelevant (like political/connectedness). As a result, the money ends up getting spent on projects that are most likely to have a significant payoff.

    Giving someone a large grant is basically giving them a lot of money and then asking them to use their own internal decision-making for deciding which projects to spend it on. In my experience, those internal decision-making tend to be driven less by merit. Merit is a factor, but so is giving money to your favorite colleagues, avoiding pissing off other colleagues, jockeying for position to end up with the most money, and other forms of politics.

    So, in terms of get the most research bang for the taxpayer buck, I tend to think that small grants are usually the most efficient, and the lion’s share of the money should be distributed in that way.

    P.S. This is probably obvious, but there’s a significant difference between “80% of the money goes to small grants” vs “80% of the grants are small grants”.

  3. > How is it working out, Andreas?

    Hard to tell so far. In Germany there are different kinds of universities and i am not sure how they map internationally. There is “Universität”, which is the general kind and focused on academia. A “Technische Hochschule” is a “Universität”, which concentrates on the hard sciences (MIT copied that). A “Fachhochschule”, which concentrates on applied science, i.e. training people for industry work, not academia. The popular opinion in academia is that the “Excellence Initiative” will degrade the lesser “Universitäten” to “Fachhochschulen”, because they cannot compete against the “excellent” universities in doing science. Their only chance is to specialize, which is bad for the students, because their education degrades in breadth. I do not see that the top universities gain much by additional money. At least in computer science, money is not a limiting issue.

    For example, a bigger problem, which is not addressed is that post-docs leave the country. There is no established “tenure track”. Instead of taking a junior professor job, it is advisable to go overseas and only return for a full professor job. This means Germany loses their good PhDs to the US.