Should granting agencies fund projects or people?

Basic research grants, awarded to individual investigators or small teams of investigators in order to pursue specific  projects, are increasingly difficult to obtain. Applications have been increasing and success rates decreasing for years at the US National Science Foundation (NSF), US National Institutes of Health (NIH), UK Natural Environment Research Council (NERC), and other government funding agencies. Success rates at many agencies are now below 10%. In response, applicants submit more and more applications in the hopes of getting at least one funded, and repeatedly revise and resubmit the same project. This creates a significant time burden on peer reviewers, as well as on the applicants themselves. Low success rates also discourage development of long-term or risky research programs by individual investigators.

The easy 'solution' is to call for more funding, but in the current economic climate that is unrealistic. More importantly, as NIH discovered a few years ago (when their funding was rapidly doubled), more funding simply causes more investigators and institutions to chase that money, so that per-capita success rates end up no higher than they were before.

I wonder how many of my colleagues are aware that there's an alternative. Canada's Natural Science and Engineering Research Council is the Canadian equivalent of NSF or NERC, funding basic research by individual investigators in non-biomedical fields. But NSERC's strategy is, as far as I know, unique among major government funding agencies. Put briefly, NSERC funds people, not projects. For details see here, but the salient points are as follows:

• NSERC Discovery Grants are (typically) 5 year awards to individual investigators to fund their research programs (as opposed to individual projects within those programs). The goal is to fund excellent research programs in a long-term, sustainable way.
• Applications are 5 pages long (not counting bibliography, budget, etc.). The emphasis is not on the technical details of specific studies, and for this reason preliminary data are rarely included. Rather, the applicant describes and motivates the long-term goals of their research program, summarizes progress over the previous 6 years, and (in the remaining 4 pages or so) lays out their plans for the next 5 years of their program.
• There's no expectation that you'll do whatever you said you'd do in your previous grant. In evaluating your renewal application, NSERC evaluates the science you've done on its merits; they don't care whether or not it's the science you proposed to do 5 years ago. You need to propose a plan for the next 5 years not because NSERC expects you to necessarily follow through on it, but because if you don't have any plan at all the odds that you're going to do good science over the next 5 years are slim, even if your track record is good.
• You can only hold one Discovery Grant at a time (because by definition, you only have one research program).
• Success rates are about 60% (a bit lower for new investigators), which is actually down from 80% 10 years ago. This is in part because 1/3 of the evaluation relies on your track record (publications and other outputs over the previous 6 years). If you do good science, and you propose a reasonable plan for the next 5 years, you're going to get renewed. You might get cut (or increased!) substantially over your previous grant, but frankly you're not going to get cut off entirely unless you deserve it. High renewal rates are necessary if your goal is to fund long-term research programs in a sustainable way (unless you think giving out 25 year grants with very low success rates would be a good idea)
• Keeping success rates high means keeping the average grant size relatively small--about $35,000/year. The very best people can get up to about $140,000/year, and the minimum is roughly $15,000/year.
• There are also some technical differences between NSERC grants and grants issued by agencies in other countries, which I'll note just for completeness. Canadian academic appointments are 12-month appointments, so applicants don't need to apply for (and NSERC won't pay) summer salary for themselves. Nor do applicants need to budget for overhead; overhead is handled separately. So the difference in size between a typical NSERC grant and, say, an NSF grant is less than it might seem at first glance, since no NSERC money is going to the investigator's summer salary or to overhead

As a Canadian academic, I love the NSERC system. I love being able to sustain a baseline level of funding for my lab by writing one 5-page grant every 5 years. I love not having to review repeated revisions of other people's grants. I love that the system cares about my track record, which considerably reduces the stochasticity of decision making. If a single referee misunderstands or doesn't appreciate your NSF application, you're doomed. But if you have a track record of publishing regularly in good journals, no referee is going to fail to appreciate that. I love that the system doesn't bog everyone down in picky technical details.

It's no problem that I can't afford to pay a grad student's entire salary, because Canadian grad students are mostly funded by TAships, scholarships, and fellowships during the academic year. I only have to pay them a bit of summer salary. Even so, I still can't afford more than about 3 students on my current grant, but that's ok because that's my ideal lab size.

Are there features of the system I don't love? Sure. I wish I had enough money to pay a postdoc or a full-time technician on my NSERC grant, but I don't (I'm not one of the 'high fliers' whose NSERC is worth >$100K/year). I'd like to get into experimental evolution, but I probably couldn't afford to do so on my current NSERC because gene sequencing is still too expensive. But for me, these drawbacks are trifles compared to the advantages.

I wonder how low success rates at NSF, NERC, and other agencies will have to drop before the Canadian system starts to look good to my colleagues in other countries. I've actually asked numerous colleagues this question, and answers have ranged from "The Canadian system already looks good to me!" to "They can never drop that low." The latter view was expressed by a colleague who likened an NSF grant to a hit of crack cocaine--it's so big that you'll do anything, and put up with anything, to get another hit!

Conversely, a Canadian colleague once suggested to me that NSERC should start a new program of big, NSF-style project-based grants, funded by cutting everyone's Discovery Grant by 5-10%. His theory was that we're all cocky enough to think that we'd be one of the rare few who got one of these big new grants. He's wrong about that--I'm cocky, but not that cocky. More precisely, I'm sufficiently risk-averse that I'm not willing to give up 10% of my NSERC for what's effectively a lottery ticket. I think, in the long run, I can do more and better science in a system where my expected grant size, and the variance around that expectation, are both modest, than in a system in which the mean might (or might not!) be higher, but the variance is also higher (with a huge mode at zero).

Who's with me?