That’s a proposal that appeared today on Citizen’s Briefing Book, the digg-style section of the Obama transition website where interested citizens can post, and vote on, suggestions for the new administration.
On first glance, this sounds good to me! Currently, the NSF has a successful graduate fellowship program, which each year awards a few dozen students full support for their Ph.D. study, along with a $30,000 annual stipend. These students are the very strongest math undergraduates in the country, and the large majority go to one of the top 5 programs.
What if, instead, the NSF gave ten times as many students a much smaller package? Say, enough to supplement a TA salary by $5,000 per year, and to offer one or two semesters of full support to be used during the dissertation year? That would make a big difference to a lot of students at state universities, who otherwise have limited time off from teaching.
The argument for using the money to fund a small number of full-tuition fellowships, I guess, is that budgets are limited even at Harvard and Princeton, and the NSF fellowships allow more students to train at top departments than otherwise would. From running graduate admissions at Princeton I know very well that many, many students who would succeed at Princeton get rejected in favor of even stronger applicants. But graduate admissions are limited by things besides money — primarily the time and attention of the faculty.
The justification for the huge stipend is even harder to imagine. It’s more than doubled since I was a grad student fellow in the mid-90s. If a student is deciding whether to go to grad school on financial grounds, $30K is the same as $15K — bubkes next to what a star math major can make in industry.
So what would happen if the NSF gave many small fellowships instead of a few big ones? The students at Harvard and Princeton would still be fully supported, maybe teaching one semester to get some classroom experience. Maybe some students who otherwise would have gone to Harvard and Princeton would go to Chicago or Columbia instead. And a lot of U.S. students at good places like Wisconsin would write better dissertations faster.
Of course, the effect would also include a big transfer of wealth from the top-5 math departments to my graduate students; so maybe my self-interest is showing here.
See also: I. Laba’s thoughtful post on Canada’s NSERC Discovery Grant program, and whether it should switch from it’s current “many small grants” model to a “few large grants” model like the current NSF Graduate Fellowship.
Quomodocumque 
One argument in favor of the current system is that the large support gap between “top 5” and private schools further discourages the (few) top students whose first choice is to attend State University X to work with Professor Y from doing so. This is a very real consideration where a top 5 school (such as Princeton) provides roughly the same support as a state school, but without the 20 hours of TA work. For a NSF award winner the two may be financially comparable. In this sense the proposal might hurt state schools. Admittedly, I can’t point to many cases where this was a determining factor.
An alternative proposal (which would be difficult to implement) would be for the NSF to “fill in the difference” between a students’ default level of support at his/her institution and some set rate. The idea being this would only be a little at a private institution and a lot at a public institution. Assuming a lot of award recipients attend private institutions, this would allow for more awards. Ironically, the system is currently set up in the inverse of this. Since the NSF awards cover “tuition” (which is always somewhat an accounting scam, as it is usually waived for supported students) which is much larger at private schools than public schools. This probably results in the top 5 university getting 10-20k more out of having an NSF winner than the state school.
What is the stipend intended to cover? Poverty level for the 48 states plus D.C. in 2008 is $10.4K, so it does indeed look to me like there’s a huge difference between $15K and $30K if you have to pay food, rent, utilities, etc. out of it, to say nothing of tuition, feeds, textbooks, etc. And that figure is kept artificially low: most states set the threshold of “indigent” at 125% or 150% of it.
Smaller fellowships? That’s the thin edge of the wedge. Graduate fellowships today, research grants tomorrow. Moreover, your costing seems a little suspicious. Tuition is usually only waived for TAs; if I support a graduate student for a semester on my grant, I need to pay their tuition (not to mention overhead).
As the following table shows:
http://www.nsfgrfp.org/about_the_program/statistics_and_past_recipients
the success rate for graduate fellowships is approximately 10%. I certainly don’t think this number should be 100% (of course, if it’s “easier” to get a grant, more people may apply), although, I agree that 30K is a little excessive. On the other hand, I think the NSF believes that one purpose of grants is to reward/promote excellence, rather than “spread the wealth”.
John: Oh, there’s certainly a huge difference between $15K and $30K; I’m only saying that I doubt the pain of living on $15K/yr would deter an NSF-level student from pursuing Ph.D. study. At least, I never encountered such a case.
Anon 3: you’re right, of course — even one semester of graduate student support without TA is mega-expensive, carving out a big chunk from a typical NSF research grant. But as expensive as it is, it is much less expensive than the 8 semesters that I believe the current NSF fellowship covers! Anyway, I certainly agree that all physical numbers in the above post are to be taken as little more than indications of inequalities.
As for research grants, there’s at least the argument that a lot of money in one person’s hands allows for certain kinds of activities (e.g. organizing conferences) that are harder when money is more dispersed. I’m not sure how much weight to give this.
Well, I always thought the main reason for the $30K NSF fellowship was to ensure that it is prestigious. At least when I was a student (up till 2005), some science departments at the top universities (most notably biology and physics) paid students up to $27K/year, and the impression I got was that the NSF was trying to exceed that.
In the past I thought about two improvements: one is to have different NSF fellowship salaries in different fields based on market values. Thus for Math, $25K might be enough, but for physics it may not be. An even better approach is to note that departments will be generally happy to take a strong student for “cheap”, not just for “free”, and condition the NSF money on matching funds from the department. This is already the case wrt tuition: the NSF pays the university $10K for tuition and the university may not charge the student any more (it is a matter of internal university accounting whether the balance is waived or billed to the department; I have no idea if the university can have it billed to a faculty member’s grant).
Thus, the NSF might say “we’ll pay $18K/year (plus $10K for tuition); the department must provide a matching fellowship of $12K/year”. Tweakings include allowing a limited amount of TA work (teaching experience is useful), or having the matching funds bring the support to at least $5K above the usual support for a graduate student rather than the fixed $30K/year.
When I was last at NSF, this idea was raised in a similar situation but the response of DMS was that it’s a lose because there are bean-counters checking what percentage of submissions get funded and they would note a huge increase in the percentage success of Math if more smaller grants were awarded. This could then give reason for a lowering of the DMS budget, to bring the percentages in line with other areas. This is why institutes are favored – one large grant (satisfies the bean-counters) but payoff to lots of people.
Nigel’s right. This is a perennial problem at DMS, since many influential NSF staff are experimental scientists. Experimental research is usually quite expensive and the amount of work you can do in a given area is more or less proportional to your funding (in particular, with no funding your lab would simply shut down). When experimentalists learn that math research is cheap, they lose respect for it. When they learn that much of it could be done successfully without an NSF grant at all, they decide it’s a crime to give money to mathematicians when it could be spent on more experiments. The net effect is that NSF is full of people who are convinced that math gets more money than it “needs” (based on their field’s definition of need, not a cost/benefit analysis). The DMS staff have to fight many political battles over this, and they can’t afford to do anything that might give ammunition to the other side.
Comments 6 and 7 seem to discount the fact (which has been previously mentioned) that the success percentages are artificially increased by self selection (students who think they aren’t competitive don’t apply). Making the awards twice as easy to win, I would bet, would more than double the applications.
Moreover, I’d imagine that changes such as those being discussed would be implemented across disciplines. This would also help address 6 and 7’s comments.
$30,000 a year now? Holy crap.
I have felt the same way that Jordan does for many years now. $30 K per year (American, of course) is comparable to what I was making as a postdoctoral fellow at McGill University. The fact that they never seem to have trouble getting top people for these positions in Montreal is a pretty convincing argument against the “We need to pay the best this much more, or they’ll leave.”
So I was very interested to read the “bean-counting” arguments 6) and 7) above, which are at least logical and may well be true. If this is really the way DMS/NSF thinks, then the mathematical community can adapt in response. For instance, we could be much stronger in our encouragement of all (U.S. permanent resident) grad school applicants to also apply for an NSF graduate fellowship, up to the point of actually requiring it as part of their school application. The way I understand things, this would cause the NSF to dramatically increase the total number of grants awarded.