Daniel Lemire's blog

, 2 min read

Jay and Return On Investment from Research Funding

Jay takes on the research funding debate:

I noticed a curious phenomenon at the NSF. They define R&D performance in a way that is completely baffling to me as an industrial scientist. Take a look at the first paragraph here. Performance is defined as money spent.

The truth is that funding bodies have no sensible way to determine how effective their funding is. Without any sensible metric, we are left with stupid things like counting papers or students. You could find worse metrics, but the tax payer is likely to be unimpressed. You don’t fund companies by the number of products they put on the market or the number of employees they have.

So yes, we should work harder to make a case for the usefulness of our research if we need lots of money. But this case cannot be made using a single metric like a patent. A patent in software engineering often makes little sense whereas it might make a lot of sense in mechanical engineering. Let the researchers give you their individual metrics. Let them tell their story as to why their work is important.

Things go sour when Jay looks at percentages…

Now Ernie thinks that basic research spending is too low. On the other side of the coin, I’m not sure how much public research spending is too much. The NSF calims that it only funds about 1/3 of all grant applications. At some point you reach the point of diminishing returns, where all the really good ideas are well funded, and you begin to fund a lot more junk.

There I disagree strongly. The long tail in science is important: you cannot guess where the important discoveries might come from and they might not be all from the same guy.

Low acceptance rates only make sense when the research overhead and scalability is high. For example, if you need half a million dollar in gear to even start the research, then you better give the funding to a couple of good places and let other researchers migrate there. But you must also take into account scalability: how many projects and subprojects can a professor run efficiently? Giving all the money to one person means that every dollar gets a very tiny fraction of this professor attention and at some point, the money will be wasted for sure. For theoretical research, you have low scalability (you can’t do 15 projects at once) and low overhead (no need for expensive gear), so you should aim to fund everyone a little bit. For other fields, the opposite is true.

So, is 1/3 low or high? For theory work, it is too low, for some high overhead research, it might be just the right number.