The NIH's "high risk" grant programs: https://commonfund.nih.gov/highrisk Anyone know on what basis they call it high risk? Do they actually track failure rates? Or is it just signalling?
-
Didn't we talk about this a while back? Incentivizing risk appropriately is a hard problem, you can get cases where people would aim to pick surefire failures. (I think akin to how optimizing for calibration over discrimination can mislead.)
Sorry, I'd forgotten, you're right. That particular issue doesn't seem difficult to sort out. If nothing else, just monitoring the issue would likely be sufficient (e.g., it may turn out to only be a small problem, in which case doing nothing but monitor would be enough).
