I think of it more as a Type 1 versus Type 2 error tradeoff
I realized earlier this morning that I had forgotten my main point, and so the sibling comment only hints at it instead of making it explicit: many people talk about plans with the assumption that all of them are on the possibilities frontier, and so the relevant thing is moving along the possibilities frontier until they’re at the right tradeoff.
But being optimal is surprising—one should assume that there is lots of room for growth, and should try to get more of everything (i.e. move perpendicular to the perceived frontier) until it’s clear that they are actually on the frontier. (In the stats case, getting more data means both less Type 1 and Type 2 error.)
I realized earlier this morning that I had forgotten my main point, and so the sibling comment only hints at it instead of making it explicit: many people talk about plans with the assumption that all of them are on the possibilities frontier, and so the relevant thing is moving along the possibilities frontier until they’re at the right tradeoff.
But being optimal is surprising—one should assume that there is lots of room for growth, and should try to get more of everything (i.e. move perpendicular to the perceived frontier) until it’s clear that they are actually on the frontier. (In the stats case, getting more data means both less Type 1 and Type 2 error.)