Erich Grunewald writes:
Since you’re interested in sabermetrics, I thought I’d write you to see if you have any advice/resources/examples on how to calculate a person’s relative counterfactual impact for different employment options?
My interest in this is to figure out how you (as an EA, say) should choose a job if you’re optimizing for impact. (“Impact” could be revenue produced, lives saved or something else.) In sports I guess it’d translate to a player figuring out on which team he or she would produce the most counterfactual goals for the league or something like that. So while WAR compares players, I’m looking for a method that players could use to compare team options. Do you have any ideas?
Here are some examples of things that’d be useful to me:
– Thoughts/ideas on how to go about calculating or thinking about this.
– Examples of models (doesn’t have to be in hockey or even sports) that do something like this.
– Resources that talk about similar problems.
– Suggestions for people to send a similar email to.
My response: I don’t know! My quick thought is that any given person doesn’t have that much choice on what job to take, but I guess that impact should be one factor. I can’t really picture this being done in any quantitative way, though. Maybe commenters have other thoughts?
One obvious example of something like this is redirecting asteroids, eg see the DART mission:
https://en.m.wikipedia.org/wiki/Double_Asteroid_Redirection_Test
A useful counterfactual requires having a model that makes accurate and precise predictions. And *that* requires using methods that can tell you the accuracy/precision of the predictions.
Fields like social science and medicine would need to get beyond that first step.
This sounds like an ideal place for fuzzy logic, rather than probabilities and statistics.
I took one stab at this: https://doi.org/10.31235/osf.io/u2jcr
This sounds like exactly what 80,000 hours is aimed at? https://80000hours.org/