I think your cost-effectiveness analysis is a little bit misleading: You’re assuming there was a binary choice between spending $0 and having a 5% chance of a seat or spending $13mn and having a 30% chance*. This is not the case, though, as they could have spent anything between those two amounts. It is quite reasonable that spending say $3mn would have led to something between 25-28% chance of winning and spending $10mn probably to something like 29.9%, so the effectiveness of most of the $13mn spent is much lower than you are suggesting.
*I’m much more sceptical than you though that the Metaculus estimate is a reasonable guess: It seems plausible to me that there could be a significant overlap between people highly excited about Flynn and people estimating on this question, which could very well have biased the estimate in Flynn’s favour.
It might be cheaper than this because perhaps most of the value was created by some small proportion of the money. Perhaps if you spent 60 million across 60 races, you’d expect more than 1 pro pandemic preparedness seat.
It might be more costly than this because metaculus was miscalibrated. I find this a little frustrating because “maybe the well-calibrated forecasters were wrong” is an exceptionally cheap attack against the best forecasting we had ahead of time. That said, we’ll find out in future races anyway.
On balance I reckon the first bullet point dominates the second, so I reckon this is an overestimate of costs. It probably costs less than $60 mil to campaign on expectation.
I think your cost-effectiveness analysis is a little bit misleading: You’re assuming there was a binary choice between spending $0 and having a 5% chance of a seat or spending $13mn and having a 30% chance*. This is not the case, though, as they could have spent anything between those two amounts. It is quite reasonable that spending say $3mn would have led to something between 25-28% chance of winning and spending $10mn probably to something like 29.9%, so the effectiveness of most of the $13mn spent is much lower than you are suggesting.
*I’m much more sceptical than you though that the Metaculus estimate is a reasonable guess: It seems plausible to me that there could be a significant overlap between people highly excited about Flynn and people estimating on this question, which could very well have biased the estimate in Flynn’s favour.
So I’m hearing the following:
It might be cheaper than this because perhaps most of the value was created by some small proportion of the money. Perhaps if you spent 60 million across 60 races, you’d expect more than 1 pro pandemic preparedness seat.
It might be more costly than this because metaculus was miscalibrated. I find this a little frustrating because “maybe the well-calibrated forecasters were wrong” is an exceptionally cheap attack against the best forecasting we had ahead of time. That said, we’ll find out in future races anyway.
On balance I reckon the first bullet point dominates the second, so I reckon this is an overestimate of costs. It probably costs less than $60 mil to campaign on expectation.
What cost effectiveness guesses would you give?
What do you think the cost-effectiveness was?