Perhaps you should re-emphasize the ~main takeaway in the intro?
Taking this decay into account shrinks the total economic benefit by 60% compared to GiveWell’s model. This would reduce the cost-effectiveness of nearly all their recommended deworming charities below their 10x bar for funding.
But you suggest that GW vastly discounted the one study that suggested very large LT benefits in an ad-hoc way.
Thus I saw the takeaway as possibly the opposite—maybe even understated benefits of deworming, but perhaps vastly more uncertainty.
And your pain point was a critique of GW’s approach. … I think you are saying “don’t do (semi-hidden) ad-hoc adjustments”
… do this as part of a transparent methodology. If you are incorporating prior beliefs, make this explicit.
If you want to include a study-specific random term as part of a Bayesian meta-analysis, make that explicit. (This is my suggestion but I think other commenters fleshed out this a bit more.)
Perhaps you should re-emphasize the ~main takeaway in the intro?
But you suggest that GW vastly discounted the one study that suggested very large LT benefits in an ad-hoc way.
Thus I saw the takeaway as possibly the opposite—maybe even understated benefits of deworming, but perhaps vastly more uncertainty.
And your pain point was a critique of GW’s approach. … I think you are saying “don’t do (semi-hidden) ad-hoc adjustments”
… do this as part of a transparent methodology. If you are incorporating prior beliefs, make this explicit.
If you want to include a study-specific random term as part of a Bayesian meta-analysis, make that explicit. (This is my suggestion but I think other commenters fleshed out this a bit more.)