You named two charities, and I ended up deciding that the case for one of them was plausible (CiWF), so for any given dollar there's a 50-50 chance ;)
More seriously, I do have what you might summarize as a strong prior against cost-effectiveness estimates. In particular, I didn't address these issues:
Bollard's picking post-hoc an animal charity and intervention with especially clear positive track records. This has the following problems:
- Regression to the mean (I mentioned this but didn't properly account for it).
- Even if you earmarked the money for such programs, I expect there's some elasticity of substitution between different programs within a charity.
(Of course some programs could be secretly better than the cage-free egg campaigns, too.)
Here are some other costs I didn't account for:
- I didn't account for costs imposed on humans at all (see Jim's comment)
- To work properly, offsets require an allocation of credit that doesn't overcount. Bollard's conservative estimate tries to account for this, but this is pretty hard to do. To some extent we have to count all the prior work done on promoting compassion for animals, and account for compounding opportunity cost.
- In general I expect my environment to be marketing to me in non-truth-tracking ways. OPP is better than many but not perfect. In particular, I expect marketing to tend towards exaggerating the benefits of things that want my money.
If OPP or someone else claiming this impact had a relevant track record of publicly registered predictions of impact, and had actually gone back and checked and found that they were well-calibrated, then that would go a long way towards getting me to update.
I'm not sure what amount I'd put at 90% - my thinking on this is pretty bimodal, most of the 50% probability that the number's off comes from it being way off, not from it being a little off. For way off, I basically shouldn't anchor on public cost-effectiveness estimates at all.