Thank you, @Kevin Xia 🔸 , for the text! I also find the Shapley value very interesting for attributing impact—I wasn’t familiar with it before, so thanks for the hint, @Vasco Grilo🔸 !
I think it depends on what decisions are being guided by the “impact share.” If the goal is to determine how a donor should allocate their money, then in your first example, the Shapley value is probably more suitable than simple counterfactuals. However, if Organization A has already decided that it has fulfilled its role in securing a corporate commitment and now Organization B is deciding whether to do the same, then counterfactuals are useful here (which are identical to Shapley values with only one actor).
Even though the Shapley value is a good reference point for donors when distributing funds, I don’t think the best overall strategy is necessarily to donate to the charities with the highest cost-effectiveness in terms of Shapley value. Instead, donations should also be “coordinated.” This becomes particularly clear in the second example: If Organization A and Organization B had nearly the same costs for referring to a grant, their cost-effectiveness would also be nearly the same, and a donor would most likely have to support either no charity or both charities for that purpose. It is obviously smarter to fund only one (or none) of them in this case.
One solution would be to first fund the project with the highest cost-effectiveness (in Shapley value) and then recalculate the Shapley values. In the second example, this would mean that first, Organization A or Organization B is funded (whichever has slightly lower costs), and then the other organization is no longer funded for this purpose.
However, in the first example, problems could arise if the total donation budget is insufficient to fund both organizations, meaning that in the end, the money has no effect at all.
Even though this scenario may seem unrealistic (since Organization A’s actions would likely still have a positive impact, even if Organization B does nothing), this problem also appears in a slightly modified model that may be more realistic. Let’s assume that if Organization A or Organization B acts alone, they would help 200,000 chickens. The Shapley value per organization would still be 300,000 chickens, but if the funds are not sufficient to support both organizations, funding one of them would only help 200,000 chickens. In that case, it would be better to fund a third charity, Charity C, which could help 250,000 chickens (in Charity C’s campaign, no other organizations would play a role).
Thank you, @Kevin Xia 🔸 , for the text!
I also find the Shapley value very interesting for attributing impact—I wasn’t familiar with it before, so thanks for the hint, @Vasco Grilo🔸 !
I think it depends on what decisions are being guided by the “impact share.” If the goal is to determine how a donor should allocate their money, then in your first example, the Shapley value is probably more suitable than simple counterfactuals. However, if Organization A has already decided that it has fulfilled its role in securing a corporate commitment and now Organization B is deciding whether to do the same, then counterfactuals are useful here (which are identical to Shapley values with only one actor).
Even though the Shapley value is a good reference point for donors when distributing funds, I don’t think the best overall strategy is necessarily to donate to the charities with the highest cost-effectiveness in terms of Shapley value. Instead, donations should also be “coordinated.” This becomes particularly clear in the second example: If Organization A and Organization B had nearly the same costs for referring to a grant, their cost-effectiveness would also be nearly the same, and a donor would most likely have to support either no charity or both charities for that purpose. It is obviously smarter to fund only one (or none) of them in this case.
One solution would be to first fund the project with the highest cost-effectiveness (in Shapley value) and then recalculate the Shapley values. In the second example, this would mean that first, Organization A or Organization B is funded (whichever has slightly lower costs), and then the other organization is no longer funded for this purpose.
However, in the first example, problems could arise if the total donation budget is insufficient to fund both organizations, meaning that in the end, the money has no effect at all.
Even though this scenario may seem unrealistic (since Organization A’s actions would likely still have a positive impact, even if Organization B does nothing), this problem also appears in a slightly modified model that may be more realistic. Let’s assume that if Organization A or Organization B acts alone, they would help 200,000 chickens. The Shapley value per organization would still be 300,000 chickens, but if the funds are not sufficient to support both organizations, funding one of them would only help 200,000 chickens. In that case, it would be better to fund a third charity, Charity C, which could help 250,000 chickens (in Charity C’s campaign, no other organizations would play a role).