I have a question. This article suggests that for a given utility function there is one single charity that is best and that’s the one one should give money to. That looks a bit problematic to me—for example, if everyone invests in malaria nets because that’s the single one that saves most lives, then nobody is investing in any other kind of charity, but shouldn’t those things get done too ?
We can get around this by considering that the efficiency function varies with time—for example, once everybody gives their money to buy nets the marginal cost of each saved life increases, until some other charity becomes best and all charitable giving switches to that one.
But we don’t have a complete and up-to-the-second knowledge of how many lives each marginal dollar will save in every charity, all we have to work with is approximations. In that situation, wouldn’t it be best to have a basket of charities one gives to, with more money going to those that save the most lives but not putting all the money on a single charity ?
Or is this consideration completely and utterly pointless in a world where most people do NOT act like this, and most people don’t give enough money to change the game, so rational actors who don’t have millions of dollars to give to charity should always give to the one that saves the most lives per dollar anyway ?
What happens in that situation is that people continue to invest in malaria nets, so much that the marginal cost of saving another life goes from say, $500 to $700, and for $600 dollars you can dig a well, saving another persons life. In essence, you donate to the most efficient charity until that money has caused the charity to have to pay more to save lives, and therefore stops being the most efficient charity.
I’ve thought about this problem before, but in the context of peer peer-to-peer file sharing.
The problem is that everyone is acting independent and with limited knowledge. It’s hard to know what other people are choosing. There may also be long delays between you and others paying and the cost changing.
Say that the optimal outcome is that out of $1000M, $200M is spent on insect nets and $800M on wells, and that you can only donate to one charity (too bothersome or high transaction costs or something). Now, if everyone is rational they are going to donate to the wells, and no one to nets. This is a suboptimal outcome. It’d also be difficult to coordinate the millions of people donating, so that just the right amount choose nets instead of wells. A solution to such coordination is to roll a dice. If everyone makes a random selection and lets the probability of choosing nets be 2/10ths then the expected outcome is just what we want.
Now, you can adjust this to how many (you think) are playing like yourselves. E.g. if you know most people are going to give to wells, perhaps it’d be better if you put higher probability on nets (perhaps 100%).
Yes, but there can be long delays between a donation happening and updates. Coordinating donations can be non-trivial, especially when flash crowds appear (e.g. sob story on reddit).
Also, such a randomized approach is not necessary if one can just donate small amounts to multiple projects instead (i.e. if transaction fees are not a problem).
I once donated some money to VillageReach a few minutes before getting the GiveWell newsletter issue announcing that VillageReach wasn’t going to be among the top charities in the next update because their founding gap had mostly closed and encouraging people to wait for the next update before deciding whom to donate money to. True story!
I have a question. This article suggests that for a given utility function there is one single charity that is best and that’s the one one should give money to. That looks a bit problematic to me—for example, if everyone invests in malaria nets because that’s the single one that saves most lives, then nobody is investing in any other kind of charity, but shouldn’t those things get done too ?
We can get around this by considering that the efficiency function varies with time—for example, once everybody gives their money to buy nets the marginal cost of each saved life increases, until some other charity becomes best and all charitable giving switches to that one.
But we don’t have a complete and up-to-the-second knowledge of how many lives each marginal dollar will save in every charity, all we have to work with is approximations. In that situation, wouldn’t it be best to have a basket of charities one gives to, with more money going to those that save the most lives but not putting all the money on a single charity ?
Or is this consideration completely and utterly pointless in a world where most people do NOT act like this, and most people don’t give enough money to change the game, so rational actors who don’t have millions of dollars to give to charity should always give to the one that saves the most lives per dollar anyway ?
What happens in that situation is that people continue to invest in malaria nets, so much that the marginal cost of saving another life goes from say, $500 to $700, and for $600 dollars you can dig a well, saving another persons life. In essence, you donate to the most efficient charity until that money has caused the charity to have to pay more to save lives, and therefore stops being the most efficient charity.
I’ve thought about this problem before, but in the context of peer peer-to-peer file sharing.
The problem is that everyone is acting independent and with limited knowledge. It’s hard to know what other people are choosing. There may also be long delays between you and others paying and the cost changing.
Say that the optimal outcome is that out of $1000M, $200M is spent on insect nets and $800M on wells, and that you can only donate to one charity (too bothersome or high transaction costs or something). Now, if everyone is rational they are going to donate to the wells, and no one to nets. This is a suboptimal outcome. It’d also be difficult to coordinate the millions of people donating, so that just the right amount choose nets instead of wells. A solution to such coordination is to roll a dice. If everyone makes a random selection and lets the probability of choosing nets be 2/10ths then the expected outcome is just what we want.
Now, you can adjust this to how many (you think) are playing like yourselves. E.g. if you know most people are going to give to wells, perhaps it’d be better if you put higher probability on nets (perhaps 100%).
The thing about that is, is that not everyone is donating at the same time, so that they can see the expected value change.
Yes, but there can be long delays between a donation happening and updates. Coordinating donations can be non-trivial, especially when flash crowds appear (e.g. sob story on reddit).
Also, such a randomized approach is not necessary if one can just donate small amounts to multiple projects instead (i.e. if transaction fees are not a problem).
I once donated some money to VillageReach a few minutes before getting the GiveWell newsletter issue announcing that VillageReach wasn’t going to be among the top charities in the next update because their founding gap had mostly closed and encouraging people to wait for the next update before deciding whom to donate money to. True story!