Wednesday, January 15, 2014

It's harder to favor a specific cause in more efficient charitable markets

Summary: People vary widely in their views on the relative importance of different causes, interventions, and charities. Those with strong idiosyncratic beliefs favoring one might expect they will have much greater impact by focusing on the favored cause. However, "smart money" which moves in pursuit of marginal returns, can mean that targeted support simply displaces flexible support, instead of adding to it. For example, attempts to favor one of GiveWell's recommended charities relative to others can easily be thwarted as GiveWell attends to room for more funding and diminishing returns in making its recommendations. Insofar as the effective altruist movement increases this dynamic, it will tend to link different causes and interventions together.

It's hard to support just one of GiveWell's recommended charities
Suppose that one thought that one of GiveWell's recommended charities, e.g. the deworming charity SCI, was much better than GiveWell judges. In other words, that at funding levels where GiveWell would be indifferent between another dollar for SCI or GiveDirectly or Deworm the World Initiative, in fact SCI does 10 times as much good with marginal dollars. In that scenario, should one believe that giving to SCI will do 10 times as much good?

[Edited to add: in comments Holden Karnofksy of GiveWell mentions that, at least within a single giving season, GiveWell and Good Ventures, fix a donation plan without real-time adjustment. However, from one giving season to the next GiveWell does drop and change recommendations based on funding levels, e.g. dropping AMF from its recommended charities.]

I think that the answer is no. GiveWell's recommendations move many millions of dollars of funds to recommended charities, and GiveWell keeps careful track of room for more funding while conveying its preferences about the quantity of funds to be received by its top charities. Extra funding for SCI would tend to push GiveWell influenced donors to other recommended charities. A large GiveWell-influenced donor could act on its own to balance out any differential effect of a donation targeted at SCI.

This problem may be even clearer with respect to funding GiveWell Labs operations: GiveWell might fill its funding gaps entirely through the support of Good Ventures, and so to a great extent donations to GiveWell could be seen as supporting "Good Ventures' marginal use of funds, whatever it may be."

Essentially, the problem is that if a charity or cause already benefits from "smart money" that would move elsewhere in response to changes in room for more funding, then marginal donations can simply displace it. This is similar to difficulties with earmarked donations to support one program within a charity: if the program was previously receiving unrestricted funds, the charity can substitute earmarked donations for those funds on a 1:1 basis.

It is possible to 'push out the GiveWell frontier' raising funding for all the recommended charities in ratios determined by the evaluations of the marginal funders and GiveWell's recommendation, but challenging to target one alone.

The impact of targeted support is determined by the new marginal donors
In the example above the marginal funder was a large centralized organization acting to keep overall spending allocations in line with its beliefs and values, which could fully compensate for any attempt to target support. But a sufficient increase in funding can overwhelm the marginal donor, shifting to a new one. With a spectrum of donors with different views about appropriate funding levels, one can derive a demand curve.

For example, suppose that there is a large population of donors who each donate $1 to one of two causes, X and Y. Both have diminishing returns to more funding, but for convenience's sake we will assume that Y's marginal returns shift negligibly in the example (perhaps it is a mature cash transfer program, with almost limitless room for more funding).

Donors differ in their views of how much funding X needs before its marginal returns equalize with Y. Thus, the donors who most favor funding for X donate, followed by those slightly less enthusiastic, until we get to the last donor, whose donation brings X to the funding level she prefers, e.g. $1,000,000.

Now suppose some new and extreme X-enthusiasts come along: they are convinced that X ought to have $10,000,000 in funding, so they collectively donate $1,000 to X. Now the marginal donor finds that X is getting $1,000 more, so she shifts her $1 donation to Y. The next donor will do the same unless she thinks X needed $999 more, and so on until a donor finds the current level of donation acceptable and remains with X.

The enthusiasts' $1,000 might deliver anywhere between nothing and $1,000 depending on the shape of the distribution of views around the marginal donor, but the extent of thee enthusiasts views doesn't matter above a minimal threshold: their impact will be the same whether they think funding for X should go up by 0.1%, 100% or 1,000,000%.

The effective altruism movement is working to increase sensitivity to marginal returns
Large philanthropies like the Bill and Melinda Gates Foundation work to pluck low-hanging fruit, and comparatively uncrowded areas. Increasing funding to a cause can easily reduce foundation support. Researchers and people choosing careers take crowdedness into account to some degree.

But the effective altruist movement is working in a number of ways to make people even more sensitive. GiveWell's recommended charities provide one of the clearest cases, but not the only one. 80,000 hours and others giving career advice take room for more talent into consideration in their recommendations. Writers and speakers discussing effective altruism seek out what they think of as neglected topics. Widespread embrace of cause-neutrality and overlap in interest in different causes makes it likely there will continue to be substantial pools of marginal resources shifting between causes in response to changes in marginal returns.

As a result, focusing on a particular cause or organization that has substantially caught the eye of the effective altruist movement may wind up being similar to targeting support for one of GiveWell's charities. The effective altruist movement's causes would be advanced, in relative proportions determined by the marginal actors, but attempts to focus on a cause favored by idiosyncratic views may matter a lot less than it would naively seem to.

10 comments:

Brian Tomasik said...

This is an important point! If this dynamic were happening, supporters of a cause might instead aim to shift the views or preferences of these cross-cause donors, although there might be a lot of them that would have to be influenced before anything moved. Or the cause's supporters might take their money out of the donation pool and use it for things off the radar to the rest of the giving community.

If the cause is sufficiently idiosyncratic that mainstream EA donors wouldn't give to it, this situation won't apply. Of course, the situation you describe is a good problem to have, so causes should still aim to make themselves more widely accepted in order to at least hit diminishing returns -- and benefit from the rising tides of total EA funding.

In some ways this could help reduce competition among causes: Once everybody is on the same frontier, everybody may just aim to maximize donation pie, unless they pursue a donor-persuasion strategy as mentioned above.

Owen Cotton-Barratt said...

I think it would be useful to think of elasticities here. The thought and effort behind the smart money certainly increase the degree to which funding one area is fungible with another (within the fairly tight set where the smart money is already going), but stickiness in the system may well be a larger factor than the difference in view between consecutive marginal donors.

If I donate a $100 to SCI, my (rather plucked from the air) guess is that will increase the amount SCI end up with relative to Give Directly by somewhere in the region of $50-$80.

I guess you might expect more elasticity than this if you expect that the quality of our critical analyses is just going to keep increasing, but I think it's also quite likely the case that one of these charities won't be at the top of GiveWell's recommendations at all in a year or two (because of new considerations rather than just because they run out of room for funding), in which case there is likely to be less elasticity than this.

Carl said...

Owen, yes this is talking about the elasticity of donations, and there are other factors.

Jürgen said...

GiveWell and Good Ventures actively wish to avoid a situation in which a donation to one recommended charity is effectively a donation to all recommended charities in the proportions we prefer. We prize what we call "donor agency": the ability of each donor to make a decision about what they are supporting, effectively as well as directly. As such,

-GiveWell has committed to grant out the "funds available for regranting" according to a set formula that is independent of what other donors do. (We haven't done so publicly but have privately emailed multiple major donors with this formula such that it can be verified that we committed to it).

-Good Ventures announced its giving plans shortly after the announcement of our top charities, without waiting to see what other donors would do; it explicitly discussed donor agency in its announcement and expressed plans not to make further donations this giving season.

-We generally discourage major donors from "funging" other donors, and we believe that they rarely do so.

I think Carl's broader point holds in many cases. Donating to a charity may reduce its fundraising efforts and may reduce its perceived need (though speeding its ability to plan and reducing the amount of effort it needs to put into fundraising have effects as well). However, I think his specific claim that "It's hard to support just one of GiveWell's recommended charities" is false because we have purposefully taken steps to make it false.

Carl said...

Thanks for sharing that Jurgen.

Carl said...

Is this true on a year-to year basis though, or just within a single giving season?

We do see GiveWell dropping AMF from its recommendations as temporarily overfunded, so the basic dynamic seems to be in play.

Holden Karnofsky said...

Sorry, I didn't realize I was posting as my old blogger account, which is called Jurgen.

The dynamic I described holds only within giving season. The goal is to avoid situations in which a donor's choice between charities ends up being "funged" by processes that are opaque to them. It is certainly true that donating to a charity brings it closer to closing its funding gap, which eventually means donations that would have gone to this charity go elsewhere. However, over that kind of time frame things can change quite a bit, and earlier donations are very meaningfully different from later ones.

Brian Tomasik said...

Carl, your post mentioned a "demand curve," and this inspired me to try drawing out what's happening here in this figure. I'm not sure if I got it right.

The demand curve represents how much value the donors place on the charity's work. The enthusiastic donors are willing to pay lots of money for one unit of output from the charity, and enthusiasm decreases as you go down the demand curve. This figure also assumes increasing marginal cost by the charity (in the short run) to produce a given amount of good, reflecting diminished room for more funding.

In the extreme example of no room for more funding, the marginal-cost curve would be a vertical line, and the funging would be complete. Since in general the marginal-cost curve shouldn't be vertical, this suggests that enthusiastic donors do make some difference (depending on elasticities) between 0% and 100% of the naive difference.

It seems like this model doesn't require extremely smart money but might apply in general, even for uncoordinated donors who just naively evaluate whether a charity is doing enough good work to justify a donation.

Anonymous said...

Hi Brian

Here's what I had in mind after reading Carl's post.

The linear curve simply shows the fact that each donor is able to donate 1$ either to X or to Y, so the funding X receives is proportional to the number of donors that donate to X. The decreasing curve shows the number of potential donors who think that X should receive at least a certain amount ("before its marginal returns equalize with Y"). The intersection of the two curves is what X actually receives.

Now new X-enthusiasts come along who think that X should receive much more funding than it currently does. The demand curve thus shifts upwards by an amount that is equal to the number of enthusiasts. Naïvely, one would expect that 1000 new X-enthusiasts who each donate 1$ to X increase the amount X receives by 1000$. However, the new amount X receives is given by the intersection with the new demand curve. As can readily seen from the figure, the increase in funding for X is less than the number of new enthusiasts.

If one marginal dollar to X leads to q previous donors shifting their donation to Y, then to first order the actual increase in donations to X is given by 1000$/(1+q). ("The enthusiasts' $1,000 might deliver anywhere between nothing and $1,000 depending on the shape of the distribution of views around the marginal donor")

Anonymous said...

I'm the second or third largest funder of a charity which is too controversial to get any of these big multi-charity or foundation grants, *and* has been consistently underfunded.

So. You want impact? Do that.