Donor lotteries: demonstration and FAQ

Sup­pose that Alice is try­ing to figure out how to do the most good with her dona­tion of $1,000 this giv­ing sea­son, and can spend var­i­ous kinds of re­sources to im­prove her de­ci­sion. Un­for­tu­nately, many in­vest­ments that could im­prove the de­ci­sion qual­ity would im­pose costs that are large rel­a­tive to her dona­tion: spend­ing hun­dreds of hours (whether her own, those of char­ity staff, or of hired eval­u­a­tors) in­ves­ti­gat­ing op­por­tu­ni­ties will cost more than her dona­tion amount. She will also be limited in the pro­jects she can fund: whereas a large fun­der can at­tract pro­pos­als for new pro­jects, and fund a new po­si­tion or startup or­ga­ni­za­tion, she seems to be limited to con­tribut­ing to ex­ist­ing pub­lic op­por­tu­ni­ties.

One solu­tion to this prob­lem would be for Alice to work with Bob, a large donor, to con­struct a ‘donor lot­tery.’ Alice donates her $1,000 to Bob’s donor-ad­vised fund, or DAF. Then Alice and Bob con­sult a ran­dom num­ber gen­er­a­tor to de­ter­mine how to recom­mend dona­tion al­lo­ca­tions to the DAF. For ex­am­ple, they might plan that with 1100 prob­a­bil­ity Alice would get to recom­mend the al­lo­ca­tion of $100,000 from the DAF, while with 99100 prob­a­bil­ity Bob al­lo­cates the DAF with­out in­put from Alice.

In the event Alice ‘wins’ the donor lot­tery, then she can pur­sue a num­ber of costly op­tions to im­prove her de­ci­sion qual­ity, offset against a 100x larger dona­tion, while the rest of the time she can avoid re­search costs. Effec­tively, the donor lot­tery cuts her re­search costs by 99% while open­ing up ad­di­tional op­por­tu­ni­ties. If $100,000 is not enough for some op­por­tu­ni­ties (e.g. if $1,000,000 is needed to fund a new startup or­ga­ni­za­tion) she could re­peat the pro­cess with still-larger donor Carol.

I have pre­vi­ously pro­posed that small EA donors who wish to cap­ture economies of scale con­sider this method. On the whole, I think that it would be bet­ter for many donors giv­ing less than $100,000 per an­num to use donor lot­ter­ies to con­soli­date into a smaller num­ber of medium-large donors and cap­ture economies of scale.

How­ever, in­so­far as sig­nifi­cant effort is re­quired to ac­cess a donor lot­tery (which must be made re­gard­less of the out­come), and it rep­re­sents a novel un­tried method, it be­comes less at­trac­tive. This post an­nounces a con­crete im­ple­men­ta­tion of the donor lot­tery con­cept, thanks to effec­tive al­tru­ist Paul Chris­ti­ano, who has agreed to back it with his DAF ac­count at Van­guard Char­i­ta­ble, and the first use cases, in hopes of work­ing out bugs and es­tab­lish­ing a prece­dent for oth­ers.

The re­main­der of the post will cover fre­quently asked ques­tions and im­ple­men­ta­tion de­tails for other effec­tive al­tru­ists who may wish to use or provide donor lot­ter­ies.

Donor lot­tery announcement

[ETA: fi­nal­ized pre-draw de­tails are in Paul Chris­ti­ano’s Jan­uary 11th (GMT) post. The draw for this first lot­tery has now been com­pleted, with Tim-Tel­leen Law­ton win­ning, al­though there will be other fu­ture lot­ter­ies, hope­fully with mul­ti­ple providers.]

The donor lot­tery will be ran­dom­ized us­ing the first 10 hex­adec­i­mal digits of the NIST pub­lic ran­dom­ness bea­con out­put, with the re­sult at 12:00 pm PST, Jan­uary 15, 2017. This will provide plenty of time for peo­ple to con­tribute in ei­ther the 2016 or 2017 tax year (tax de­duc­tions ap­ply for the year of dona­tion to the DAF).

Par­ti­ci­pants will trans­fer funds to Paul Chris­ti­ano’s DAF prior to the draw, with each $1,000 pur­chas­ing 1% of the pos­si­ble val­ues of the ran­dom­ness bea­con. If one of a con­trib­u­tor’s num­bers are drawn Paul in­tends to recom­mend al­lo­ca­tion of $99,000 from his DAF at the win­ner’s di­rec­tion to qual­ify­ing non­prof­its. The differ­ence from $100,000, a 1% ‘hair­cut,’ would be recom­mended by Paul to a char­ity of his choice as com­pen­sa­tion for offer­ing the lot­tery from his char­i­ta­ble funds.

How­ever, as is gen­er­ally the case with DAFs, and in similar struc­tures such as the Long Now Foun­da­tion’s Long Bets, Van­guard Char­i­ta­ble is legally free not to fol­low Paul’s recom­men­da­tion (al­though such a case would be quite ex­cep­tional), and no bind­ing con­tract ex­ists. The par­ti­ci­pants thus far are fa­mil­iar with Paul and one an­other in many cases, and the lot­tery re­lies on the honor sys­tem and the im­plau­si­bil­ity of ac­cept­ing the rep­u­ta­tional costs of visi­ble failure to fol­low through on a pub­lic com­mit­ment in or­der to recom­mend differ­ent char­i­ties to the DAF.

The DAF has a min­i­mum con­tri­bu­tion size of $5,000, so those donat­ing less than that will be co­or­di­nat­ing to donate in groups, but with num­bers as­signed sep­a­rately. [ETA: dona­tions un­der $5,000 can now be ac­cepted through dona­tion swap­ping with Paul, no pool­ing nec­es­sary.]

[ETA: per Paul’s fol­low-up post, the draw num­bers have been rescaled to en­sure a first-round win­ner (for the to­tal donated to the lot­tery), with the win­ner then hav­ing the choice of plac­ing a sec­ond bet to reach $100k]

Con­trib­u­tor Amount ($) Low # (in dec­i­mal) High # (in dec­i­mal) Prob­a­bil­ity
Ti­mothy Tel­leen-Law­ton 5050 0 121632721144.5 11%
Gre­gory Lewis 5000 121632721144.5 242061157921.5 11%
Ajeya Co­tra 2200 242061157921.5 295049670103.5 5%
Ro­hin Shah 2800 295049670103.5 362489594699.5 6%
He­len Toner 2500 362489594699.5 422703813087.5 5%
Ni­cole Ross 500 422703813087.5 434746656765.5 1%
Howie Lem­pel 5000 434746656765.5 555175093542.5 11%
Re­becca Raible 2000 555175093542.5 603346468253.5 4%
Pablo Staffor­ini 2000 603346468253.5 651517842964.5 4%
Aaron Gertler 500 651517842964.5 663560686641.5 1%
Bray­den McLean 5000 663560686641.5 783989123418.5 11%
Ben­jamin Hoff­man 100 783989123418.5 786397692154.5 0.2%
Cather­ine Ols­son 500 786397692154.5 798440535832.5 1%
Eric Her­boso 500 798440535832.5 810483379509.5 1%
Ian David Moss 2500 810483379509.5 870697597898.5 5%
Glenn Willen 500 870697597898.5 882740441576.5 1%
Ja­cob Stein­hardt 4000 882740441576.5 979083190997.5 9%
Bran­don Rein­hart 5000 979083190997.5 1099511627775 11%

[ETA: added Howie Lem­pel De­cem­ber 7th, 2016; Re­becca Raible and Pablo Staffor­ini De­cem­ber 12th; Aaron Gertler De­cem­ber 16th; Bray­den McLean and Ben­jamin Hoff­man De­cem­ber 24th; Cather­ine Ols­son Dec. 26th; Eric Her­boso and Ian David Moss Dec. 31st; Glenn Willen Jan. 2; Ja­cob Stein­hardt Jan. 4th; Bran­don Rein­hart Jan 8th.]

Others who wish to par­ti­ci­pate, please con­tact Paul F Chris­ti­ano (at first­namemid­dleini­tial­last­name at gmail) for de­tails re­gard­ing funds trans­fer to Van­guard Char­i­ta­ble, etc.

I would also hope that these early adopters might in­spire other donor lot­ter­ies, both other providers and users.

Q: Step-by-step, how do I make use of a donor lot­tery as a small donor?


[ETA: one les­son from the first lot­tery was that the min­i­mum dona­tion size for DAF con­tri­bu­tion was a has­sle, and dona­tion swap­ping (dona­tion to a char­ity the lot­tery spon­sor was oth­er­wise donat­ing to, so they can re­duce their dona­tion to it ac­cord­ingly and recom­mend chan­nel­ing the funds to the win­ner’s recom­men­da­tion) was more effi­cient.]

  1. Find a larger donor with a DAF will­ing to provide a donor lot­tery, such as Paul Chris­ti­ano.

  2. Deter­mine the dona­tion amount you would like a chance at, and agree to terms, e.g. “I donate $1,000 to this DAF in ex­change for a 1% chance at al­lo­cat­ing $99,000 from it.”

  3. Jointly an­nounce the terms, and the pub­lic ran­dom­ness source that will be used to de­ter­mine the out­come, e.g. the NIST bea­con, as dis­cussed be­low.

  4. En­sure copies of the an­nounce­ment are se­cured, e.g. by tweet­ing a hash of the post.

  5. Donate to the DAF.

  6. The pub­lic ran­dom­ness source pro­vides its re­sult.

  7. If you ‘won’ the ran­dom­iza­tion, take steps to gather in­for­ma­tion or pur­sue dona­tion op­por­tu­ni­ties that make sense as a large donor.

    1. Post about your de­ci­sion-mak­ing pro­cess and in­vite feed­back and in­for­ma­tion about key points.

    2. Talk to char­i­ties and ex­perts who can af­ford to spend more time ad­vis­ing you on the larger dona­tion.

    3. Call for pro­pos­als, or re­quest them from par­tic­u­lar or­ga­ni­za­tions and ex­perts, for ‘chunky’ grants or pro­j­jects with min­i­mum scale.

    4. Spend more time think­ing about and study­ing the op­tions.

    5. Hire re­search sup­port, au­di­tors, and other means of eval­u­a­tion.

  8. Make your recom­men­da­tions to the DAF.

  9. Your large dona­tions go through.

Q: This idea seems novel. Is there an es­tab­lished prece­dent?

A: The Long Now Foun­da­tion has op­er­ated a closely re­lated ser­vice, Long Bets, since 2002. This al­lows two peo­ple to make bets on fu­ture events, with the stakes go­ing to the win­ner’s char­ity of choice. The Long Bets rules page states:

There is no max­i­mum amount. The bet money, treated as a dona­tion to the The Long Now Foun­da­tion, must be paid at the time the Bet is made, and is tax-de­ductible im­me­di­ately...

The char­ity des­ig­nated should be IRS-ap­proved if in the US. For­eign char­i­ties also are el­i­gible. The Long Now Foun­da­tion can­not legally guaran­tee that the win­nings will go to the win­ner’s preferred char­ity. It is highly mo­ti­vated to do so, of course, since how a bet is re­solved is in the pub­lic eye.

This pro­vides an es­tab­lished prece­dent of dona­tions con­di­tioned on un­cer­tain events. Like this, and like DAFs, a donor lot­tery would not in­volve a legally bind­ing con­tract, but such mechanisms can be quite re­li­able in prac­tice.

Q: How can I be con­fi­dent that the large dona­tion will be forth­com­ing if I ‘win’?

A: The terms of the donor lot­tery can be an­nounced pub­li­cly, com­mit­ting the back­ing of rep­u­ta­tion. Money donated to a DAF can only be passed on to legally rec­og­nized char­i­ties. Us­ing a trusted donor’s DAF, and one for whom the amount is rel­a­tively small (so that rep­u­ta­tion mat­ters more) would also provide ad­di­tional as­surances. The an­nounce­ments should be copied to se­cure and un­change­able pub­lic stor­age (some par­ti­ci­pants in this lot­tery will be tweet­ing hashes).

Ul­ti­mately, how­ever, this setup re­lies on the honor sys­tem and rep­u­ta­tion.

Q: Where can we find a trusted pub­lic ran­dom num­ber gen­er­a­tor?

A: The Na­tional In­sti­tute Stan­dards and Tech­nol­ogy in the United States offers a pub­lic ran­dom­ness bea­con, which “gen­er­ates full-en­tropy bit-strings and posts them in blocks of 512 bits ev­ery 60 sec­onds. Each such value is se­quence-num­bered, time-stamped and signed, and in­cludes the hash of the pre­vi­ous value to chain the se­quence of val­ues to­gether and pre­vent even the source to retroac­tively change an out­put pack­age with­out be­ing de­tected.”

To ran­dom­ize, the par­ties can pub­li­cly state a par­tic­u­lar minute (e.g. 5:04 pm, De­cem­ber 8th, 2016) for the de­ter­min­ing out­put value. Then one can take the first 10 hex­adec­i­mal digits (which can show num­bers from 0 to 1,099,511,627,775) and use them to cash out a prob­a­bil­ity. For ex­am­ple, a 1% chance could be im­ple­mented as “the first 10 digits lie in the range from 0 to 10,995,116,116,278.”

Other sources, from weather to mar­ket move­ments, are abun­dant.

Q: If I be­lieve I would do more good per dol­lar with a smaller chance of a larger dona­tion, why not just gam­ble or make risky in­vest­ments?

A: In prin­ci­ple, as dis­cussed in my blog posts on donor lot­ter­ies, these meth­ods could also be used by small donors to take ad­van­tage of economies of scale in ex­pec­ta­tion. The ad­van­tage of the donor lot­tery via DAF is just in re­duc­ing over­head and trans­ac­tion costs.

Q: Does it mat­ter what other donor lot­tery par­ti­ci­pants would do with their funds?

A: Sup­pose you are one of 10 peo­ple each con­tribute $10,000 to a pool, and one ran­domly gets to al­lo­cate all $100,000. Should you care what char­i­ties the other par­ti­ci­pants would recom­mend? On av­er­age, you are not trans­fer­ring any net dol­lars to the char­i­ties recom­mended by the other par­ti­ci­pants (90% of the time you pay out $10,000 to one of those char­i­ties, and 10% of the time $90,000 gets di­rected to your prefer­ence in­stead). How­ever, the rea­son to par­ti­ci­pate in the first place would be if marginal re­turns to dona­tions to a char­ity vary with the amount, and donors hope they are in­creas­ing.

If other donors would sup­port char­i­ties which one thought were ac­tively bad (even with the chance to think more care­fully in cases where they ‘win’ the donor lot­tery), and do so more effec­tively per dol­lar with larger dona­tions, then one could do harm by helping harm­ful dona­tions.

For char­i­ties with pos­i­tive or neu­tral effects, the worry would be that par­ti­ci­pants are mis­taken about in­creas­ing re­turns to dona­tion size. For ex­am­ple, a donor who gives $10,000 to an or­ga­ni­za­tion with a bud­get of $50,000 might well ex­pect that the or­ga­ni­za­tion would much pre­fer $10,000 with cer­tainty over a 10% chance of $100,000. Such diminish­ing re­turns can give rea­son for risk-aver­sion. If one is fo­cused on a small cause or or­ga­ni­za­tion, where diminish­ing re­turns mat­ter over small ranges, this is rea­son to worry about the chance of the area fac­ing a fi­nan­cial short­fall due to chance lot­tery out­comes. One way to ad­dress that is by con­duct­ing a lot­tery among sup­port­ers.

Q: How can the lot­tery provider offset the risk their DAF takes?

A: While for dona­tion amounts that are small rel­a­tive to fund­ing for one’s causes, there is a good case for al­tru­ists be­ing ap­prox­i­mately risk-neu­tral (e.g. for small donors), diminish­ing re­turns mean that for large pools of money there is some cost to tak­ing on risk for donors who are large rel­a­tive to the causes they work on. So a lot­tery provider could ar­range a ‘hair­cut,’ e.g. Alice donates $1,000 to Bob’s DAF, and with 1% prob­a­bil­ity gets to al­lo­cate $99,000, rather than $100,000. The differ­ence serves to com­pen­sate for the risk that Bob’s DAF will be de­pleted of cash to fund char­i­ta­ble op­por­tu­ni­ties he pri­ori­tizes.

The op­ti­mal ‘hair­cut’ would vary de­pend­ing on al­ign­ment be­tween Alice and Bob, Bob’s bankroll, diminish­ing re­turns in the causes Bob sup­ports (rel­a­tive to the size of his dona­tions), etc.

Q: How large an amount should I con­sider us­ing a lot­tery to reach? Does this im­ply all EA dona­tions should be merged?

A: There are both helpful economies of scale and harm­ful dis­ec­onomies of scale. For ex­am­ple, if all effec­tive al­tru­ist dona­tions were con­soli­dated in a sin­gle gi­ant foun­da­tion, then they would all be at risk of mishap to­gether, e.g. if a catas­tro­phe befell it. If one dona­tion risked at­tract­ing con­tro­versy or strain­ing man­age­ment bot­tle­necks for the whole pool, the cost would be much greater than for sev­eral smaller pools. A sin­gle pool would also suffer from re­duced di­ver­sity of ap­proaches, and could miss out on lo­cal knowl­edge of par­tic­u­lar op­por­tu­ni­ties (which should give donors rea­son to think that they have diminish­ing re­turns to larger dona­tions, not in­creas­ing re­turns).

A donor lot­tery ex­e­cuted through a DAF also re­quires that dona­tions go to reg­istered char­i­ties, ex­clud­ing such things as poli­ti­cal con­tri­bu­tions and in­for­mal help to par­tic­u­lar in­di­vi­d­u­als (the lat­ter be­ing a case where lo­cal knowl­edge is es­pe­cially im­por­tant).

I would sug­gest that small donors con­sider us­ing lot­ter­ies to move into the range of $50,000-$500,000. This is most at­trac­tive if one would oth­er­wise be donat­ing to or­ga­ni­za­tions or ar­eas where one’s funds are small rel­a­tive to to­tal fund­ing, and fun­gible with other donors, but have hope of iden­ti­fy­ing or cre­at­ing bet­ter op­por­tu­ni­ties as a larger donor. It is least at­trac­tive when one one would oth­er­wise be a large com­po­nent of a small fund­ing pool, es­pe­cially ex­ploit­ing lo­cal knowl­edge, so that one ex­pects diminish­ing re­turns over the rele­vant range.

Ad­di­tional lot­tery stages to reach quan­tities in the mil­lions or more face high risk of miss­ing low-hang­ing fruit and suffer­ing from diminish­ing re­turns. How­ever, they might make sense in some cir­cum­stances if one ex­pects strong ne­glected op­por­tu­ni­ties in the cre­ation of new or­ga­ni­za­tions or other foun­da­tion-like grants.

Q: What if my tar­get dona­tion amount is too low or high for a known donor lot­tery?

A: For the small­est dona­tions, e.g. $10 or $100, I would sug­gest an in­for­mal ran­dom­iza­tion with a friend to start. If that ran­dom­iza­tion de­liv­ers a larger dona­tion, then it could be fun­neled through a more sub­stan­tial donor lot­tery.

If a friend is donat­ing to a char­ity with bud­get that is large rel­a­tive to their dona­tion (e.g. GiveDirectly or AMF) they can eas­ily offer a sim­ple honor sys­tem donor lot­tery with­out much worry about diminish­ing re­turns.

If you would like to have a chance at a quite large dona­tion, e.g. mil­lions of dol­lars, I would recom­mend a multi-stage pro­cess. First reach a dona­tion to­tal in the hun­dreds of thou­sands, ei­ther di­rectly or via a donor lot­tery. At that point, try to be con­nected with much larger donors

Q: What if I want to de­lay my dona­tion over time af­ter win­ning?

A: Some of the op­por­tu­ni­ties for im­proved de­ci­sion-mak­ing as a larger donor take time. If funds will be sit­ting in the DAF for some time, com­pound­ing due to in­vest­ment re­turns, you may want to make an ar­range­ment with the donor pro­vid­ing the lot­tery as to the treat­ment of in­vest­ment re­turns, e.g. the stakes grow­ing in line with re­turns for the DAF as a whole.

Q: Are there any in­ter­est­ing al­ter­na­tives to a ran­dom num­ber gen­er­a­tor?

A: In­stead of us­ing the re­sults of a pub­lic ran­dom num­ber gen­er­a­tor, one could in­stead ‘bet’ (in the style of Long Bets, but with less of a hair­cut) on pre­dic­tions that are rele­vant to char­i­ta­ble al­lo­ca­tion, or the rel­a­tive pre­dic­tive skill of the par­ties. For in­stance, a pay­out might oc­cur con­di­tional on a par­tic­u­lar elec­tion re­sult, or sci­en­tific bench­mark. This could be used to in­sure against spe­cial needs (e.g. per­haps you care more about dona­tions if cer­tain tech­nolo­gies are de­vel­oped, or if cer­tain prob­lems be­come more se­vere), and to make it more likely that bet­ter de­ci­sion-mak­ers al­lo­cate the dona­tions.

Q: Why not in­stead al­lo­cate the dona­tions of a group of small donors us­ing ran­dom­iza­tion?

A: Another way to cap­ture some of these benefits is to team up with a few friends donat­ing on the same scale, and use ran­dom­iza­tion (weighted by dona­tion amount) to pick one of you to al­lo­cate the dona­tions of the group (us­ing the honor sys­tem). It may be sig­nifi­cantly more difficult to buy chances of large dona­tions, but this is a use­ful al­ter­na­tive.

Q: Won’t donor lot­ter­ies re­duce the amount of prac­tice peo­ple get think­ing about char­ity?

A: Think­ing care­fully about one’s dona­tions can be a helpful ed­u­ca­tional ex­er­cise, and build im­por­tant skills for a would-be effec­tive al­tru­ist, as well as par­ti­ci­pate in var­i­ous so­cial pro­cesses. Giv­ing Games let peo­ple al­lo­cate money as a way to in­crease their in­ter­est in donat­ing effec­tively and in in­ves­ti­gat­ing their op­tions. This may be a rea­son not to al­lo­cate one’s en­tire dona­tion bud­get to a lot­tery, and to donate a smaller amount to one’s best guess, for prac­tice, while other funds go into a lot­tery for higher ex­pected qual­ity.

If the in­di­rect benefits such as prac­tice and sig­nal­ing are suffi­ciently great, and economies of scale suffi­ciently low, this could be a rea­son to avoid donor lot­ter­ies en­tirely.

Q: My ques­tion isn’t an­swered above, or I would like to par­ti­ci­pate in match­mak­ing for a donor lot­tery.

A: Please con­tact Paul F Chris­ti­ano (at first­namemid­dleini­tial­last­name at gmail) or my­self (at first­name­last­name at gmail). Public dis­cus­sions in the com­ments are also helpful.