Fewer but poorer: Benevolent partiality in prosocial preferences

Link post

A new pa­per in Judg­ment and De­ci­sion Mak­ing finds that:

  • Peo­ple of­ten choose to help peo­ple from more dis­ad­van­taged groups “even when this trans­par­ently im­plies sac­ri­fic­ing lives.”

  • How­ever, peo­ple are much more likely to make the de­ci­sion that saves the most lives pos­si­ble if they are asked to ex­plic­itly re­flect on and rank which crite­ria they should use to make.

The pa­per ex­plic­itly dis­cusses EA on the first page and the first study in­volves par­ti­ci­pants choos­ing to donate to ei­ther SCI or The END Fund, and are pre­sented with cost-effec­tive­ness in­for­ma­tion from GiveWell.

The stud­ies are all based around roughly the same ba­sic de­sign: pre­sent­ing par­ti­ci­pants a choice be­tween a less effec­tive char­ity that serves peo­ple from a more dis­ad­van­taged group or a more effec­tive char­ity that serves a more ad­van­taged group (though note that the groups in ques­tion are benefi­cia­ries in Nige­ria and Ethiopia, so both would likely be con­sid­ered highly dis­ad­van­taged rel­a­tive to the par­ti­ci­pants). The effects are re­li­ably pretty large (around 25.5% pick the less effec­tive char­ity in the con­trol con­di­tion in study 1, and over 43% in the ex­per­i­men­tal con­di­tion, where which set of benefi­cia­ries is more dis­ad­van­taged is made salient (e.g. in­for­ma­tion is pro­vided about the GDP per cap­ita and liter­acy rate of the coun­tries where the two char­i­ties work). Similarly the differ­ences are 15.8% → 42.4% in study 2A and 20.5% → 40.79% in 2B. In Study 3, over 46% picked the less effec­tive char­ity when de­mo­graphic in­for­ma­tion was made salient but they weren’t asked to re­flect be­fore­hand on what crite­ria to use, but only 23% did when asked to re­flect be­fore­hand about which crite­rion they should use to make the de­ci­sion (e.g. save the most lives per dona­tion, the av­er­age in­come of the coun­try etc.) and rank their im­por­tance.

The effect of en­courag­ing peo­ple to ex­plic­itly re­flect on what de­ci­sion pro­ce­dure to use prior to mak­ing a de­ci­sion seems of par­tic­u­lar in­ter­est to EA. One par­tic­u­lar ad­van­tage of this ap­proach is that it is non-pa­ter­nal­is­tic (i.e. we don’t as­sume a par­tic­u­lar con­clu­sion is nor­ma­tively cor­rect and try to nudge peo­ple to­wards it). It is also pre­sum­ably epistem­i­cally salu­tary (con­di­tional on the as­sump­tion that more re­flec­tion tends to be benefi­cial, which is cer­tainly open to ques­tion). Of course, there are lots of differ­ent ways that EAs could en­courage oth­ers (and them­selves) to re­flect more about their de­ci­sion-mak­ing in ad­vance (I’m think­ing pri­mar­ily about in­sti­tu­tional de­sign, even if on a very small level, but of course this oc­curs in­di­vi­d­u­ally as well), so per­haps we should think more about the best ways to do this.

Of note, when asked ex­plic­itly, about 92% of re­spon­dents ranked sav­ing the most lives per dona­tion as the most im­por­tant crite­rion, over the mark­ers of dis­ad­van­tage. That said, I sus­pect this effect would have been much less im­pres­sive had the ranked op­tions in­cluded fac­tors which made the pos­si­bil­ity of pri­ori­tis­ing the dis­ad­van­taged even more salient. I would also ex­pect the ‘dis­ad­van­tage’ effect the pa­per found to be much stronger in a lot of cases where the pull of pri­ori­tis­ing the worst off is more salient, for ex­am­ple, where more jus­tice re­lated in­tu­itions are elic­ited. Choos­ing to help par­ti­ci­pants in one Afri­can coun­try as op­posed to an­other slightly more so­cio-eco­nom­i­cally dis­ad­van­taged Afri­can coun­try, doesn’t strike me as a case that would gen­er­ate a strong in­tu­itive pull to­wards pri­ori­tis­ing more dis­ad­van­taged groups.

One caveat about the pa­per, was that when I looked at the open com­ments where re­spon­dents ex­plained why thought a par­tic­u­lar de­ci­sion crite­ria was im­por­tant (e.g. in­come), a small num­ber sug­gested that they wouldn’t want to donate to a coun­try with too high an in­come, be­cause they thought that in such cases the would-be benefi­cia­ries should be able to pay for them­selves, which isn’t re­ally in line with the ex­pla­na­tion sug­gested by the pa­per (peo­ple want­ing to help the worst off, rather than not want­ing to help peo­ple who don’t need help). Fur­ther re­search into what con­di­tions are re­quired for this effect and ex­actly what is mo­ti­vat­ing par­ti­ci­pants is re­quired.