Downloaded by guest on September 30, 2021 eiinmaking decision neglect policy. default evidence-based of and implications influence, social discuss making, We decision for defaults. where use cases in even do average, on they effect people default most antici- the in yet to believe effects, not seem default do drive People that effect. mechanisms default some inves- the pate we to Third, related defaults. beliefs use systematically tigate not feed- do still and they experience back, for when opportunities Second, multiple all. have at participants defaults (n using systematically samples not 11 tar- people across the default- with default set one-shot participants the of as in 50.8% option only First, get that neglect.” find oth- we “default influence games, show to setting peo- defaults they that use i.e., and/or find understand ers, we to the predictions, fail as theoretical often option ple target to target a contrast that selecting present In toward to default. whether others choosing nudge by We to option quo. adults) US status and the dents, is stu- law/business/medical option is managers, (including which whichever participants effect, asked choose default to the tendency relevance: policy the special stud- with predictions widely bias a these consider ied We However, attention. empirical others. exploit little received 2017) influence to have 20, how to July review understand biases for people (received common 2017 that 29, suggest September theories approved and Current NJ, Princeton, University, Princeton Fiske, T. Susan by Edited and Kong; Hong Technology, a www.pnas.org/cgi/doi/10.1073/pnas.1712757114 tradition- of members many thou- of save lives the could including change lives, simple of This HIV sands (19). provide default not by did imple- and screenings not had recommendations recom- hospitals policy most similar these 2016, mented a of as with Yet, 2007. followed Control in mendation Organization and Disease default, Health for a World Centers such the of US application the recommended 2006, Prevention and In of 20). portion (19, large agnosed a help could professionals default the by healthcare screenings having randomized HIV that provide example, demonstrate For trials underutilized. clinical remain they defaults, 14–18). (8, behavioral choices better in make interest to that people of and defaults) “nudge” (including surge makers, tactics influence a policy social (1, and been cost interventions academics, little has among at there and Accordingly, organizations, scale money, 13). large evi- lives, 12, a default As save on 8, help (7–11)]. the often can bias environment, as defaults quo the set and wise status setting or is above, effect option denced default whichever the as choose [known to tend people percentage 90 default behaviors. are friendly the that environmentally are promoting use (6), options utility higher energy points energy-efficient good green of public which rates underprovided, in have environ- but regions world, critical, Yet the a (5). Across is elderly. the protection among improv- mental vastly life (4), of points can quality percentage default ing 50 companies by by plan Yet rates savings (3). savings retirement all increase a Most at in employees nothing deaths. and enroll saved (2), that of have retirement for people thousands money of little preventing millions too (1), far saving by now higher are donors Americans points organ are percentage organ an 40 organ they as rates for than donation of people waiting to undersupply enrolling lead while could Yet severe default die year. a people every is 8,000 donation over there and States, donors, United the in D architecture choice influence social Zlatev J. at Julian attempts in neglect Default rdaeSho fBsns,Safr nvriy tnod A94305; CA Stanford, University, Stanford Business, of School Graduate oee,apzl ean:Dsieeiec ftepwrof power the of evidence Despite remains: puzzle a However, that demonstrated has research of body growing and vast A 5,0 mrcn h r I-oiiebtrmi undi- remain but HIV-positive are who Americans ∼150,000 elh nnil n niomna ucms Currently, outcomes. important on environmental impact and enormous financial, an health, have can options efault | oilinfluence social a,1 ai .Daniels P. David , c colo at,Eeg niomna cecs tnodUiest,Safr,C 94305 CA Stanford, University, Stanford Sciences, Environmental & Energy Earth, of School | eal effect default b ai Kim Hajin , | nudges ,4) consistent 2,844), = c n agrtA Neale A. Margaret and , | b colo uiesadMngmn,TeHn ogUiest fSineand Science of University Kong Hong The Management, and Business of School htsca n ontv scooit aedsrbdoe the biases over the described of have most psychologists of cognitive impact, and the social “people and that that existence, oth- proposing the psychology influence arguments, in recognize to complementary researchers Several made them (34). will have harness directions and desired optimally biases in expectations; to others’ ers rational about able influ- beliefs have be to correct strategi- they therefore hold biases who that will people those they assume thus, of exploiting nudges models use at economic cally adept Current thus them. ence and biases decision systematic others of aware in be should people almost that predict been has question fundamental this 33], overlooked. and completely refs. 32, [e.g., multidisciplinary 29, and lit- vast 24, the is while influence social Surprisingly, sta- on (31)? erature the strategically nudges them default toward use understand to biased actually enough people be strategically, do However, will option quo. choices tus default others’ a these that of present all anticipating In could pursue. their to people give goal can a situations, selecting leaders when and default read; a to followers book when a default bedtime choosing a which children when deciding their exam- default give a For can parents patients effect. treatment; imple- their default medical to give the can easy as physicians such are ple, effective, that change widely tactics to and with tactics ment how influence especially and use behavior, whether 30). can others’ (29, about anyone others known influence virtually to is Moreover, tactics little these use very sim- tac- strategically influence these, be people while to as respond However, to people such behavior. how found about tics others’ known been alter is deal have to great (28) a ways reliable reciprocity and of ple norm (27) technique the foot-in-the-door influ- the and and as such persuasion Tactics in (23–26). involved ence factors psychological the stand profes- policy? healthcare clear-cut simple, have this then implement to why failed (21); sionals groups stigmatized ally 1073/pnas.1712757114/-/DCSupplemental at online information supporting contains article This 1 aadpsto:Dt n aeil r vial at available are materials and Data deposition: Data the under Published Submission. Direct PNAS a is article This interest. the of conflict wrote no M.A.N. declare authors and The H.K., D.P.D., J.J.Z., and data; analyzed paper. D.P.D. and J.J.Z. research; uhrcnrbtos ...adDPD eindrsac;JJZ n ...performed D.P.D. and J.J.Z. research; designed D.P.D. and J.J.Z. contributions: Author owo orsodnesol eadesd mi:[email protected]. Email: addressed. be should correspondence whom To in.Teefidnssgetta h cp o improving for vast. is scope interventions the policy behavioral that via welfare suggest social to findings excep- fail These interesting some often tions. with defaults, adults use managers, US and/or relevance, that understand and policy find We we students, special Overall, others. with law/business/medical choices. bias others’ default influence influence the studied to using to widely at a noth- them successful tac- effect, are almost use these people them, whether understand respond strategically examine people on people and whether used how about tics are about known that is known ing tactics is influence deal to great a While Significance nteoehn,svrliprattertclperspectives theoretical important several hand, one the On under- to sought have researchers (22), 1950s the least at Since a NSlicense. PNAS . www.pnas.org/lookup/suppl/doi:10. https://osf.io/4efdv/ NSEryEdition Early PNAS . | f6 of 1

PSYCHOLOGICAL AND COGNITIVE SCIENCES past few decades” [ref. 35, p. 369] as a result of observational tect” (CA) who decides which of two options to present as the or instructional learning. Further, one review of the literature default option to player 2 (8). Player 2 is a “Choice Maker” (CM) on social influence suggests that “influence agents” commonly who chooses an option. The CA’s objective is to select a default succeed when trying to influence “targets” in desired directions option that will influence the CM toward choosing a randomly (24). Even if people do not fully anticipate the magnitudes of assigned target option. If the CM does choose the target option, others’ biases, they could still anticipate the directions of these the CA may receive a monetary bonus (of up to $4). CAs who biases (36, 37), which is often sufficient for using nudges optimally decide to present the target option to CMs as the default option (8). Consistent with this possibility, one recent study suggests that are considered to have made the “optimal” CA decision.† people think others are influenced toward national-level defaults All experiments were approved by Stanford University’s Insti- when deciding whether or not to be an organ donor, even though tutional Review Board, and all participants gave informed people underestimate the magnitude of this default effect (36). consent. In summary, current economic models of people who use nudges and related perspectives in psychology suggest that people should Studies 1a to 1k: Do People Use Defaults in One-Shot be good at understanding and/or using defaults in a sophisticated Games? way, suggesting the following three hypotheses. Materials and Methods. Using the default game, we asked 11 Hypothesis 1: People should be good at using defaults to influ- samples of participants (total n = 2,844) to influence others ence others in desired directions (i.e., the rate of using optimal toward selecting a target option by choosing whether to present defaults should be better than chance). the target option or a nontarget option as the default. In sam- Hypothesis 2: People should hold correct beliefs about what ples 1 to 3 (studies 1a to 1c), US adults were asked to influence defaults do (i.e., the rate of holding correct beliefs about the another person’s choice of job by setting a default job (based on a direction of the default effect should be better than chance). vignette adapted directly from ref. 53). In samples 4 and 5 (stud- ies 1d and 1e), Stanford law (JD) students and US adults were Hypothesis 3: People who hold correct beliefs about the direc- asked to imagine they were a legislator and wanted to influence tion of the default effect should be better at using defaults to consumers’ choice of insurance policy by setting a default insur- influence others in desired directions. ance policy. In samples 6 to 8 (studies 1f to 1h), Stanford busi- On the other hand, people might hold incorrect beliefs about ness (MBA) students, business executives, and US adults were others’ systematic decision biases—perhaps stemming from asked to imagine they were a CEO and wanted to influence an idiosyncratically erroneous (39) or systematically biased (40) lay employee’s choice of job by setting a default job. In samples 9 to theories about what factors influence others’ behavior.∗ As a 11 (studies 1i to 1k), Northwestern and Stanford medical (MD) result, people might be unable to optimally harness these biases students and doctors, and US adults were asked to imagine they to influence others in desired directions. While, in principle, peo- were on a hospital panel and wanted to influence a patient’s ple could adjust their incorrect beliefs by learning from past choice of medication by setting a default medication (see Sup- experiences (41), in practice, effective learning typically requires porting Information for full instructions). Five of the CA samples immediate and accurate feedback, which is unlikely to be avail- (studies 1a, 1b, 1c, 1i, and 1k) made incentivized decisions involv- able in most real-world environments (42). Even with immedi- ing real CMs, while the other six CA samples (studies 1d, 1e, 1f, ate and accurate feedback, however, additional opportunities for 1g, 1h, and 1j) made hypothetical decisions. learning can still fail to improve patterns of suboptimal decision making (43, 44). In noisy environments, learning about system- Results. Across studies 1a to 1k, we find evidence that people atic causal relationships—such as the default effect—is hard failed to use default nudges optimally. Results indicated an over- (45), despite its importance for good decision making in inter- all CA decision pattern consistent with random chance (i.e., active situations (43, 46, 47). Consistent with this view, Johnson consistent with CAs not using defaults at all), with a narrow and Goldstein (48) conjecture that people may hold incorrect 95% confidence interval that included 50% (fixed effects model: beliefs about the default effect. In summary, it is possible that peo- 50.8%, 95% CI [49.0%, 52.7%]; random effects model: 51.9%, ple may show “default neglect” (48)—i.e., they may fail to under- 95% CI [47.4%, 56.5%]). These results were thus inconsistent stand and/or use defaults to effectively influence others, which with hypothesis 1. To investigate potential heterogeneity across would be inconsistent with some or all of hypotheses 1 to 3. our studies, we split the samples into two groups; the “non- In addition to considering the aforementioned theoretical pre- professional samples” included online adult studies and non- dictions, we also asked experts in the field of judgment and professional university affiliates, while the “professional sam- decision making to predict whether people would be good at using ples” included executives, business students, medical students, the default effect to influence others. Interestingly, the predic- and law students. The meta-analysis of the nonprofessional sam- tions of these experts were in line with the predictions of current ples (k = 7, n = 2,187) indicated an overall CA decision pattern economic models of people who use nudges. In an email survey of consistent with random chance, with a 95% confidence interval members of the Society for Judgment and Decision Making (n = that included 50% (fixed effects model: 48.3%, 95% CI [46.2%, 133), the overwhelming majority of experts—90.1%—predicted 50.4%]; random effects model: 48.4%, 95% CI [44.8%, 52.0%]). that people would successfully use defaults to influence others The meta-analysis of the professional samples (k = 4, n = 657) in desired directions. Only 2.3% of experts predicted that peo- indicated an overall CA decision pattern that was significantly ple would fail to use defaults altogether. These predictions may better than chance, with a 95% confidence interval that did help explain the dearth of existing research on default neglect. not include 50% (fixed effects model: 59.8%, 95% CI [55.9%, To experimentally investigate whether people are able to 63.5%]; random effects model: 59.0%, 95% CI [50.2%, 67.3%]). strategically use default nudges, we developed a “default game” See Fig. 1 for the meta-analytic results for each group. Thus, involving two players, inspired by similar interpersonal and the professionals used default nudges at a rate that was better intrapersonal influence games used in previous research [e.g., than chance and better than that of the nonprofessional sam- refs. 49–51]. In the default game, player 1 is a “Choice Archi- ples, although they were still closer to random chance than to optimality. Differences in CA behavior across populations could

∗By incorrect belief, we mean either (i) believing that there is no default effect, (ii) believing that there is an antidefault effect, or (iii) having no belief about the default †Random assignment of the target option lets us evaluate how good people are at using effect but making decisions as if holding one of the two aforementioned incorrect defaults to influence others independent of idiosyncratic preferences regarding the beliefs [e.g., a CA who has no beliefs about the default effect might set defaults ran- individual options (see experiment 1 in ref. 52, which asked University of California, domly, perhaps appealing to the principle of insufficient reason (38)]. In this paper, we San Diego undergraduates to set a default for hypothetical others but did not randomly do not focus on the third type of incorrect belief mentioned here. assign a target option).

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1712757114 Zlatev et al. Downloaded by guest on September 30, 2021 Downloaded by guest on September 30, 2021 ewe opeeso hc ucmsadC performance correlations CA significant and outcomes (see no check were comprehension before There pay- between questions the decisions. these and their rounds, completed making of Participants number structure. the ment options, CM the regarding Thus, option. target the to chose decisions. up who earn CM could a not CAs with was paired it when were than a they default earned the These CAs was likely default.) 1. it more the when points the study option percentage an 25.2 in used choose were as to CMs (We percentage the CAs 25.2 is, of not. with That effect points. treatment match or default to a CM option demonstrated CMs the CMs target whether of told the sample were same chosen CAs then, actually that choice CA); default that had the the within by with choice selected of environment job was choice response a the real made (i.e., previously the each environment had with After who option. matched having CM default randomly a the job round, were as each set CAs one In randomly rounds. to round, was 20 and option all jobs which for pay two chose option the jobs CAs target more of the two be one having of to CA, choice assigned job each a For one present vacation. participants to more with Specifically, how CMs, row. decided a and to in CAs times as 20 acted 1a study from game Methods. and Materials feedback. and experience neglect to default robust the is decisions whether CA test in to observed to us we allowed forming robust repeated game row, The this are (54). a of game nature in default 1k times one-shot” 20 “repeated to each game 20-round which a default 1a in one-shot study studies the follow-up played CM. a of CA ran a results CAs we to feedback, which present the and experience to in whether default games which test default about To decision “one-shot” single used a 1k made to 1a One-Shot Studies Repeated a in Defaults Game? Use People Do 2: Study in differences including motivation. factors, or of practice, education, number a reflect potentially shows bar left students, The MBA students). students, MD (i.e., JD SEs. and managers, bar samples (i.e., right are samples The nonprofessional professional affiliates). all all bars university includes nonprofessional includes and Error studies bar adult studies. middle online The 11 results. all overall across nudges 1. Fig. ltve al. et Zlatev eas nlddsvncmrhninceksre questions survey check comprehension seven included also We uprigInformation Supporting 100% 25% 50% 75% 0% td :mt-nlsso ecnaeo A sn pia default optimal using CAs of percentage of meta-analysis 1: Study Percentage ofPeople UsingOptimalDefault vrl Non−Professional Overall ntpo hi aepybsdo their on based pay base their of top on $5 atcpnspae h n-htdefault one-shot the played Participants .5bnsfrec on nwhich in round each for bonus $0.25 o details). for Samples Professional Samples pial,wihwssgicnl os hnrno chance nudges random default (all than used rounds CAs worse 20 of significantly [χ the 33% was round), of which first one (the optimally, random any 1 from round in In different chance pos- significantly than maximum ter not a of was (z (out chance which times 20), 9.86 sible optimally options default Results. fCsue eal ugsotmly hc a significantly was which (z optimally, behavior 1 nudges 54% round round), than default final better (the used 20 round CAs In 1. of study in decisions CA the of on eetdoeso eal ae ro asaeSEs. are bars Error game. default one-shot repeated round 2. Fig. 1.8% by signif- selection round nudge additional optimal each we of 10, participant, rate [t to the by 1 increased clustered rounds icantly sep- errors across halves that, standard these find of regres- robust (OLS) each to with 1 least-squares analyzed rounds sion ordinary and into using 20 data Specifically, the to sec- split arately. 11 the hoc rounds versus post and half we study, 10 first difference the the a of be in half to ond played appeared games there learning) one-shot Because of between off. amount first leveled modest selection then nudge a but optimal (suggesting of somewhat rate the increased time, over game default [χ chance random from different icantly .. hi rdcin bu o hnigadfutwl change will default a changing how about predictions their i.e., defaults. use to fail often CAs many why able on be light might additional effects, shed default to understanding or drive of their to seem and (mis)believe that investigation mechanisms do, the An of defaults people judg- what effect? about do in beliefs default experts people’s What the and making. about nudges (mis)understand decision use 1k who eco- and to current people both 1a ment of of studies predictions models both the to across nomic opposition defaults in 2, optimal study set and to failed CAs the About Beliefs Effect? Accurate Default Hold People Do 3: Study feedback. 1. in hypothesis and with neglect experience inconsistent default thus with that were even results find persist, These we does 20, decisions round CA in behavior only sider [t selection increased nudge significantly optimal longer of no rate round the additional each that find 19 4 = (149) 2 i.2sosta,a A eetdypae h one-shot the played repeatedly CAs as that, shows 2 Fig. ebgnb netgtn A’blesaottedfuteffect, default the about beliefs CAs’ investigating by begin We con- or rounds 20 all across behavior pool we whether Overall, 1 16 = (1) 100% 25% 50% 75% 0% td :pretg fCsuigotmldfutnde na20- a in nudges default optimal using CAs of percentage 2: Study oln cosrud o2,o vrg A used CAs average on 20, to 1 rounds across Pooling 0 = .29, 1234567891011121314151617181920 .01, .27, Percentage ofPeople UsingOptimalDefault p p < < p .Hwvr cosrud 1t 0 we 20, to 11 rounds across However, 0.001]. .Ti a ulttvl iia osome to similar qualitatively was This 0.001]. 0 = .I at atcpnswr obet- no were participants fact, In .78). Round 3 = .89, 19 1.17, = (149) p 2 < 1 0.81, = (1) NSEryEdition Early PNAS u o signif- not but 0.001) p p 0 = p 0 = s .12]. > | . 37]. 0.19). f6 of 3

PSYCHOLOGICAL AND COGNITIVE SCIENCES CMs’ choices. We also investigate CA beliefs related to the CA strategies in the default games. In the stay-or-switch three most common explanations advanced in the literature for default game, CAs presented the optimal default 46.9% of why defaults affect CM behavior (1). First, CMs might stick to the time, which was not significantly different from chance defaults because of the effort costs associated with making an [χ2(1) = 0.99, p = 0.32]. This is consistent with results from stud- active decision. Switching away from a default requires cogni- ies 1 and 2. However, in the preselect default game, CAs tive (and sometimes physical) effort, and so people may stay presented the optimal default 79.0% of the time, which was with the default option because doing so is relatively effortless significantly better than chance [χ2(1) = 102.54, p < 0.001] and (9, 55). Second, CMs might stick to defaults because of loss significantly better than the behavior of CAs who played the stay- aversion (56). Default options often represent the status quo, or-switch default game (z = 7.93, p < 0.001). These patterns pro- and the disadvantages or losses that result from switching loom vide mixed support for hypothesis 1. larger than the advantages or gains. Third, CMs might believe Relationship between CA strategies and CA beliefs. Overall, that defaults signal an implicit recommendation or suggestion there was a small but significant positive correlation between by a CA (52). Relatedly, default options may affect the mean- whether CAs demonstrated correct beliefs and whether they pre- ing CMs assign to actions (36) or may be perceived by CMs as a sented the optimal default (r = 0.10, p = 0.01). Among CAs who social norm (7), both of which can affect how CMs value default played the stay-or-switch default game, this correlation was sig- options relative to nondefault options. Thus, in a final study, nificant (r = 0.18, p = 0.002). However, among CAs who played we investigate CAs’ beliefs about the default effect overall and the preselect default game, this correlation was not significant about each of these three mechanisms in two different types of (r = 0.08, p = 0.19). These patterns provide mixed support for default games: a version of the original “stay-or-switch” default hypothesis 3. game from studies 1 and 2, in which CMs initially start with one CA beliefs about mechanisms driving the default effect. Next, option and then decide whether to stay or switch (53), and a new we analyzed CA beliefs related to the three mechanisms—effort “preselect” default game, in which one option is initially prese- costs, loss aversion, and implicit recommendations—that may lected/prechecked before CMs make a choice (1). drive default effects. To do so, we ran an OLS regression pre- dicting CA beliefs about the percentage of CMs who chose a Materials and Methods. CAs were randomly assigned to play one target option using the following predictor variables: a target- of two default games, both in a medical decision-making context. option-is-default dummy variable equal to 1 if CAs were told The first default game was a stay-or-switch default game, iden- the target option was the default option; three CM subtype tical to the default game in study 1j. CAs decided whether to dummy variables equal to 1 if CAs were told that CMs were tell CMs to imagine they were already taking a medication that more influenced by effort costs, loss aversion, or implicit rec- was less expensive or a medication that had to be taken less fre- ommendations; and (most importantly) interactions between the quently. CMs could either stay with their current medication or target-option-is-default dummy and each of the three CM sub- switch to the other medication. The second default game was a type dummies, using robust standard errors clustered by par- preselect default game, and was newly designed for study 3. CAs ticipant. We found that CAs believed that loss aversion would were told that CMs would indicate which medication they would increase CM selection of the default (b = 0.07, p = 0.007) (57). take by clicking a radio button. CAs then decided which medica- However, CAs did not believe that effort costs (b = 0.03, tion option to initially preselect or precheck on the CM’s decision p = 0.18) or implicit recommendations (b = −0.01, p = 0.68) screen (see Supporting Information for full instructions). would increase CM selection of the default. Additionally, to recover CA beliefs about the default effect, we Considering each default game separately, in the stay-or-switch elicited predictions from CAs about CM behavior under differ- default game, CAs believed that loss aversion (b = 0.15, p < ent defaults. Specifically, we asked CAs to predict what percent- 0.001) and effort costs (b = 0.12, p < 0.001), but not implicit rec- age of CMs would choose each option when it was and was not ommendations (b = −0.04, p = 0.23), would increase CM selec- the default. The difference between these two predictions reveals tion of the default. In the preselect default game, CAs did not CAs’ beliefs about the effect of setting a default. Using analo- gous methods, we also recovered CA beliefs related to the three specific mechanisms commonly discussed in analyses of default Percentage of People With Directionally effects: effort costs, loss aversion, and implicit recommendations. Correct and Incorrect Beliefs We did so by asking CAs to make predictions about (respec- 100% tively) CMs who were more lazy than most people, CMs who cared more about the medications’ drawbacks than most peo- ple, and CMs who were more influenced by their doctors’ rec- ommendations than most people. We counterbalanced whether CAs first played the default game or first made predictions about 75% CM behavior.

Results. Directional Belief Type CA beliefs about the default effect. Overall, only 39.6% of CAs 50% Correct had directionally correct beliefs about the default effect (i.e., Incorrect predicted that more CMs would choose an option if it was the default), which was the minority of CAs by a significant mar- gin [χ2(1) = 25.58, p < 0.001]. Thus, CAs often demonstrated default neglect not only in their decisions about how to use 25% defaults but also in their understanding of what defaults do. Fig. 3 breaks down CAs’ beliefs by the type of default game they played. Among CAs who played the stay-or-switch default

game, 43.2% had directionally correct beliefs about the default 0% effect. Among CAs who played the preselect default game, Stay−or−Switch Preselect 36.3% had directionally correct beliefs about the default effect. Default Game Default Game These patterns are inconsistent with hypothesis 2. CAs who played the stay-or-switch default game had marginally more cor- Fig. 3. Study 3: percentage of CAs revealing directionally correct beliefs and directionally incorrect beliefs about the effect of defaults on CM behavior in rect beliefs than CAs who played the preselect default game 2 the stay-or-switch default game and the preselect default game. Error bars [χ (1) = 2.71, p, = 0.099]. are SEs.

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1712757114 Zlatev et al. Downloaded by guest on September 30, 2021 Downloaded by guest on September 30, 2021 ltve al. et Zlatev feedback. people and experience when for even opportunities that persisted multiple found decisions given we CA were across 2, in default, study neglect the In default con- that as participants. this in option found 2,844 target 1, totaling we the samples study predictions, set 11 In people expert interest. of and best 50.8% theoretical only their choices, both that in others’ to use is find influence trast so and/or we to doing understand Yet when trying to even die. fail when or frequently strategically live CAs defaults as people acting whether people even for deci- and friendly money sions, environmentally enough make have people people whether retirement, whether influence can Defaults Discussion General default which of regardless considered. accurate, the is about from game beliefs far CA are that demonstrate effect 3 default study of results the cess, Information them Supporting of majority opti- significant a the [78%; game, chose default preselect [43% the chance played random with default; mal consistent CAs defaults the these rate optimal as set when a they option expect, at game, an might default setting one stay-or-switch of the As played effect zero. exactly the was that for default 38.8% believed evidence the who of Additional strategies CAs able the 61). of examining being (60, from comes why without possibility reasons this making consis- the decision be articulate optimal would to which in This in cases 59). engage demonstrated has 58, people that (15, research previous not all with need used at tent the processes beliefs CAs in decision one’s default that intuitive consider optimal as possible the game, is select default to it preselect process Second, decision intuitive effect. by an guided beliefs default by was than the rather process in measure about this not default did that we optimal used that but of deliberations the CAs other game, select one that default to suggests preselect possible process the This is decision it do. deliberative about First, to a beliefs possibilities. likely their alternative are considering two defaults which deliberatively about preselect decisions by what their set made to CAs that default notion the with the tent drive to thought mechanisms three the beliefs effect. default of correct seem any had not anticipate they did CAs to if Furthermore, no 3). strategies were hypothesis optimal with also (inconsistent use CAs to 2). hypothesis likely chance with random more of than (inconsistent likely direction correct more be the no (consistent to about again chance were beliefs effect random CA default than the However, better 1). much hypothesis rate with a at defaults effect. default the not but drive game, to stay-or-switch thought mechanisms the recommendations) three in implicit the costs of effort two and aversion anticipate (loss did partially however, least CAs, at defaults. to using were seem marketers and CAs understanding of and majority both firms the at some defaults), bad use least effects systematically at to default seem why using do explain and help understanding may “above-average” both (consistent (which some at beliefs while good correct words, were other had CAs correct In they be 3). if hypothesis to strategies with with likely chance optimal more (inconsistent random use were CAs to than However, chance 2). likely hypothesis default random with more the (inconsistent of no than direction were the better about effect beliefs their no and 1), rate hypothesis a default at of type each in sup- hypotheses exhibiting three the effect, of game. default one the only use for port and understand to ability Discussion. p (b eiv htls vrin(b aversion loss that believe 0 = eetees eadeso h aueo h Adcso pro- decision CA the of nature the of regardless Nevertheless, inconsis- were game default preselect the in choices Thus, optimal used CAs game, default preselect the in contrast, By defaults optimal used CAs game, default stay-or-switch the In = −0.05, ol nraeC eeto ftedefault. the of selection CM increase would .47) χ 2 1 41 = (1) vrl,Csi td eosrtdol partial a only demonstrated 3 study in CAs Overall, p χ 0 = 2 1 1.80, = (1) ,o mlctrcmedtos(b recommendations implicit or .16), .55, p o ute analyses). further for < p hs h pia eal (see default optimal the chose 0.001] 0 = = .Hwvr hnteeCAs these when However, .18]. −0. 003, p 0 = ,efr costs effort .93), 0 = .02, l pin htaees oudrtn rwdl beneficial, widely or understand peo- to give to easy want who are simply bureaucrats that may than options (who more tar- policy ple defaults specific public use toward develop to people help professionals seem influence marketing options) why to get explain incentivized help are may about (who This knowledge 23). tac- own (8, influence social their tics and architecture in choice use invest effectively to to how people motivate might CA better into or (67). decisions translate thinking and could judgments critical that above-average attract abilities possess might perspective-taking fields already some who example, condi- For boundary people neglect. potential few default a for suggests tions deci- This CA still others. better although than significantly 1), sions made study optimality, achieving in from samples” far “professional the (e.g., in tions help out tactics. seek available use would to they default how that the about that likely decisions believe less their even it not makes do addi- exists CAs In effect made. many decision that every fact on the them tion, real- with of consult number to enormous CAs the and world expect between, to far unreasonable and be design few would in are to it groups Team helping these Insights Kingdom), of Behavioral United the purpose the (e.g., express architecture the choice with better exist and tanks organizations influence think some CAs, to Although defaults directions. beneficial use average in to people opportunities pol- that public many implement overlook 15). and suggests will devise icy, to (8, default help participants will the media who of those many power including popular the among in and belief of effect academic lack startling in nudges the using Moreover, behavior of change popularity increasing to and widespread the given people of tactics. processes influence decision of tac- target the influence the in are using flaws who are potential who only people not of tics, flaws processes potential decision consider the to people’s suggest in begin findings other should Our theories in choices. behavioral biases their that influence common to use making people to decision that how argue which recognize social psychology, should powerful in and perspectives models simple related economic current and a challenges understand therefore it and tactic; influence use to to failure ing using seem for opportunities 66). naturally (65, notice others better influence training CAs to mechanisms a CAs defaults of help the that part to aversion—as highlight effect intervention loss to and default costs be anticipate—effort the might approach underlying of One rate sions. high game. default naturally preselect CAs’ one-shot could the given game in utilization CAs, default default preselect by a the improvement whether was of show examine (i.e., version might with one-shot prominent research set repeated choice Future more that probability. was select higher to option a them target leading choice preselected), the the already to which attention in greater they allocated set target have decisions specific a thus visually the toward might CMs affect option influence or to can looking unique CAs which (64). more 63), make to (62, stimuli attention nondefault prominent respective allocate their verbally People to (a options. relative game action), than default game stay-or-switch “stay” prominent default the described visually preselect in more option the was default in than button) the option reason game radio one default preselected default that the (a preselect speculate that the We be why game. at could is default better then, stay-or-switch much arises, the so that one were question choosing natural CAs than intu- A rather be strategy (15). may nudge deliberatively CAs good with that a all possibility where selecting at the itively cases pre- comport suggests the in not This in did beliefs. Even (i.e., decisions CAs’ CM. these defaults game), a using default systematically affect select at to good likely were is CAs default a setting lentvl,ln-emicniesrcue ncranfields certain in structures incentive long-term Alternatively, popula- certain that is findings our of aspect interesting One viewpoint, policy a from surprising also is neglect Default strik- a represents beliefs and strategies CA in neglect Default deci- CA improve to ways explore also should work Future how about beliefs incorrect revealed CAs most 3, study In NSEryEdition Early PNAS | f6 of 5

PSYCHOLOGICAL AND COGNITIVE SCIENCES without necessarily wanting to influence them toward a specific interventions—especially those that people fail to naturally use target option) (68, 69). as social influence tactics—is vast. At the same time, however, Overall, our research highlights the fact that the poten- it seems unlikely that this tremendous potential will be realized tial scope for improving social welfare via behavioral policy without external assistance.

1. Johnson EJ, Goldstein D (2003) Do defaults save lives? Science 302:1338–1339. 38. Fox CR, Rottenstreich Y (2003) Partition priming in judgment under uncertainty. Psy- 2. Bernheim BD, Forni L, Gokhale J, Kotlikoff LJ (2000) How much should Americans be chol Sci 14:195–200. saving for retirement? Am Econ Rev 90:288–292. 39. Van den Steen E (2011) Overconfidence by Bayesian-rational agents. Manage Sci 3. Kirkham E (2016) 1 in 3 Americans has saved $0 for retirement. Time.com Money. 57:884–896. Available at time.com/money/4258451/retirement-savings-survey/. Accessed July 19, 40. Heath C (1999) On the social psychology of agency relationships: Lay theories of 2017. motivation overemphasize extrinsic incentives. Organ Behav Hum Decis Process 78: 4. Madrian BC, Shea DF (2001) The power of suggestion: Inertia in 401(k) participation 25–62. and savings behavior. Q J Econ 116:1149–1187. 41. Kagel JH, Levin D (1986) The winner’s curse and public information in common value 5. Tavoni A, Dannenberg A, Kallis G, Loschel¨ A (2011) Inequality, communication, and auctions. Am Econ Rev 76:894–920. the avoidance of disastrous climate change in a public goods game. Proc Natl Acad 42. Tversky A, Kahneman D (1986) Rational choice and the framing of decisions. J Busi- Sci USA 108:11825–11829. ness 59:S251. 6. Sunstein CR, Reisch L (2014) Automatically green: Behavioral economics and environ- 43. Ball SB, Bazerman MH, Carroll JS (1991) An evaluation of learning in the bilateral mental protection. Harv Environ Law Rev 38:127–158. winner’s curse. Organ Behav Hum Decis Process 48:1–22. 7. Loewenstein G, Bryce C, Hagmann D, Rajpal S (2015) Warning: You are about to be 44. Selten R, Abbink K, Cox R (2005) Learning direction theory and the winner’s curse. nudged. Behav Sci Pol 1:35–42. Exp Econ 20:5–20. 8. Thaler RH, Sunstein CR (2008) Nudge: Improving Decisions About Health, Wealth, and 45. Brehmer B (1980) In one word: Not from experience. Acta Psychol 45:223–241. Happiness (Penguin, New York). 46. Neale MA, Bazerman MH (1991) Cognition and Rationality in Negotiation. (Free Press, 9. Samuelson W, Zeckhauser R (1988) Status quo bias in decision making. J Risk Uncer- New York). tain 1:7–59. 47. Sloman SA, Hagmayer Y (2006) The causal psycho-logic of choice. Trends Cogn Sci 10. Steffel M, Williams EF, Pogacar R (2016) Ethically deployed defaults: Transparency 10:407–412. and consumer protection via disclosure and preference articulation. J Mark Res 53: 48. Johnson EJ, Goldstein DG (2013) Decisions by default. The Behavioral Foundations of 865–880. Public Policy, ed Shafir E (Princeton Univ Press, Princeton), pp 417–427. 11. Suri G, Sheppes G, Schwartz C, Gross JJ (2013) Patient inertia and the status quo bias: 49. Blount S, Larrick RP (2000) Framing the game: Examining frame choice in bargaining. When an inferior option is preferred. Psychol Sci 24:1763–1769. Organ Behav Hum Decis Process 81:43–71. 12. Allcott H, Mullainathan S (2010) Behavior and energy policy. Science 327:1204–1205. 50. Gneezy U, Rustichini A (2000) Pay enough or don’t pay at all. Q J Econ 115:791–810. 13. Larrick RP, Soll JB (2008) The MPG illusion. Science 320:1593–1594. 51. Rogers T, Milkman KL (2016) Reminders through association. Psychol Sci 27: 14. Johnson EJ, et al. (2012) Beyond nudges: Tools of a choice architecture. Market Lett 973–986. 23:487–504. 52. McKenzie CR, Liersch MJ, Finkelstein SR (2006) Recommendations implicit in policy 15. Kahneman D (2011) Thinking, Fast and Slow (Farrar Straus Giroux, New York). defaults. Psychol Sci 17:414–420. 16. Madrian BC (2014) Applying insights from behavioral economics to policy design. 53. Brenner L, Rottenstreich Y, Sood S, Bilgin B (2007) On the psychology of loss aver- Annu Rev Econ 6:663–688. sion: Possession, valence, and reversals of the endowment effect. J Consumer Res 34: 17. Milkman KL, Chugh D, Bazerman MH (2009) How can decision making be improved? 369–376. Perspect Psychol Sci 4:379–383. 54. Andreoni J, Croson R (2008) Partners versus strangers: Random rematching in public 18. Soll JB, Milkman KL, Payne JW (2014) A User’s Guide to Debiasing, Handbook of Judg- goods experiments. Handbook of Experimental Economic Results, eds Plott CR, Smith ment and Decision Making, eds Keren G, Wu G (Wiley-Blackwell, Chichester, UK), VL (Elsevier, New York), Ed 1, Vol 1, pp. 776–783. Vol 33. 55. Baron J, Ritov I (1994) Reference points and omission bias. Organ Behav Hum Decis 19. Montoy JCC, Dow WH, Kaplan BC (2016) Patient choice in opt-in, active choice, and Process 59:475–498. opt-out HIV screening: Randomized clinical trial. BMJ 532:h6895. 56. Kahneman D, Knetsch JL, Thaler RH (1991) Anomalies: The endowment effect, loss 20. Centers for Disease Control and Prevention (2017) HIV in the United States: aversion, and status quo bias. J Econ Perspect 5:193–206. At a glance. Available at https://www.cdc.gov/hiv/statistics/overview/ataglance.html. 57. Imas A, Sadoff S, Samek A (2017) Do people anticipate loss aversion? Manage Sci Accessed July 19, 2017. 63:1271–1284. 21. Dovidio JF, Fiske ST (2012) Under the radar: How unexamined biases in decision- 58. Loewenstein GF, Weber EU, Hsee CK, Welch N (2001) Risk as feelings. Psychol Bull making processes in clinical interactions can contribute to health care disparities. Am 127:267–286. J Epidemiol 102:945–952. 59. Suter RS, Pachur T, Hertwig R (2016) How affect shapes risky choice: Dis- 22. Hovland CI, Janis IL, Kelly HH (1953) Communication and persuasion: Psychological torted probability weighting versus probability neglect. J Behav Decis Making 29: studies of opinion change. Am Sociol Rev 19:315. 437–449. 23. Cialdini RB (1984) Influence: The Psychology of Persuasion (Harper Business, New 60. Klein G (2008) Naturalistic decision making. Hum Factors 50:456–460. York). 61. Neale MA, Northcraft GB (1990) Experience, expertise, and decision bias in negotia- 24. Cialdini RB, Goldstein NJ (2004) Social influence: Compliance and conformity. Annu tion: The role of strategic conceptualization. Research in Bargaining and Negotiating Rev Psychol 55:591–621. in Organizations, eds Sheppard B, Bazerman M, Lewicki R (JAI Press, Greenwich, CT), 25. Petty RE, Cacioppo JT (1986) The elaboration likelihood model of persuasion. Adv Exp Vol 2, pp 55–75. Soc Psychol 19:123–205. 62. Taylor SE, Fiske ST (1978) Salience, attention, and attribution: Top of the head phe- 26. Wood W (2000) Attitude change: Persuasion and social influence. Annu Rev Psychol nomena. Adv Exp Soc Psychol 11:249–288. 51:539–570. 63. Treisman A, Gormican S (1988) Feature analysis in early vision: Evidence from search 27. Freedman J, Fraser SC (1966) Compliance without pressure: The foot-in-the-door tech- asymmetries. Psychol Rev 95:15–48. nique. J Personal Soc Psychol 4:195–202. 64. Halevy N, Chou EY (2014) How decisions happen: Focal points and blind spots in inter- 28. Gouldner A (1960) The norm of reciprocity: A preliminary statement. Am Sociological dependent decision making. J Personal Soc Psychol 106:398–417. Rev 25:161–178. 65. Bazerman MH, Chugh D (2005) Bounded awareness: Focusing failures. Negotiation in 29. Malhotra D, Bazerman MH (2008) Psychological influence in negotiation: An intro- Frontiers of Social Psychology: Negotiation, ed Thompson LL (Psychol Press, College duction long overdue. J Manage 34:509–531. Park, MD), pp 7–26. 30. McGinn KL, Milkman KL, Noth¨ M (2012) Walking the talk in multiparty bargaining: 66. Morewedge CK, et al. (2015) Debiasing decisions: Improved decision making with a An experimental investigation. J Econ Psychol 33:278–291. single training intervention. Pol Insights Behav Brain Sci 2:129–140. 31. Camerer CF (2003) Behavioral Game Theory: Experiments in Strategic Interaction 67. West RF, Toplak ME, Stanovich KE (2008) Heuristics and biases as measures of critical (Princeton Univ Press, Princeton). thinking: Associations with cognitive ability and thinking dispositions. J Educ Psychol 32. Chong D, Druckman JN (2007) Framing theory. Annu Rev Polit Sci 10:103–126. 100:930–941. 33. Crawford VP (1990) Explicit communication and bargaining outcomes. Am Econ Rev 68. Lohr S (October 15, 2011) The default choice, so hard to resist. NY Times. Avail- 80:213–219. able at www.nytimes.com/2011/10/16/technology/default-choices-are-hard-to-resist- 34. DellaVigna S (2009) Psychology and economics: Evidence from the field. J Econ Lit online-or-not.html. Accessed July 19, 2017. 47:315–372. 69. Sunstein CR (2013) Simpler: The Future of Government (Simon & Schuster, New York). 35. Pronin E, Lin DY, Ross L (2002) The bias blind spot: Perceptions of bias in self versus 70. Daniels DP, Zlatev JJ (2017) Choice architects reveal a bias toward positivity and others. Personal Soc Psychol Bull 28:369–381. certainty. SSRN. Available at https://ssrn.com/abstract=2832703. Accessed July 19, 36. Davidai S, Gilovich T, Ross LD (2012) The meaning of default options for potential 2017. organ donors. Proc Natl Acad Sci USA 109:15201–15205. 71. Becker GM, DeGroot MH, Marschak J (1964) Measuring utility by a single-response 37. Van Boven L, Dunning D, Loewenstein G (2000) Egocentric empathy gaps between sequential method. Behav Sci 9:226–232. owners and buyers: Misperceptions of the endowment effect. J Personal Soc Psychol 72. Johnson EJ, Hershey J, Meszaros J, Kunreuther H (1993) Framing, probability distor- 79:66–76. tions, and insurance decisions. J Risk Uncertainty 7:35–51.

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1712757114 Zlatev et al. Downloaded by guest on September 30, 2021