Unethical responds more to instrumental than to hedonic motives Fabio Galeotti, Charlotte Saucet, Marie Claire Villeval

To cite this version:

Fabio Galeotti, Charlotte Saucet, Marie Claire Villeval. Unethical amnesia responds more to in- strumental than to hedonic motives. Proceedings of the National Academy of Sciences of the United States of America , National Academy of Sciences, 2020, 117 (41), pp.25423 - 25428. ￿10.1073/pnas.2011291117￿. ￿halshs-02952265￿

HAL Id: halshs-02952265 https://halshs.archives-ouvertes.fr/halshs-02952265 Submitted on 18 Dec 2020

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Downloaded by guest on December 18, 2020 dishonesty future justify show to we self-excuse decisions. a Thereby, immoral as likely. occurs more undeserved motivated keep is that to forgetting whether on apart. motivated deciding for wk money, as justification 3 such a of action, as instead future serve a day can same amnesia unethical the when elicited However, (n are experiment recalls follow-up which a in in considera- cheating confirmed hedonic past is of that forgetting This find the behavior. motivate we to game, sufficient not mind are a tions of ( variant experiment a online using incentivized misbe- large-scale future a of In by anticipation havior. also in but motives, self-image, strategic moral or one’s instrumental such maintain reasons, to affective willingness or the hedonic behav- may purely as by unethical individuals motivated past be why may these Forgetting ior reasons vividly amnesia. two less unethical explores remember observed experience paper or is remember This amnesia not actions. actions Unethical do in self. people engage moral when 2020) often 3, their June they review contradict Yet, for morality. (received that 2020 about 30, August care approved and Humans NJ, Princeton, University, Princeton Fiske, T. Susan by Edited a Galeotti Fabio motives than hedonic instrumental to to more responds amnesia Unethical o onthv oudt hi oa efve fte are they restore if therefore self-view www.pnas.org/cgi/doi/10.1073/pnas.2011291117 and moral transgressions their behav- past update unethical their to in forget to have engaged able not who do Individuals ior (10). preservation to the image and affect, relates behavior, of of inappropriate motives regulation of the of justification as the such category cleansing, first moral hedonic A purely amnesia. their unethical of vate externalities the downplay (22). pivotality to their (19– or narratives time actions onto over using behavior blame or moral the balancing 21), shifting 18), (17, (16), else uncertainty someone (14, adopt norm others information on can exploiting avoiding behavior their 15), as people of such consequences that negative self-image, the com- strategies about positive thus of a preserve range to Motivated wide passes. the recalls time their past plements as that their that so fuzzier of experiences, fact memory become unwanted the the or characterize shape actions and unethical to forget coined actively “motivated been individuals broadly, have more (13) or, them- memory” (10) forget of amnesia” forgetting” to “unethical think “motivated terms can able The (11), (10–12). they are persons transgressions, honest people past as selves If their reactivate of distorted. such details may and the if behavior obfuscated except be unethical distress, can and of mis- discomfort costs past generate a their and moral for about desire think the people their they deeds, time contradict (5–9), Each that cheating self-image. actions from moral as cost in seen intrinsic engage being an sometimes about perceive care and people their moral most in people Although ordinary lives. dodge affect everyday also or They not criminals. investigation, do concern impairments the only memory respon- impede However, avoid questions. misdeeds, to embarrassing attempt past an and for as (3), serves sibility violence amnesia domestic Often, (4). (2), fraud harassment sexual (1), murders E self-image and ruedAayee eTh de et d’Analyse Groupe oee,w tl nwltl bu h esn htmoti- that reasons the about little know still we However, c ioiso ea eoiin.Cam fansaocrfor occur amnesia of tes- Claims congressional depositions. in legal abound or amnesia timonies apparent of xamples nttt fLbrEoois 31 on Germany Bonn, 53113 Economics, Labor of Institute | memory a,1 hrot Saucet Charlotte , | oiae forgetting motivated oi cnmqe M52,Ui yn NS -93 cly France; Ecully, F-69130 CNRS, Lyon, Univ UMR5824, Economique, eorie ´ b,1 | ntia behavior unethical n ai lieVilleval Claire Marie and , n 1,005) = 1,322) = | doi:10.1073/pnas.2011291117/-/DCSupplemental at online information supporting contains article This 2 1 is ulse etme 8 2020. 28, September published First BY-NC-ND) (CC 4.0 NoDerivatives License distributed under is article access open This Submission.y Direct PNAS a is article This research, interest.y competing performed no declare research, authors The designed paper. M.C.V. y the wrote and and data, C.S., analyzed F.G., contributions: Author moral for desire their people sustain whether to explore 28– both we memory 12, accurately, their (11, to manipulate actions incentivized one past monetarily (27), first are their individuals the memory where on setting motivated a only In of focused 32). values have tests strategic empirical and both almost affective recognize remained models has the economic misbehave theoretical to While when opportunity unexplored. engage a future can as a litera- people manipulation facing which psychological memory in empirical (10–12), mechanism recent amnesia self-management a unethical and on 24–26) ture and 6 refs. unethical behind sec- motives the possible This action, decisions. two amnesia. future future these of any anticipation investigates of the study by present-oriented absence triggered is is the one one in use ond first develop also the can can While and are exclusive. they instrumental, mutually irrespon- and but hedonic not future mechanisms, it) justify two These from to acts. sible strategically utility memories positive positive persons these a clean eco-friendly a as derive have themselves envi- (and only of not the thinking from will forget conscience they systematically ones, actions irresponsible but eco-friendly ronmentally the undertake remember sometimes individuals they if example, postvi- For as [that] actions. only the justifications “previolation not unethical lessen as used future also but be justify justifications can olation amnesia to manipulation instrument unethical memory an as Indeed, as forward-looking, serve is sec- can A motive self. moral possible their ond and memories their between consistency owo orsodnemyb drse.Eal [email protected] Email: addressed. be may correspondence whom work.y To this to equally contributed M.C.V. and C.S., F.G., oiyipiaini htipoigehc eursmaking individuals. requires for difficult ethics more improving amnesia that amnesia. unethical is unethical implication of policy function fur- cost A suggest the and of investigations memory ther and dishonesty the future between on in light engage interplay shed to findings not These behavior. excuse more responsible an forgetting morally lies as to past serve can their value amnesia forget amne- instrumental when Individuals an unethical amnesia. of such motivate addition triggers purely to the that but sufficient sia, found not positive We a are of game. maintenance self-image, the mind as such a considerations, hedonic in past behavior their about cheating forgetting tested individuals’ of we origins possible experiments, two online incentivized large-scale Using Significance ept atltrtr nuehclbhvo frsressee surveys (for behavior unethical on literature vast a Despite a,c,1,2 PNAS anticipated b | NS cecsP,737Prs France; Paris, 75337 Po, Sciences CNRS, coe 3 2020 13, October hett h oa ef rf 3 .1). p. 23, (ref. self” moral the to threat . y raieCmosAttribution-NonCommercial- Commons Creative | . y o.117 vol. https://www.pnas.org/lookup/suppl/ | o 41 no. | 25423–25428

ECONOMIC SCIENCES self-image, and/or as an excuse not to engage in subsequent but because they had to recall a different distribution rela- morally responsible behavior (giving back some undeserved tive to honest individuals. We thus ran two control treatments: money). Our study presents an experimental test of the impact of Hedonic-Control (n = 163) and Instrumental-Control (n = 163). anticipated decisions on memory manipulation. It complements The only difference with the previous treatments is that partici- previous work (11, 12, 33) that, in contrast to us, investigated pants could not cheat in the first part. They had to choose one memory biases as a consequence of past unethical behavior, empty square in each wheel by clicking on it instead of choos- relied on attitudinal measures of memory rather than on behav- ing it in their head before the numbers were displayed on the ior, and did not incentivize participants for providing truthful wheel, and they could not report a different number than the recalls. one displayed in the chosen square. To allow comparisons, the numbers displayed on the chosen squares (not wheels) Testing Hedonic and Instrumental Motives behind reflected the aggregate distribution of the numbers reported in Unethical Amnesia the respective Hedonic and Instrumental treatments. In these To study 1) whether people tend to forget their past dishonest control treatments, participants had no reason to manipulate decisions and 2) whether this results more from the maintenance their memory. The comparison of the main treatments with of moral self-image or from using memory as an instrument the control treatments allows us to separate motivated from to justify future decisions, we ran a large-scale two-part online nonmotivated forgetting. experiment (n = 1,322) (see details in SI Appendix, section A1 and Table S1). Measuring Dishonesty. To identify the individuals who misre- In the first part of the experiment, we designed a variant of a ported numbers to their advantage in the wheel task of part mind game (8, 34–37) inspired by cheating games (5, 38). Par- 1 in the two main treatments, each individual’s distribution ticipants played 20 rounds of a “wheel game” where they could of reported numbers is compared with a uniform distribution misreport the outcome of a random draw to increase their pay- (i.e., the expected distribution of a participant who truthfully off at no risk of detection. The program displayed a wheel with reported all numbers). A participant is classified as dishonest six empty squares, and participants were asked to choose one if the average report is greater than 3.5 (i.e., the mean of a square in their head. Then, the program randomly displayed a uniform distribution) and the reported distribution differs sig- number between 1 and 6 in each square of the wheel. Each num- nificantly from the uniform at the 10% level (χ2 test). Otherwise, ber appeared only once. Participants were asked to report the the participant is classified as honest. We chose a 10% thresh- number displayed in the square they had previously chosen. Par- old to classify participants as honest or dishonest instead of ticipants’ earnings for this part were determined by the number 5% because a stricter threshold would lead us to classify as they reported in one randomly selected round. The higher the honest individual participants whose distribution of reported reported number, the higher the earnings. numbers is farther from the uniform distribution. There is a In the second part of the experiment, conducted 3 wk later, trade-off between the threshold used to define the two cate- participants were first informed that, depending on a random gories of participants and the size of the difference between the draw, they might be given the possibility to reduce their pay- two distributions. We favored the 10% threshold to generate off. If given this possibility, they would have to decide at the a larger difference between the two distributions and provide end of the experiment whether or not to reduce their pay- a better test of the difference in recalls between a more uni- off by a fixed amount, and they were encouraged to do so if form and a less uniform distribution. Also, setting a threshold they had misreported several numbers to their advantage in at 10% is equivalent to conducting a one-sided test, and this the wheel game. All participants were told about the two con- is justified since we only consider left-skewed distributions (by ditions, so that the salience of the moral dimension was kept imposing that the average report is greater than 3.5) in the def- constant across conditions. Then, participants were informed inition of a dishonest subject. Note that our results are robust about their own condition, and they learned that in this part they to classifications using different thresholds (see SI Appendix, would be asked to recall the distribution of the 20 numbers they section A3). reported in the first part of the experiment. Precisely, they had The same method serves to separate participants in the control to recall how many times they reported the numbers 1, 2, 3, 4, 5, treatments. They are classified as having to recall a nonuni- and 6. form distribution if the average of the numbers displayed in We varied between subjects whether only hedonic or both their selected squares is greater than 3.5 and their distribu- hedonic and instrumental reasons could motivate memory. The tion differs significantly from the uniform distribution at the Hedonic treatment corresponds to the condition in which par- 10% level (χ2 test). Otherwise, they are classified as having to ticipants (n = 488) learned before performing the memory task recall an almost-uniform distribution (“almost” because they that they were not given the option to reduce their payoff at the end of the experiment. Thus, biased memory errors, if any, result did not see exactly the same frequency for each number in the from purely affective reasons. The Instrumental treatment corre- selected squares). Note that we also impose a condition on the sponds to the condition in which participants (n = 508) learned mean number to be able to compare the memory of partici- before performing the memory task that they were given the pos- pants classified as dishonest in the main treatments and that sibility to reduce their payoff at the end, and encouraged to do of participants having to recall a nonuniform distribution in so if they overreported their outcomes. Biased memory errors in the control treatments. Overall, 27.25% (26.77%) of the partic- this treatment may be motivated not only by the maintenance of ipants from the Hedonic (Instrumental) treatment are classified moral image but also by strategic reasons, that is, not giving back as dishonest individuals, and 30.06% (30.06%) of the participants part of the undeserved money by persuading oneself that one is from the Hedonic-Control (Instrumental-Control) treatment not concerned. had to recall a nonuniform distribution. Descriptive statistics Since accurate recalls were incentivized, any memory errors on participants’ cheating behavior are provided in SI Appendix, from honest participants can be attributed to random rather section A4. than , and there should be as many errors Some dishonest participants cheated to the full extent by below as above the actual numbers. However, it might be eas- always reporting the highest number. We classify these partic- ier to recall a uniform or close-to-uniform distribution rather ipants as “full cheaters.” The remaining dishonest participants than a biased distribution. Thus, dishonest individuals could dis- are classified as “partial cheaters.” Out of 996 participants allo- play more memory errors not because of motivated memory cated to the Hedonic and Instrumental treatments, 55 cheated to

25424 | www.pnas.org/cgi/doi/10.1073/pnas.2011291117 Galeotti et al. Downloaded by guest on December 18, 2020 Downloaded by guest on December 18, 2020 orcl ouiomdsrbto ( distribution nonuniform Having a treatment. Hedonic-Control recall the more to in significantly than forget Hedonic Instrumental-Control almost-uniform not the respective do in an the Participants in or treatments. reported Instrumental nonuniform numbers and a the of reflecting not distribution could numbers aggregate participants faced controls, the but In cheat Hedonic. for HEDO and Instrumental 1. Fig. almost-uniform an recall to had who individuals 1, of error (Fig. memory distribution form almost-uniform memory an 1, recall average (Fig. treat- to had distribution the Instrumental-Control who participants displays and for errors. both which Hedonic-Control ments, such 1, the on Fig. in impact by error no attested naturally is has This limitations. amnesia motives cognitive unethical different trigger for from would that result conditions Unethical. the Manipulating errors Not Are memory Actions treatments, When Errors Memory Results memory. motivated captures average thus positive error A is memory capture forgetting. error nonmotivated may memory and error motivated negative memory both a positive While a out- recalled nonmotivated, part. average reported the exclusively second and the average experiment the in the of outcome part between first the difference in come the as Errors. defined Memory Measuring in provided A6. is and A5 checks, sections robustness cheaters, full and additional briefly partial of A with then behavior cheaters. the partial and of analysis and detailed cheaters full more between we full differences Below, may the excluding numbers. on it results different comment same: the than the 6s report be have twenty not first cheaters recall may to full behavior easier their and be recall partial to their that less provide manipulate care effort to to need may the the cheaters feel Finally, Full not memory. may nature: are and very cheaters image con- their their full the about in and between differ partial excluding to comparison Second, Hence, likely treatments. clean 6s. main a twenty and allows recall trol cheaters to full had treatments, subjects the control the the in of First, and none reasons. partial several of behavior for the cheaters separately full analyze We extent. full the ro asidct tnaderr ftema.Semi etfrstatistics. for text main See mean. the of ***P errors standard indicate nonuniform 4 bars vs. Error Model (almost-uniform distributions), 7 of (almost-uniform Model distributions) 1 and Model distributions), from (nonuniform computed (Left are distribution almost-uniform els an recall to having than aetie al. et Galeotti Average Memory Error <

1 S=ntsignificant. not = NS 0.01, 0 .2 .4 .6 vrg eoyerri h oto ramns NTUi for is INSTRU treatments. control the in error memory Average NTUCnrlHEDO_Control INSTRU_Control Almost-uniform distribution al S2 Table Appendix, SI (mean) treatment NS n hs h a orcl nonuni- a recall to had who those and Left) atcpns eoyerr are errors memory Participants’ .I hw htteaverage the that shows It Right). Right *** sealso (see *** 0 .2 .4 .6 ed omr eoyerrors memory more to leads ) NTUCnrlHEDO_Control INSTRU_Control Non-uniform distribution eto A7 ). section Appendix, SI (mean) treatment NS nteControl the In .Sgicnelev- Significance ). IAppendix, SI al S3 of models Table All variables. Appendix, control various SI including A5), section in had reported treat- analyses who Pooled-Control participants the in the (P distribution ment indistin- of nonuniform is a error recall individuals average to dishonest the of from nonuni- error guishable a treatment, memory Hedonic recall the average In to cheat. the not had could but and who distribution Instrumental-Control form individuals between for errors nonuni- in Hedonic-Control a difference no recall was to (Mean treatment had Pooled-Control who 0.29; the individuals in (Mean distribution than treatment form and Hedonic test) the Wald (Mean in treatment Instrumental than the in errors memory atwe ogtigcnb sda efjsicto ontgive not to self-justification a back. as money used undeserved be can impor- forgetting Instrumental more when is tant treatments, amnesia the unethical Hedonic whether Comparing the indicates and respectively, image. Pooled-Control the moral and them treatment allows their dishon- it maintain when whether treatment forgetting to indicates motivated Hedonic exhibit treatment the individuals est in Pooled-Control had distribution the Memory. who nonuniform and participants Motivated a of of recall error Source to memory Main average the the Comparing Is Amnesia Instrumental dis- S4). Table almost-uniform (Mean and A5 an treatment recall Pooled-Control to P the participants had but to in cheat compared tribution not are could honest who as classified participants ewe atcpnscasfida oeti h eoi and Hedonic the in P honest (Mean as treatments Instrumental classified participants 2, between Fig. honestly. reported Left participants when amnesia motivated almost-uniform difficult an more than outcomes. is their “Pooled- distribution it misreport as that nonuniform (P them shows a one to also recall 1 refer Fig. and to that analysis Note the Control.” of remainder the respec- tively; treatments, Instrumental-Control and Hedonic-Control (Mean together participants all pooling when treatments Instrumental-Control and Hedonic-Control in significantly the 0.29 not distribu- again (P and is nonuniform treatment different which Instrumental-Control a treatment, the recall Hedonic-Control in the to 0.28 had statistically is who tion is individuals which of error treatment, and Hedonic-Control (P treatment indistinguishable the Instrumental-Control in the in 0.08 0.11 is distribution h he ak?I h ntuetltetet h average the treatment, in Instrumental times twenty the “6” In a task)? reported wheel who (those the cheaters Amnesia. what but full Instrumental cheaters, partial about of for Need forgetting instrumental in of evidence Less Are Cheaters to Full self-justification amnesia. not unethical a are generate as motives to study hedonic serves our purely it in sufficient while when money, is, forget- undeserved for that keep motive present, instrumental is an the exhibit ting when in than individuals errors distribution memory errors dishonest nonuniform higher memory Thus, a higher treatment. recall exhibit to Pooled-Control had not who treatment, do individuals Hedonic latter mem- the the in higher individuals whereas significantly dishonest than exhibit errors ory treatment Instrumental the 0 = 0 = i.2, Fig. unn otemi ramns hr sn vdneof evidence no is there treatments, main the to Turning hw osgicn ifrnei vrg eoyerrors memory average in difference significant no shows .105 adts) hr sn ifrneete hnthe when either difference no is There test). Wald .593, P P 0 = 0 = 0 = 0 = Right and . .B otat i.1, Fig. contrast, By .006). vnfrpriiat h a ocac to chance no had who participants for even 008) . .Teerslsaespotdb h regression the by supported are results These .963). .Teeoe epo h oto ramnsfor treatments control the pool we Therefore, 854). 0 = P hw htdsoetpriiat xii higher exhibit participants dishonest that shows PNAS .Teei odfeec ihrbetween either difference no is There .904). 0 = al S3 Table Appendix, SI epciey (see respectively) .245, 0 = | coe 3 2020 13, October adts) h vrg memory average The test). Wald .764, hwta ihns niiul in individuals dishonest that show ME .2ad00,respectively; 0.04, and 0.02 = ME | Right o.117 vol. ME 0 = as see (also IAppendix SI 0 = .15 hw htthere that shows .29; | n .6i the in 0.16 and o 41 no. IAppendix, SI ME ME P hr is There 0 = section , 0.09; = 0 = | ME .006, 25425 .54) =

ECONOMIC SCIENCES trast, partial cheaters most likely faced a trade-off in part 1 Almost-uniform distribution Non-uniform distribution between overreporting the numbers to increase their earnings *** and maintaining a positive self-image. For these participants, for- *** getting their past cheating behavior could have been the only .6 .6 viable self-management strategy to self-justify not returning the undeserved money.

.4 Hedonic Motives Do Not Affect the Quality of Short-Term Mem- .4 NS ory Either. Finding no difference in average memory errors between the Hedonic and the Pooled-Control treatments indi- cates that affective motives alone are not sufficient to trigger NS .2 NS .2 motivated memory. However, participants in the control treat-

Average Memory Error Average NS ments reported numbers that they did not get to choose. Thus, memory errors in these treatments may also capture inattention, which may have produced more volatile souvenirs than when 0 0 people were actively dishonest. This would be consistent with INSTRU HEDO CONTROL INSTRU HEDO CONTROL pooled_treatment pooled_treatment Saucet and Villeval (30), who found more memory errors in dic- tator games when the amount to recall was randomly assigned by the program than when it was chosen by the dictator. Therefore, Fig. 2. Average memory error of the honest and dishonest participants in we conducted a follow-up experiment (n = 1,005) (see details in the main treatments. INSTRU is for Instrumental, HEDO for Hedonic, and CONTROL for the Pooled-Control treatment. The average memory error of SI Appendix, section A1 and Table S9). The Short-Term-Memory honest participants in the main treatments does not differ from the average treatments replicated the same treatments as in the original error of participants who had to recall an almost-uniform distribution in the experiment, except that the wheel game and the recall task were Pooled-Control treatment (Left). The average memory error of dishonest performed in the same session. Thus, even if participants were participants in the Hedonic treatment does not differ from that of partic- possibly still less attentive in the control than in the main treat- ipants who had to recall a nonuniform distribution in the Pooled-Control ments, they had a higher chance to recall their reported number treatment (Right). However, the average error of dishonest participants in distribution. In this setting, the comparison between the new the Instrumental treatment differs from the average error of those who Hedonic and control treatments (all indexed by ST for Short- had to recall a nonuniform distribution in the Pooled-Control treatment. Term-Memory) provides additional insight on the existence of Significance levels are computed from Model 1 in SI Appendix, Table S3 (for dishonest subjects) and SI Appendix, Table S4 (for honest subjects) (see also an affective motive behind motivated memory. SI Appendix, section A5). Full cheaters are not included. Error bars indicate Not surprisingly, in each Short-Term-Memory treatment the standard errors of the mean. ***P < 0.01, NS = not significant. average memory error is lower than in our original experi- ment because there is less time to forget. Like in the original experiment, there is no memory bias when participants reported memory error of full cheaters (n = 29, MeanME = 0.26) is sig- honestly (Fig. 3, Left and SI Appendix, Table S11). None of the nificantly lower than that of partial cheaters (n = 88, MeanME = treatment pairwise comparisons are significant (P > 0.10). By 0.54) (P < 0.001, Mann–Whitney U test). This does not result contrast, when considering dishonest participants (Fig. 3, Right), from differences in cognitive memory abilities between these two groups, as no difference in performance is observed between partial and full cheaters in a word memory task based on ref. 39 that we administered at the end of part 1. On aver- Almost-uniform distribution Non-uniform distribution age and on aggregate, partial cheaters recalled correctly 25.58 (25.86 in the Instrumental treatment) words out of 35 and full cheaters 24.63 (24.24) (P = 0.426 and 0.510, Mann–Whitney U .6 .6 tests). Also, the difference does not seem to be driven by the fact that recalling is cognitively easier for full cheaters than .4 .4 ** for partial cheaters. Indeed, in the Hedonic treatment, there NS is no significant difference in the mean memory error between NS full cheaters and partial cheaters (MeanME = 0.39 and 0.29,

respectively, P = 0.448). Moreover, for the whole population, .2 .2 the correlation between the average memory error and the SD NS of the numbers actually reported in the wheel task is significant NS

Average Memory Error Average NS neither at the aggregate level (ρ = −0.015, P = 0.792, pairwise Pearson’s correlation), nor at the treatment level (ρ = −0.020, 0 0 0.040, and 0.050; P = 0.858, 0.674, and 0.595 in the Pooled- Control, Hedonic, and Instrumental treatments, respectively). INSTRU_ST HEDO_ST CONTROL_ST INSTRU_ST HEDO_ST CONTROL_ST The regression analysis reported in SI Appendix, Table S3 con- pooled_treatment pooled_treatment firms that a lower SD in the distribution of the numbers reported does not significantly decrease memory errors. Hence, the higher Fig. 3. Average memory error of the honest and dishonest participants in memory errors exhibited by partial cheaters does not seem the Short-Term-Memory (ST) treatments. The average memory error of hon- to be explained by the higher variation in the numbers they est participants in the ST treatments does not differ from that of participants had to recall. A more likely explanation for the difference in who had to recall an almost-uniform distribution in the Pooled Control ST memory errors between full and partial cheaters in the Instru- treatment (Left). Only the average memory error of dishonest participants in the Instrumental ST treatment differs from the average error of participants mental treatment is that these two types of individuals differ who had to recall a nonuniform distribution in the Pooled-Control ST treat- in terms of intrinsic motivation. Full cheaters, who deliber- ment (Right). Significance levels are computed from Model 1 of SI Appendix, ately chose to maximize their payoff in the wheel task, might Table S10 (dishonest subjects) and SI Appendix, Table S11 (honest subjects). care less about their image and, thus, may not need additional Full cheaters are not included. Error bars indicate standard errors of the arguments to not return part of the undeserved money. By con- mean. **P < 0.05, NS = not significant.

25426 | www.pnas.org/cgi/doi/10.1073/pnas.2011291117 Galeotti et al. Downloaded by guest on December 18, 2020 Downloaded by guest on December 18, 2020 hw httervrecuaiyi lotu:Pol experience People true: also is over causality repeatedly study reverse Our the unethically (11). that act amnesia shows unethical to experience they likely because more time demonstrating are research people prior that complement self- also results past Our self-management tool. with consider future-oriented to a cope as crucial also is manipulation it to memory that show strategy we decisions, present-oriented threatening image a as ulation keeping for self-justification uneth- needed money. reason past undeserved they possible their because underestimated A behavior people recalls. that ical moral of is future quality results a these the of for affect engage addition not to the did have making, their decision not decision again did ethical engaged prior individuals that in when make contrast, to they By that decision morality. informed future being a when have accuracy would less with trig- behavior their motive past recalled instrumental individuals Dishonest an forgetting. of self-serving gered addition the amnesia, unethical compensate ego-utility. in somewhat loss might a for utility initially Monetary had (32). they that suppressed to traces people memory induce image-threatening may recalls recover for incentives monetary suffi- high participants Providing ciently our recalls. cheat- accurate whereas for of recalls, incentives monetary truthful memories received incentivize investigating not most did experiments Second, ing negatively. previous memory the of higher may of quality which the the treatments, of impacted control the have because in Short-Term-Memory test participants our of conservative previ- passiveness even a contradict First, future provides necessarily 33). not treatments 12, no does (11, when made evidence absence amnesia be ous This unethical to before. has behind decision decision moral motive distribution any nonuniform hedonic in errors similar of engage a memory not recall did higher to but had significantly who exhibit those than not did but treatment decisions. misconduct immoral a future justify after to self-image self-excuse a moral as neces- their more not restore memory to motivated result sarily experience important people This memory. that a their indicates motivated as past misbe- serve the has who in can individuals amnesia haved of action—that forgetting future unethical a forgetting when when for is, self-justification motivate only value—that is to instrumental It an sufficient behavior. not cheating investigated past are that study showed motives We our amnesia. hedonic unethical making, behind motives decision the possible two in dishonest forgetting of motivated of context evidence clean providing (not Besides so do to Discussion reason this strategic of a realization the has to money). one returning time the when in when closer even memory is one’s decision, decision manipulate a to of difficult retrieval cheat. more to be opportunity no may had It who individuals recalls sta- of short-term a that of to quality generate compared the to in strong difference for significant sufficiently motives tistically not hedonic Thus, are negative. distortion even memory is coefficient its of sign (P control the Instrumental the in errors in in reported analysis sion Pooled-Control Hedonic the the by lowed vdneo ntuetlfretn loi h hr em How- term. short the Hedonic in the ever, also forgetting instrumental of evidence (P the is error memory average The Instrumental the clearly: in highest appears ranking a aetie al. et Galeotti hl rvoseprclsuisivsiae eoymanip- memory investigated studies empirical previous While generate to insufficient were reasons hedonic purely While Hedonic the in dishonestly behaved who individuals Indeed, eto A9 section Appendix, SI 0 = .029 nMdl1, Model in > 0.10 Tteteti o infiatydfeetfrom different significantly not is treatment ST Ttetet(Mean treatment ST nalmdl) n oto h iethe time the of most and models), all in Ttetet(Mean treatment ST al S10 Table Appendix, SI P Ttetetdfe rmtecontrol the from differ treatment ST 0 = Ttetet(Mean treatment ST niae httema memory mean the that indicates .054 nMdl4,soigsome showing 4), Model in ME 0 = ME .Teregres- The .07). n summarized and 0 = ME ,adby and .18), 0 = ,fol- .26), clTr;105idvdas(9 ae;ma g 68 ,S 12 y) 11.24 = SD Hedonic y, the 36.87 = in age 254 mean total, Mechan- males; in Amazon (597 participated through individuals recruited 1,005 were Turk; and ical States United the that in sure located earned make They to part. parts 0.48). second = second the plus (SD $4.31 complete and part, average to second first on return the the would completing in participants for part, made most $1.5 second earnings the of of joint payoff end the the fixed At a participation. received received their participants for they part, $1.5 first of the payoff power of fixed end statistical a the At The completed. y). were parts 10.68 two = SD y, 37.62 in = treat- described Instrumental-Control age is the analysis in mean 163 males; and males; y), (75 (74 10.93 mean treatment = ment Hedonic-Control males; SD the y, (275 in 38.21 488 treatment 163 = y), age Instrumental part); 11.22 mean the = = in SD first y, age 508 40.23 the mean y), = age males; 11.40 completed wk = = (235 SD (1,550 SD 3 treatment y, y, Hedonic parts study 40.01 39.58 the = two the in age participated the of mean individuals males; completed part (659 second y) individuals 1,322 the 11.22 were part; and complete first study the could online after two-part they our that in participate informed to Turk Mechanical zon Groupe the (2018-0015). of procedures Board the Review approved The Th experiment. de the et of d’Analyse beginning the found at be sent can tables and procedures and in design experimental the of Details Methods and Materials infer to translate helpful particularly would recommendations. be func- costly policy would cost behavior more ethical the more amnesia which 37), into unethical to (8, Exploring making unknown. cheating totally debate extent to is of amnesia a able function unethical is of are cost tion there they moral may while that the costs However, on anticipate these memories. can but their individuals 9) manipulate if (5–7, exploit- reduced opportunities The fully be from cheating cheaters. individuals their full most prevent ing than lying amnesia of exhib- costs intensity unethical cheaters moral the strategic partial with experiment more our behavior ited In unethical amnesia. that of unethical function cost of the moral of the characteristics also relates be the would investigate metacog- It to on sophistication. important debate of degree Conducting the individuals’ to future-oriented. and contribute whether nition nicely or on would present- depending studies such differ is forget- also amnesia genuine might to unethical networks compared These costs networks, neural ting. of and computation different areas activate controlled might brain manipulation retrieval, more memory such or a benefits, and involve over would control which inhibit indi- deliberately If (46). viduals techniques neurophysiology models and with these (45) methods neuroimaging Confronting experimental combine 44). to (27, forget- require might deliberate that empirically and so active individuals, models is to other ting control (40–43), memory process metacognitive theo- heuristic and large a some grant naive of While result are the individuals deliberate. are biases that less consider or models more retical justify is to forgetting excuse vated an as serve can behavior. unethical this future when amnesia unethical 12mls enae=3.2y D=1.8y,ad28i h Instrumental- the in 248 and y), 12.08 = Hedonic-Control SD y, the 36.72 in Control = 252 age y), 11.44 mean = males; SD (152 y, Instrumental 37.77 the in = 251 age y), mean 10.31 = SD y, 36.13 = age atcpnsa h 09Erpa cnmcSineAscainmeeting Association Science Economic European seminar 2019 and the comments helpful at very their participants for Zimmermann F. and Tarroux, B. ACKNOWLEDGMENTS. 10.17605/OSF.IO/QMC8J ). (doi: Availability. Data 0.48). = (SD $4.35 average on earned IAppendix, SI atcpnsi h olwu hr-emMmr xeietwr also were experiment Short-Term-Memory follow-up the in Participants the of all until provided was earnings or performance on feedback No Ama- through recruited were States United the in located Individuals moti- whether establish to needed is evidence empirical More Ttetet(4 ae;ma g 68 ,S 10 ) They y). 11.05 = SD y, 36.88 = age mean males; (142 treatment ST l ftepriiat aeterifre con- informed their gave participants the of All A1. section aahv endpstdi pnSineFramework Science Open in deposited been have Data oi cnmqea h nvriyo ynrvee and reviewed Lyon of University the at Economique eorie ´ PNAS . A2 section Appendix, SI eaegaeu oP ebih,S Suetens, S. Seabright, P. to grateful are We | coe 3 2020 13, October Ttetet(5 ae;mean males; (156 treatment ST | o.117 vol. Ttetet(4 males; (147 treatment ST | o 41 no. Ttreatment ST | 25427

ECONOMIC SCIENCES in Dijon for their feedback. We also thank MTurk Data for assisting with the French National Research Agency (ANR-19-CE26-0019). The research has data collection. This research has been funded by the Laboratory of Excel- benefited from the support of Initiative of Excellence Lyon (IDEXLYON) from lence CORTEX (construction, cognitive function, rehabilitation and repair Universite´ de Lyon (project Institutional Design and Economic Preferences: of the cortex) (ANR-11-LABX-0042) of Universite´ de Lyon within the pro- Theory and Experiments - INDEPTH) within the Program Investissements gram “Investissements d’Avenir” (ANR-11-IDEX-007) operated by the French d’Avenir (ANR-16-IDEX-0005). C.S. gratefully acknowledges the European National Research Agency and by the DECISION project (A study on Dis- Research Council, grant 850996, MOREV (Motivated Reading of Evidence) honesty: from Emotions and Cognition to Institutions and Organizations) of project.

1. P. J. Taylor, M. D. Kopelman, Amnesia for criminal offences. Psychol. Med. 14, 581–588 25. B. Irlenbusch, M. C. Villeval, Behavioral ethics: How psychology influenced eco- (1984). nomics and how economics might inform psychology?. Curr. Opin. Psycol. 6, 87–92 2. D. Bourget, J. M. Bradford, Sex offenders who claim amnesia for their alleged offense. (2015). Bull. Am. Acad. Psychiatry Law 23, 299–307 (1995). 26. C. Jacobsen, T. R. Fosgaard, D. Pascual-Ezama, Why do we lie? A practical guide to the 3. G. Swihart, J. Yuille, S. Porter, The role of state-dependent memory in “red-outs”. Int. dishonesty literature. J. Econ. Surv. 32, 357–387 (2018). J. Law Psychiatry 22, 199–212 (1999). 27. R. Benabou,´ J. Tirole, Self-confidence and personal motivation. Q. J. Econ. 117, 871– 4. M. D. Kopelman, R. Green, E. Green, P. Lewis, N. Stanhope, The case of the amnesic 915 (2002). intelligence officer. Psychol. Med. 24, 1037–1045 (1994). 28. K. K. Li, “What determines overconfidence and memory recall bias? The role of 5. U. Fischbacher, F. Follmi-Heusi,¨ Lies in disguise—An experimental study on cheating. feedback, awareness and social comparison” (Mimeo, City University of Hong Kong, J. Eur. Econ. Assoc. 11, 525–547 (2013). 2017). 6. J. Abeler, D. Nosenzo, C. Raymond, Preferences for truth-telling. Econometrica 87, 29. S. H. Chew, W. Huang, X. Zhao, Motivated . SSRN (2020). https://papers. 1115–1153 (2019). ssrn.com/sol3/papers.cfm?abstract id=2127795. Accessed 15 September 2020. 7. M. Dufwenberg, M. A. Dufwenberg, Lies in disguise–A theoretical analysis of 30. C. Saucet, M. C. Villeval, Motivated memory in dictator games. Game. Econ. Behav. cheating. J. Econ. Theory 175, 248–264 (2018). 117, 250–275 (2019). 8. U. Gneezy, A. Kajackaite, J. Sobel, Lying aversion and the size of the lie. Am. Econ. 31. R. W. Carlson, M. A. Marechal,´ B. Oud, E. Fehr, M. J. Crockett, Motivated misremem- Rev. 108, 419–453 (2018). bering of selfish decisions. Nat. Commun. 11, 2100 (2020). 9. K. Khalmetski, D. Sliwka, Disguising lies–Image concerns and partial lying in cheating 32. F. Zimmermann, The dynamics of motivated beliefs. Am. Econ. Rev. 110, 337–361 games. Am. Econ. J. Microecon. 11, 79–110 (2019). (2020). 10. M. Kouchaki, F. Gino, Dirty deeds unwanted: The use of biased memory processes in 33. L. L. Shu, F. Gino, M. H. Bazerman, Dishonest deed, clear conscience: When cheating the context of ethics. Curr. Opin. Psycol. 6, 82–86 (2015). leads to moral disengagement and motivated forgetting. Pers. Soc. Psychol. Bull. 37, 11. M. Kouchaki, F. Gino, Memories of unethical actions become obfuscated over time. 330–349 (2011). Proc. Natl. Acad. Sci. U.S.A. 113, 6166–6171 (2016). 34. T. Jiang, Cheating in mind games: The subtlety of rules matters. J. Econ. Behav. Organ. 12. M. L. Stanley, P. Henne, V. Iyengar, W. Sinnott-Armstrong, F. De Brigard, I’m not the 93, 328–336 (2013). person I used to be: The self and autobiographical memories of immoral actions. J. 35. S. Shalvi, C. K. De Dreu, Oxytocin promotes group-serving dishonesty. Proc. Natl. Exp. Psychol. Gen. 146, 884–895 (2017). Acad. Sci. U.S.A. 111, 5503–5507 (2014). 13. J. A. Singer, P. Salovey, “Motivated memory: Self-defining memories, goals, and 36. J. Potters, J. Stoop, Do cheaters in the lab also cheat in the field? Eur. Econ. Rev. 87, affect regulation” in Striving and Feeling: Interactions Among Goals, Affect, and 26–33 (2016). Self-Regulation, L. L. Martin, A. Tesser, Eds. (Psychology Press, 1996), pp. 229–250. 37. A. Kajackaite, U. Gneezy, Incentives and cheating. Game. Econ. Behav. 102, 433–444 14. L. Feiler, Testing models of information avoidance with binary choice dictator games. (2017). J. Econ. Psychol. 45, 253–267 (2014). 38. S. Shalvi, J. Dana, M. J. Handgraaf, C. K. De Dreu, Justified. ethicality: Observing 15. Z. Grossman, J. J. Van Der Weele, Self-image and willful ignorance in social decisions. desired counterfactuals modifies ethical perceptions and behavior. Organ. Behav. J. Eur. Econ. Assoc. 15, 173–217 (2017). Hum. Decis. Process. 115, 181–190 (2011). 16. C. Bicchieri, E. Dimant, S. Sonderegger, It’s not a lie if you believe it: Lying and 39. H. L. Roediger, K. B. McDermott, Creating false memories: Remembering words not belief distortion under norm-uncertainty. SSRN (2019). https://papers.ssrn.com/sol3/ presented in lists. J. Exp. Psychol. Learn. Mem. Cognit. 21, 803–814 (1995). papers.cfm?abstract id=3326146. Accessed 15 September 2020. 40. D. Kahneman, A. Tversky, Subjective probability: A judgment of representativeness. 17. B. Bartling, U. Fischbacher, Shifting the blame: On delegation and responsibility. Rev. Cognit. Psychol. 3, 430–454 (1972). Econ. Stud. 79, 67–87 (2011). 41. S. Mullainathan, A memory-based model of bounded rationality. Q. J. Econ. 117, 735– 18. R. Oexl, Z. J. Grossman, Shifting the blame to a powerless intermediary. Exp. Econ. 16, 774 (2002). 306–312 (2013). 42. P. Bordalo, N. Gennaioli, A. Shleifer, “Memory, , and choice” (NBER Work- 19. M. Ploner, T. Regner, Self-image and moral balancing: An experimental analysis. J. ing Paper No. 23256, National Bureau of Economic Research, Cambridge, MA, Econ. Behav. Organ. 93, 374–383 (2013). 2017). 20. U. Gneezy, A. Imas, K. Madarasz,´ Conscience accounting: Emotion dynamics and social 43. A. L. Bodoh-Creed, Mood, memory, and the evaluation of asset prices. Rev. Finance behavior. Manage. Sci. 60, 2645–2658 (2014). 24, 227–262 (2020). 21. D. Cojoc, A. Stoian, Dishonesty and charitable behavior. Exp. Econ. 17, 717–732 (2014). 44. D. Gottlieb, Imperfect memory and choice under risk. Game. Econ. Behav. 85, 127–158 22. R. Benabou,´ A. Falk, J. Tirole, “Narratives, imperatives and moral reasoning” (CEPR (2014). Discussion Paper 13056, Center for Economic and Policy Research, Washington, DC, 45. M. Anderson, S. Hanslmayr, Neural mechanisms of motivated forgetting. Trends 2018). Cognit. Sci. 18, 279–292 (2014). 23. S. Shalvi, F. Gino, R. Barkan, S. Ayal, Self-serving justifications: Doing wrong and 46. X. Hu, Z. Bergstrom, G. Bodenhausen, J. Rosenfeld, Suppressing unwanted autobio- feeling moral. Curr. Dir. Psychol. Sci. 24, 125–130 (2015). graphical memories reduces their automatic influences: Evidence from electrophys- 24. S. M. Rosenbaum, S. Billinger, N. Stieglitz, Let’s be honest: A review of experimental iology and an implicit test. Psychol. Sci. 26, 1098–1106 evidence of honesty and truth-telling. J. Econ. Psychol. 45, 181–196 (2014). (2015).

25428 | www.pnas.org/cgi/doi/10.1073/pnas.2011291117 Galeotti et al. Downloaded by guest on December 18, 2020