UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Negative indirect reciprocity : theory and evidence Chudek, Matthew 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2013_fall_chudek_matthew.pdf [ 1.5MB ]
JSON: 24-1.0074109.json
JSON-LD: 24-1.0074109-ld.json
RDF/XML (Pretty): 24-1.0074109-rdf.xml
RDF/JSON: 24-1.0074109-rdf.json
Turtle: 24-1.0074109-turtle.txt
N-Triples: 24-1.0074109-rdf-ntriples.txt
Original Record: 24-1.0074109-source.json
Full Text

Full Text

Negative Indirect ReciprocityTheory and EvidencebyMatthew ChudekB.A. University of Melbourne, 2003M.A. University of British Columbia, 2009a thesis submitted in partial fulfillmentof the requirements for the degree ofDoctor of Philosophyinthe faculty of graduate and postdoctoralstudies(Psychology)The University Of British Columbia(Vancouver)August 2013? Matthew Chudek, 2013AbstractExplanations of humans? evolutionary origins that invoke the ratchet of cu-mulative cultural learning must confront the ?cooperative dilemma of cul-ture?. Adaptive cultural knowledge is a widely shared but easily degradedpublic good. How did our ancestors avoid the temptation to hoard valuableknowledge and to deceive and manipulate each other, before the advent ofcomplex social institutions? I present one possible solution: negative in-direct reciprocity (NIR). I use a series of mathematical models to reasonabout how our ancient ancestors? dispositions to gainfully exploit one an-other could have supported more complex forms of cooperation, providinga foundation for our rapidly evolving corpus of shared cultural know-how.Together these models show how reputation-based, opportunistic exploita-tion can play a pivotal role in sustaining cooperation in small scale societies,even before the advent of complex institutions.I also present two empirical tests of the assumptions made by thesemodels. First, I measure contemporary reputational judgements in circum-stances that the models predict are relevant. In the process I also map myparticipants? judgements to the full set of first and second-order reputationassessment rules described by indirect reciprocity theory. Second, I testwhether a recently observed peculiarity of people?s moral reasoning?ourtendency to ascribe blame to those who profit from others suffering becauseof mere good fortune?is consistent with the constraints assumed by NIR.The results of both empirical studies support the assumptions made by NIR.iiPrefaceChapters three, four and five of this thesis are being prepared for publication.Chapters 3 and 4 are coauthored by Joseph Henrich.Chapter 3 is a theoretical model initially suggested by Joseph Henrich.I developed and refined numerous variants of this model. Henrich has pro-vided in depth conceptual feedback throughout and has also provided de-tailed feedback of many versions of this manuscript.The initial idea for the empirical study in chapter 4 emerged from dis-cussions between Henrich and myself. I designed the details of the study,built a web-survey system capable of administering it, gathered and analysedthe data, produced figures and prepared the manuscript. Henrich provideddetailed feedback on the manuscript.Henrich also provided all funding to pay participants in the studies de-scribed in chapters 4 and 5.The data for study one in chapter 5 were gathered by Erik Thulin.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . ixDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 The cooperative dilemma of an emerging cultural species 92.1 The cooperative dilemma of culture (a.k.a. the evil teacherproblem) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.1.1 Why does culture require cooperation? . . . . . . . . . 162.1.2 Why does (human-like) cooperation require culture? . 202.2 Three kinds of solutions . . . . . . . . . . . . . . . . . . . . . 222.2.1 Route 1: The big step . . . . . . . . . . . . . . . . . . 232.2.2 Route 2: The arms race . . . . . . . . . . . . . . . . . 242.2.3 Route 3: The ratchet . . . . . . . . . . . . . . . . . . . 26iv3 Negative indirect reciprocity . . . . . . . . . . . . . . . . . . 283.1 Mathematical model . . . . . . . . . . . . . . . . . . . . . . . 493.1.1 Context and overview . . . . . . . . . . . . . . . . . . 493.1.2 Model definition . . . . . . . . . . . . . . . . . . . . . 513.1.3 Reputational dynamics . . . . . . . . . . . . . . . . . . 543.1.4 Behavioural dynamics . . . . . . . . . . . . . . . . . . 583.1.5 Combining reputations and behaviour . . . . . . . . . 593.1.6 Discrete strategy approach . . . . . . . . . . . . . . . 623.1.7 Evolving continuous traits interpretation . . . . . . . . 773.2 Additional supplemental materials . . . . . . . . . . . . . . . 864 Surveying indirect reciprocity: how do people assign rep-utations? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1124.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1314.4 Supplemental . . . . . . . . . . . . . . . . . . . . . . . . . . . 1365 It is better to profiteer on the guilty: is moral condemna-tion sensitive to reputation? . . . . . . . . . . . . . . . . . . 1375.1 Methods, study one . . . . . . . . . . . . . . . . . . . . . . . 1415.2 Results, study one . . . . . . . . . . . . . . . . . . . . . . . . 1435.2.1 Do these questions index the same underlying construct?1435.2.2 Was condemnation of the lucky profiteer sensitive tothe victims? reputation? . . . . . . . . . . . . . . . . . 1445.3 Methods, study two . . . . . . . . . . . . . . . . . . . . . . . 1445.4 Results, study two . . . . . . . . . . . . . . . . . . . . . . . . 1505.4.1 Do these questions index the same underlying construct?1505.4.2 Was the manipulation effective? . . . . . . . . . . . . . 1505.4.3 Did participants condemn the profiteer less for profit-ing on the suffering of the wicked? . . . . . . . . . . . 1525.4.4 Did participants reward the trader for taking a loss inmiracle scenarios? . . . . . . . . . . . . . . . . . . . . 153v5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1566 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1616.1 Near future directions . . . . . . . . . . . . . . . . . . . . . . 1676.2 Distant future directions . . . . . . . . . . . . . . . . . . . . . 169Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172viList of TablesTable 3.1 Summary of parameters . . . . . . . . . . . . . . . . . . . 54Table 4.1 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92Table 4.2 The cues used to establish the Target?s reputation . . . . . 101Table 4.3 A map from dichotomous reputations reputation changesto their continuous interpretations . . . . . . . . . . . . . . 108Table 4.4 Correlations between our four measures of participants?reputational judgments . . . . . . . . . . . . . . . . . . . . 113Table 4.5 Summary of key results for Chapter 4 . . . . . . . . . . . . 118Table 4.6 Frequencies of discretely categorised judgements of Targets 124Table 4.7 Frequencies of discretely categorised judgements of Coop-erators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125Table 4.8 Frequencies of discretely categorised judgements of Defectors126Table 4.9 A discrete partition of participants? reputational judgements130Table 5.1 Likelihood-maximising linear regression parameters ? ofratings of the lucky profiteer . . . . . . . . . . . . . . . . . 146Table 5.2 Correlations among study two dependant variables . . . . 150Table 5.3 Participants? ratings of Floret residents in Study 2 . . . . . 152Table 5.4 Participants? ratings of the lucky profiteer in disaster sce-narios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154Table 5.5 Participants? ratings of the lucky profiteer in miracle sce-narios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156viiList of FiguresFigure 3.1 The NIR decision tree . . . . . . . . . . . . . . . . . . . . 35Figure 3.2 NIR-P basins of attraction and equilibrium reputations . 39Figure 3.3 NIR?s logic, testable assumptions and predictions . . . . . 48Figure 3.4 Simulations showing the accuracy of analytic predictionsabout Resident and Invader equilibrium reputations. . . . 61Figure 3.5 Simulations showing the accuracy of analytic predictionsabout equilibrium reputations of three discrete strategies. 64Figure 3.6 Invasability plots for the three discrete strategies in NIR . 69Figure 3.7 Meta-barycentric plots of NIR boundary equilibria . . . . 76Figure 3.8 Selection gradients for three version of NIR . . . . . . . . 84Figure 3.9 Relative volunteering rates for NIR-V and NIR-M . . . . 87Figure 4.1 American Participants? absolute judgements of Targetsand relative judgements of Actors . . . . . . . . . . . . . . 115Figure 4.2 Indian Participants? absolute judgements of Targets andrelative judgements of Actors . . . . . . . . . . . . . . . . 116Figure 5.1 Likelihood maximising estimates of the means of partici-pants? ratings . . . . . . . . . . . . . . . . . . . . . . . . . 145Figure 5.2 Means of participants? ratings of Floret residents . . . . . 151Figure 5.3 Means of participants? ratings of the profiteer . . . . . . . 155viiiAcknowledgmentsFirst and foremost I?d like to thank Joseph Henrich for support, adviceand stimulating discussion throughout my time at the University of BritishColumbia. This would not have been possible without him.I would also like to thank my committee members (Ara Norenzayan,Susan Birch and Michael Doebeli) and the amazingly bright and diversefaculty of the University of British Columbia?s psychology department foradvice, support, collaboration and feedback.I would like to acknowledge and thank the University of British Columbia,and especially the Faculty of Graduate Studies, for funding in the form ofgrants, fellowship and scholarships. I would like to thank the Departmentof Psychology for several travel grants and a terrific education.I would especially like to thank Faye d?Eon-Eggertson for all her care andsupport, and for her careful proofreading. Your reading experience wouldbe far less polished and far more jarring were it not for her heroic efforts.Finally, I would like to thank my friends, family and loved ones for theirsupport during these interesting times.ixDedicationI dedicate this work to my brother (Mark) and sister (Georgina). I?ve beenso far away from them and missed so much of their lives to undertake thisapprenticeship into the academy.They?re amazing people and I miss them.xChapter 1Introduction?What am I??Like ?why is the sky blue?? and ?why is my puppy sick??, it is an entirelyreasonable questions for a curious child to ask. But unlike those others, itis not a question we have a good answer to.Our understanding of modern humans lacks a central explanatory paradigm.We know of no simple set of laws or principles which integrates and makessense of the staggering range of things people do, make, say, think andbecome. Much of our understanding of humans is, by Thomas Khun?s1definition, pre-scientific.The explanations offered by sociologists?who focus on the institutionsthat emerge from our interactions?are entirely disconnected and often con-tradict those offered by social psychologists?who focus on how we, as in-dividuals, are influenced by our social institutions. Economists? consistentpreference optimising rational agents have little or nothing to do with his-torians? tension between Marxist or internalist interpretations of the causesof societal change. Anthropologists? relativism sits at stark odds with therobust universalism implied by psychologists? broad generalisations from un-dergraduate student samples. A psychologist studying crime who stumblesinto a criminology conference would be just as perplexed by symbolic interac-1a twentieth-century philosopher of science, who?s descriptions of scientific progress(Kuhn, 1996) still inform many scientists? understanding1tionism as a criminologist would be, were their situations reversed, by subtledistinctions between attitudes, attributions, appraisals, attachments, acti-vations, arousals, affects, automatic processes, beliefs, bias, construals, cog-nitions, dissonances, deindividuations, emotions, ego-threats, perceptions,representations, self-verbs, and the panoply other functionally-defined enti-ties psychologists suppose occupy human minds.It is rarer still for any of these disciplines? explanations to explicitlyconnect with the principles of biology. Even though we know that humansevolved by natural selection, most disciplines proceed as though humansplay by entirely different rules from other animals, without ever explicitlyaccounting for why or how this came to be.Even within psychology the theoretical picture is little more integrated.Cognitive psychologists, personality psychologists, developmentalists and so-cial psychologists are engaged in conversations that rarely overlap. Whenthey do it is by the idiosyncratic connections made by individual researchers,rather than due to the a priori, formally derived predictions of a shared cen-tral theory of psychology. Rather than systematically testing and buildinga shared paradigm, most psychological research consists of thorough butdisconnected local descriptions of particular phenomena and exciting noveleffects.Within social psychology in particular, the theoretical landscape is par-ticularly cacophonous. Many disconnected mini-theories cluster closely aroundthe empirical effects they describe and the methods used to investigatethem. Though individual researchers strive to make connections betweenthese mini-theories, these connections are typically ad hoc and rely on re-searchers? intuitive and common-sense understandings of their explanatoryconcepts rather than precise and formal definitions.When asked ?what am I??, we can offer many reasonable answers to littlequestions. Why do we sometimes help each other and sometimes not? Whyare we sometimes willing to spend more money than other times? Why dowe forget things when we?re old? Why did Rome fall? Why do people fromthat other country act so strangely?However we lack a shared set of central explanatory principles that could21. say how something like us can emerge from biology2. predict a priori what we would be like; individually, collectively andhistorically3. begin to integrate (and, sometimes, outright reject) the many verydifferent, sometimes contradictory explanations for different aspectsof our behaviour.This absence of core theoretical principles might be a malfunction of thesociology of our science, a consequence of how we incentivise and motivatesocial scientists. Walter Miscel, then president of the Association for Psy-chological Science, implied as much when he turned his peers? attention asthe ?toothbrush problem? (Mischel, 2009):Psychologists treat other peoples? theories like toothbrushes, noself-respecting person wants to use anyone elses.It is also possible that our scientific institutions (including those thatmotivate researchers to invest primarily in their own novel, distinct micro-theories) are merely shaping themselves to the contours of our object ofstudy. Central explanatory principles of the human phenotype may be per-versely difficult to discover, even with the full arsenal of modern empiricism,or worse, there may not be any principles to find.It is entirely conceivable that no such central explanatory principles ex-ist. Perhaps the best answer we will ever be able to give to ?What am I??is a scattering of local descriptions of disparate human-phenomena. Onefor cognitive dissonance. Another, that invokes very different theoreticalassumptions, for demographic transitions. A third for extraversion, andso on. In such a world, our current sociology of isolated, idiosyncraticallyconnected inquiry might be ideal.I believe that, difficult though they may be to untangle, such principlesdo exist and are worth searching for. The argument for their existence issimple and I find it persuasive. Our species evolved by natural selection.3It did so very recently (on evolutionary timescales) and in a sudden, un-precedented explosion of complex novel behaviour. What?s more, no otherspecies did the same. This suggests that, though some aspects of humanpsychology may have accreted independently over millennia, there were alsosome key recent ingredients that sparked an inferno of positive feedback andnew dynamics. If there were simple, discrete causes then chances are we canfind a simple explanatory principles that reconstruct them and predict theirmanifold consequences.The value of searching for these principles, let alone the means of doingit, are harder to argue for. One could make a good case that our cur-rent hodge-podge of cross-disciplinary inquiry is the best way of discoveringthem. By careful, if initially disconnected, empiricism we might graduallyestablish a solid foundation of accurate descriptions of human-phenomena.This would accumulate, bottom-up, until they converged on the central ex-planatory principles. The counter-point to this argument is that our domainof inquiry?human behaviour?may be complex enough that this process isnon-terminal. We could go on making empirical discoveries about humansforever. Or rather, like the proverbial blind men and their elephant2, wewould merely rediscover too-simple, low-dimensional redescriptions of thesame underlying higher-dimensional phenomena, cloaked in different, incom-patible theoretical jargon (for a philosophical explication of this possibilitysee Dennett, 1991).For instance, one regularity that will be relevant later in this disserta-tion is ?negativity bias?. People seem to have a robust tendency to respondmore strongly to subjectively bad or unpleasant events than good ones. In2001, Roy Baumeister and his colleagues documented a terrific diversity ofdomains and investigations in which this same pattern had been rediscov-ered (Baumeister et al., 2001). In each domain?from emotion research, tomorality, to the trajectory romantic relationships, to economic choices aboutpecuniary profit and loss?it was interpreted as a novel insight, a local anddistinct piece of evidence for a distinct theory, couched in distinct theoreti-2if you?re unfamiliar with the proverb, it can be readily found on the internet4cal terms and entailed by different assumptions. Identifying these seeminglydisconnected effects as facets of a single pattern was a great achievement.Even though negativity bias is a very simple pattern?a linear relation-ship between the valence of an event and its effect3?it was by no meanssimple or obvious to spot. If some or most of the real patterns that explainhuman behaviour are more complex than that?say non-linear relationships,or interactions between multiple factors, or patterns that spread across cul-tural timescales?then we may well be pessimistic about whether an industryof bottom-up empiricism will ever converge upon them.If such concerns have you worried, you may see the worth of simultane-ously pursuing a top-down strategy. That is, you might consider trying toderive the explanatory patterns a priori. To do so, you would start from ei-ther our understanding of the biological principles that launched our species,or another sufficiently abstract, theoretically rigorous description of humans.You would then postulate as few principles, and as simple principles as youcould and attempt to derive from them the gambit of human-phenomena(individual, collective, psychological, historical, etc.). Successful theorieswould be those that were simple, fit seamlessly with biology, and accuratelypredicted the greatest breadth of human-phenomena, especially previouslyundiscovered phenomena.Such top-down attempts are valuable because, if successful, they canintegrate disconnected empirical efforts and push our understanding forwardin leaps and bounds. But they are also risky. You can invent an almostlimitless diversity of such grand theories, but only a few are accurate oruseful. Decades of valuable researcher time could be squandered fruitlesslypursuing them, at the price of empirical phenomena that could have beenbetter documented in the meantime. This is, perhaps, why such ambitiousand unlikely speculation is often frowned upon by psychologists.However the difficulty of hitting upon an accurate top-down theory is also3Actually, given the null-hypothesis testing inferential framework in which most of thiswork is done, the relationship being formally tested is often magnitude-free and so simplerstill: a dichotomy. More bad means more strong, regardless of how much more of eitheror the specific functional relationship between them.5an asset: a low false-positive rate. Top-down theories are unlikely to broadlypredict novel contemporary empirical phenomena unless they are accurate.By ?broadly?, ?novel? and ?predict? I mean that the theory is consistent with awide range of phenomena that it was not explicitly designed to explain. Themeasure of top-down theories is the breadth of their explanatory power, theirability to easily integrate phenomena that previously seemed disconnected.Relax. I will not propose a grand top-down theory in this dissertation.Instead, I hope to contribute to a broader paradigm of already successfultop-down explanation. I will refer to this paradigm, which I review below, as?culture-gene coevolution?, even though scholars from many disciplines haveconverged on this same theoretical space and called it by different names.Specifically, I hope to contribute by refining the theoretical terrain on whichculture-gene coevolutionary theories contend.In chapter 2, I spotlight a set of top-down explanations of human be-haviour that share these traits: 1) they are grounded in biology, and 2)they give a central role to ?culture??the social transmission of complex, en-coded, phenotype-shaping information. I argue that though these attemptsare already somewhat successful and very exciting, they are plagued by anunder-recognised theoretical challenge: the cooperative dilemma of culture.I review existing solutions to this dilemma.In the remainder of this dissertation I derive and test a novel solution tothe cooperative dilemma of culture.In chapter 3, I pick out one of the most plausible and commonly citedsocio-ecological dynamics that may have established the cooperative foun-dation on which ancestral culture thrived: reputation. I argue that exist-ing formal models of reputation do not address the cooperative dilemmaof culture and offer a new model that does. This new model, which I callNegative Indirect Reciprocity (NIR), shifts the theoretical focus from themutual-aid emphasised in most approaches to human cooperation (positivecooperation), to our countless unrealised opportunities to gainfully exploitone another (negative cooperation).Many scholars have assumed that positive and negative cooperation aretwo sides of the same coin, and are equally well represented by our existing6models of positive cooperation. In chapter 3 argue that they are not, detailsome of the important asymmetries between them and argue that negativecooperative dilemmas likely shaped our cultural species? early evolution morestrongly than positive ones did.Succinctly, NIR is a model of ?indirect reciprocity??reputation-basedcooperation. Unlike existing models, NIR minimises the cognitive prerequi-sites of reputation-based cooperation by assuming that cooperation or de-fection by ?inaction? (i.e., having an opportunity to act, but choosing not to)does not change reputations. That is, it does not assume that communitiesare able to coordinate their understanding of abstract opportunities, roles,responsibilities and so on. NIR show that, in the context of cooperation by?not exploiting?, such simple reputational systems can form the foundationof more complex societies.?Negative cooperation? and ?reputation without meaningful inaction? havea special synergy. Together they create selective pressure for individuals todo whatever it takes to improve their reputation?that is, to please theirpeers on average. This pressure to attend to one?s community?s behaviouralexpectations can help explain how more sophisticated forms of cooperationarose.NIR tells a specific, theoretically-rigorous story of how human reputa-tions may have first emerged and how they formed the substrate of otherforms of cooperation. It entails several clear empirical predictions aboutcontemporary humans psychology. In the remaining chapters I test some ofthese predictions.In chapter 4, I attempt to provide the first theoretically-informed cata-logue of people?s reputation-based intuitions by simply asking them. I findthat the valence of cooperation (whether it is helping or exploiting) mat-ters a lot (as predicted by NIR) and that negative non-cooperation causesparticularly strong reactions across cultures (as predicted by NIR).Chapter 5 documents a study that, while simple, is important because ittests NIR against a novel theoretical phenomenon that it was not designed toexplain. I find that this phenomenon, of people disliking those who inciden-tally profit while others suffer, fits quite nicely to the psychological biases7that NIR predicts we should have.8Chapter 2The cooperative dilemma ofan emerging cultural speciesSomewhere between our split from chimpanzees (circa 6-3 mya, Pattersonet al., 2006, but cf. Yamamichi et al., 2012) and the emergence of fullyanatomically modern humans (circa 200 kya, Day, 1969; McDougall et al.,2005) our ancestors began behaving very strangely. Their new ways of think-ing and behaving, and the social dynamics these engendered, were ultimatelyrooted in the same (relatively) well understood processes that shape the be-haviour of our many non-human relatives: evolution by natural section. Yetsomehow this species experience an unprecedented explosion of complex anddiverse behavioural forms like opera, foot-binding and hopscotch. On theface of it, it seems as though some new dynamics began interacting withnatural selection.Psychologists and other social scientists work to understand the stag-gering behavioural and cognitive diversity that has emerged in our species.For instance, cultural psychologists document deep cognitive differences be-tween humans raised among different communities (e.g., Nisbet, 2003), evo-lutionary psychologists try to identify behavioural similarities which can beexplained as adaptations to a shared ancestral environment (e.g., Barkowet al., 1992), while social psychologists tend to treat social environments asextrinsic and investigate how individuals respond to them, often implicitly9assuming that the breath of diversity is well represented by North Americanuniversity undergraduates (Henrich et al., 2010).Alongside this industry of bottom-up inquiry, some scholars also con-front the top-down challenge of disentangling the circumstances that firstlaunched this process and the core laws or principles that continue to driveit forward. Scholars from across disciplines have proposed mechanism anddynamics at the core of humans? distinct evolutionary trajectory, includingarchaeologists (Mithen, 1996), anthropologists (Boyd & Richerson, 2005;Deacon, 1998; Richerson & Boyd, 2004), primatologists (Tomasello, 1999)psychologists (Csibra & Gergely, 2009; Tomasello, 1999), biologists (La-land, 2004; Laland et al., 2001; Wilson, 2012), mathematicians (Nowak &Sigmund, 2005), linguists (Pinker, 2010) and philosophers (Sterelny, 2012),among others.Refining the glut of theory is not easy. Early human societies no longerexist so we cannot directly observe their behaviour, experiment on their cog-nition nor watch them change. We cannot directly measure the dynamicsthat gave rise to our species. However, we may have some chance of recon-structing them from the convergence of several indirect sources of evidence.Close inspection of our genome offers hints of ancestral speciation events(Garrigan & Hammer, 2006) and recent rates of genetic selection (Lalandet al., 2010). However much variability among contemporary humans seemsto be cultural, not genetic (Bell et al., 2009; Cavalli-Sforza & Cavalli-Sforza,2000). Archaeological remains help fill this gap, but are limited to materialsthat survive decomposition and more sparse the further into the past wewish to gaze. Cross-species comparisons with our nearest cousins (e.g., Deanet al., 2012; Herrmann et al., 2007, 2010) provide a source of direct and evenexperimental evidence about evolved behavioural and cognitive differences.But while contemporary primates share our ancestors, they are not ourancestors and have experienced several million years of distinct evolutionarypressures.Observation of contemporary forager societies can help here (e.g., Bellet al., 2009; Henrich et al., 2006), since their ecology, group size and residencepatterns may be more similar to our ancestors?. However the mere existence10of contemporary foragers is reason to suspect their societies are unlike ourearliest ancestors?. Contemporary foragers somehow escaped the explosivespread of agriculture (Gignoux et al., 2011), while typical foragers did not.A final empirical avenue for testing these theories is to carefully derivetheir implications for contemporary cognition, and compare these to howpeople think and behave today. This is the method I pursue in this disser-tation.We can also make purely theoretical progress by narrowing the windowof plausible theories. One way to do this is to reject willy-nilly post-hocexplanations of particular empirical phenomena (e.g., Frankenhuis, 2010;Navarrete & Fessler, 2005). Another way is to notice and define theoreticalpuzzles (e.g., Rogers, 1988) whose solutions (e.g., Boyd & Richerson, 2005;Enquist et al., 2008; Laland, 2004) push our theories towards assumptionsmore likely to be true. This dissertation highlights one such puzzle, thecooperative dilemma of culture, for a subset the of these hypotheses: culture-gene coevolutionary theories.Culture-gene coevolutionary theories explain humans by emphasising cu-mulative cultural learning. When they say ?culture? these theories are re-ferring to behaviour-shaping information that is transmitted across genera-tions socially, not genetically. While many species show evidence of somecultural transmission (Brown & Laland, 2003; Laland, 2004; Rendell et al.,2010; Whitehead et al., 2004; Whiten et al., 1999), only among ours did thiscultural information begin accumulating into ever more complex forms, thateventually included ?calculus?, ?maps? and ?dancing the Macarena?.The revolutionary importance of this transition is worth spending a fewparagraphs on. To give it some more intuitive, less jargony traction, I?llborrow a metaphor from cognitive psychology. Our minds, and other an-imals?, are information processing systems; software instantiated upon thehardware of our brains. They are computers designed to take sensory input,interpret it, make some sense of the world outside, and make decisions abouthow to behave. To do this, they must make assumptions about that world.For instance we can, in a sense, directly perceive some of the assump-tions our visual system makes when we look at a visual illusion and see11something we know is not really there. Developmental scientists have at-tempted to identify other, more conceptual assumptions that children useto make sense of the world. Proposed assumptions range from intuitionsabout the existence and properties of objects, the geometry of space, the ex-istence and behaviour of goal-directed agents and even mathematical princi-ples (see Spelke & Kinzler, 2006, for a review). They even include complexintuitions about the relative importance different kinds of information (e.g.,Barrett & Broesch, 2012), the existence of ?minds? that are distinct frombodies (Chudek et al., forthcoming) and something functionally equivalentto Bayesian priors about the nature of causality (Griffiths et al., 2011; Kalishet al., 2007; Yeung & Griffiths, 2011).Many of the assumptions we use to make sense of the world, like thosethat process vision, are encoded genetically. Our minds also seem to be ableto infer other assumptions as they develop. For instance, rats rapidly makestrong assumptions about which olfactory cues signal toxic food merely byobserving other rats? reactions (Galef & Whiskin, 2008). Though we don?tknow whether rats explicitly represent this information in the same wayhumans do, we can observe their minds translating sensory information intobehaviour as though they were assuming this of the outside world.Though our minds (and rats?) can make some sense of the world beyondthat encoded in our genes, the accuracy and complexity of these represen-tations is limited by the scant sensory experience we are exposed to duringour short lives. The advent of cumulative cultural learning was a revolutionin the potential complexity of these representations. It let our minds alsotap into the experiences of our conspecifics. In addition to learning aboutthe world from our own experience and by observing others, we began totap into the accumulated wisdom of minds long dead.Imagine all the different assumptions our minds could make about theoutside world as a vast space that stretches off in as many directions asyou can conceive. Before the revolution of cumulative culture our individualjourneys through this space were like sparks flying off of the slow-burningfuse of genetic evolution. Even today this fuse continues to endlessly wend itsway through the space of possible external realities, guided by the selective12retention of more successful variants. Like other animals, our brains wereconstrained to perceive the world in whichever ways had led their ancestorsto be more robust, fecund or in other ways fitter.With the advent of cumulative culture something entirely new happened.Our ancestors detached from the fuse. They began incorporating others?representations of the outside world into their own, building on others? as-sumptions, sparking into the unknown from a new position far from theirgenetically imbued starting point. We are still on this journey together.We each aggregate the experiences, knowledge and assumptions of our peersand each new generation inherits this amalgam. To continue the metaphor,our cultures are like great fireballs travelling through the space of ?possi-ble assumptions our minds could make about the world outside?. We blazemost brightly around an evolving core of shared assumptions, practices andunderstandings. From there our individual sparks still fly in all directions,innovating their own unique sense of the world outside.Culture-gene coevolutionists work to understand this phenomenon ofcumulative cultural learning; to theoretically and empirically describe itand discover any laws or principles that govern it. The questions they askinclude:? How did this process start, and why did it seemingly happen only inour phylogenetic line (e.g., Boyd & Richerson, 1988, 1996; Richerson& Boyd, 2004)?? What cognitive adaptations and behavioural traits make it possible,and what new cognitive adaptations does it favour (e.g., Boyd & Rich-erson, 2005; Henrich, 2009; Henrich & Gil-White, 2001; Laland, 2004;Nakahashi et al., 2012)?? Are there rules or principles that govern the trajectories such confla-grations of cultural learning will take through the space of all possibleculture (e.g., Boyd & Richerson, 1988; Cavalli-Sforza & Feldman, 1981;Laland et al., 2001; Richerson & Boyd, 2004)?? Can our evolving cultural corpus cause us to fracture into distinct,13symbolically marked ethnicities or social classes (e.g., Henrich & Boyd,2008; McElreath et al., 2003)?? How and why does this process produce cross-generational institutions,like religion (e.g., Gervais et al., 2011; Norenzayan & Shariff, 2008) ormarriage (e.g., Henrich et al., 2012), and how do these evolve?? How does cultural evolution redirect genetic evolution (e.g., Lalandet al., 2010); how does it reshape our ecology and adaptive landscape(e.g., O?Brien & Laland, 2012; Rendell et al., 2011)?? Does this process ever break down, how and why (e.g., Henrich, 2004)?? How does this our evolving culture interface with our capacity forcooperation (e.g., Boyd et al., 2011a; Chudek & Henrich, 2011, andthis dissertation)?This culture-gene coevolutionary explanatory story is rapidly maturing,and has already had considerable success predicting many emergent featuresof contemporary society and psychology. However there is something trou-bling about it. Why did it make sense?from the perspective of our genes?for our genome to give up so much control over our adult phenotype, andlet cultural information shape us instead?2.1 The cooperative dilemma of culture (a.k.a.the evil teacher problem)The cooperative dilemma of culture, which I also like to call the ?evil teacherproblem?, is pervasive. One need not agree with any of the specifics of anyparticular CGC theory to be trouble by it. It should concern you whateveryour stance in controversies over group selection (e.g., Abbot et al., 2011;Nowak et al., 2010; West et al., 2011) and other evolutionary processes. Tomeet it, you merely need to accept the following premises:1. Humans are cultural To understand us you need to explain the corpusof technology, ideas and knowledge that we have accumulated overgenerations, non-genetically.142. Humans are strangely cooperative To understand us you need toexplain how we became so cooperative. We regularly make choices thatentail a relative cost (or forfeiture of a possible benefit) for us and rela-tive benefit for others. Those others regularly include non-kin, peoplewe know distantly or only by reputation, and even complete strangerswho we are unlikely to meet again. The ways we cooperate are oftenunique to particular societies, and in any case vary dramatically acrossshort stretches of time and space. No other species cooperates like wedo (for detailed arguments for the peculiarity of human cooperationsee Chudek et al., 2013b). Though many species cooperate in theirown unique ways, human cooperation is distinct in its variety and rateof change. Humans in some places build houses together, in othersthey queue, and in others they construct and police far flung tradenetworks whose purpose is taking other humans as slaves. They alsocooperate on vastly different scales. In some regions, villages are per-manently at war with each other. In others, huge empires maintainrelative internal harmony. Furthermore the rates at which the scaleand form of human cooperation changes?some regions have movedfrom villages to nations within a single lifetime?is unprecedented andcannot be explained by genetic mutation and natural selection alone.3. Culture requires cooperation For a species to accumulate a corpusof complex cultural knowledge, its members must accurately sharevaluable, fitness-relevant, phenotype-shaping information. Each indi-viduals acquires this information from the minds of others, who couldeasily distort it to their advantage or keep it to themselves but, typi-cally, do not. When we share cultural knowledge, we cooperate. Thisis especially true for the complex, hard-to-observe ecological masterythat allows small scale foragers to thrive in some of the worlds harshestecologies.4. Explanations of human cooperation invoke culture Culture typ-ically changes far more quickly than genomes do. Any mechanismthat could keep us cooperatively sharing valuable cultural knowledge,15across its staggering breadth of domains, would need to keep up withculture. For example, a set of genetically adapted intuitions that man-aged to keep us honestly sharing food would do us little good whenwe began culturally transmitting warfare techniques, inheritance sys-tems, queuing etiquette or intellectual property laws. Models of pow-erful, domain-general, human-specific cooperation-sustaining mecha-nisms do exist; I discuss some below and in Chapter 3. However thesemechanisms typically presuppose that social groups are able to rapidlyconverge on and even enforce behavioural norms. That is, they pre-suppose that we are already a fairly cultural species.5. There is an explanatory regress Accounts of human emergence facethe challenge of explaining the origins of human culture and coopera-tion simultaneously, without assuming each to explain the other.I will assume you agree with the first two claims, and will try to convinceyou of the others.2.1.1 Why does culture require cooperation?I am aware of three interrelated cooperative challenges posed by the emer-gence of human culture.The first challenge: for complex culture (e.g., behavioural patterns no a-cultural individual could plausibly devise alone, like ?trebuchets? and ?opera?)to accumulate, some individuals must share valuable information with otherswhen they could do nothing. The second: those same individuals must notdistort the cultural information to their own benefit, even though othersmust (to some degree) trust them. The third, language makes lying cheap.For a clearer sense of these three arguments, lets imagine two individuals:a learner and a teacher. They are unrelated and the teacher has someinformation that would be useful to the learner. Each of us plays both theseroles in our lives, sometimes both simultaneously.At very early stages of the emergence of culture, when only simple be-haviours are transmitted, it is plausible that the teacher has no choice but16to transmit their information. For example, imagine the information is aforaging technique. Unless the teacher is willing to not forage at all, theymight have no choice but to demonstrate their technique to the onlookinglearner. As long as cultural information is simple enough to be transmittedby mere looking-on, learners are in a position of power. However as soon asculture becomes complex enough that it requires any deliberate demonstra-tion of a technique (e.g., stone-tool making), or a process must be performedprecisely and rarely (e.g., making tools for making tools, house building), orcan be done out of view of other (e.g., food processing) or requires specialisedknowledge (e.g., tracking, foraging, medicinal plant use), the power changeshands. As soon as culture gets really interesting, interesting enough thatsome specialised teaching is required, teachers must cooperatively transmitvaluable information to unrelated others.I have heard two counter-arguments to this first thesis.The first is that perhaps such teaching could begin within the family,piggy-backing on kin-recognition mechanisms for cooperation. I have seenno formal models of such a process, and whether it could allow enoughhorizontal transmission for cultural accumulation to ever get off the ground.I do not think such models are necessary for two reasons:? Humans don?t just behave that way. We do not show a widespreadproclivity for limiting our cultural transmission to family members(e.g., Henrich & Broesch, 2011) and certain eye- and body-cues seemto put children into a ?pedagogical stance? where they are highly cred-ulous to information transmitted by any adult, even stranger-scientistsstudying them (Csibra & Gergely, 2009).? Even if cultural transmission started within families, the cooperativedilemma is merely pushed back (or forward in time) to whenever hu-mans started sharing enough of this information with non-kin for ourcomplex societies to begin developing (Chudek et al., 2013b).The second counter-argument is usually voiced after people reflect onwhy they themselves are not evil teachers. A selfish teacher would soon17find that others refused to teach them too. The long-term costs of others?exclusion would be too great. More formally, this intuition suggests thatdirect, pairwise reciprocity might have solved the cooperative dilemma ofculture. Below I argue that it cannot (section 2.1.2).The dilemma runs deeper. Imagine that the learner and teacher passthe first hurdle and the teacher begins sharing valuable information withthe learner. This puts the learner in a very vulnerable position. While cul-tural information is simple enough that the learner can readily apprehendthe causal logic of what they are doing, they need not ?trust? the teacherin any fitness-relevant sense. However as soon as the corpus accumulatesinformation complex enough that it is causally opaque to us (which, I sus-pect, need not be very complex at all), the learner must trust. That is,adaptations must enter their genome which cause it to give up some controlover its adult phenotype. Instead, their behaviour starts to be shaped byinformation that has passed through others? minds. This information mustbe, on average, adaptive for the learner or such adaptations would not befavoured.The second challenge: complex cultural accumulation implies that somecontrol over the learner?s behaviour is in the hands of the teacher. Thisshould select for exploitative adaptations in the teacher. Teachers who findways to mutate and distort cultural information to their own advantageshould be favoured by selection.Such evil teachers could, for instance, teach foraging techniques or di-etary prohibitions that result in the teacher eating some of the learner?sfood. While I have occasionally seen older children teaching younger chil-dren rules unabashedly biased in their interests, the endless possibilities forsuch Machiavellian manipulation of cultural information do not even occurto most adults. The more the first challenge is overcome, the more this sec-ond challenge is exacerbated. The more valuable information we share, themore culture-dependent learners come trust this causally-opaque informa-tion, the more opportunities there are for evil teachers to distort it to theirbenefit. The importance of this second challenge is underscored by just howtrusting contemporary children are of non-kin adults (Harris, 2012).18The most common answers I have heard to this challenge are that ma-nipulative teaching would result in a) retaliation and b) the teacher beingignored in the future. These are ways of restating the dilemma rather thansolutions. How do learners know when to retaliate? Say I teach you a medic-inal remedy but omit a scarce ingredient so I can gather more myself. Howare you to know that your slower recovery times are not a consequence ofsome other aspect of your lifestyle, your poorer skill at preparing medicinesor the ill will you earned of a sorcerer? How can I infer that you snare thebig game more often than me because you deliberately taught me inferiortechniques, and not due to the countless other differences between us?If retaliation is error-prone (sometimes you weren?t deceiving me) howis it not more costly for the learner than blind trust or walking away? Whydo the benefits of exploitation for the evil teacher not exceed the costs ofpotential retaliation or being ignored? Answering these questions requiresspecific hypotheses about ancestral socio-ecological dynamics and how theretaliation or ignoring affected them. It requires detailed models of howthese ecologies interacted with genetic and cultural evolution. It requires asolution to the cooperative dilemma of culture.The third challenge of the cooperative dilemma of culture: languagemakes lying cheap. One plausible way to meet the first two challenges is toargue that it is more costly for the teacher to withhold information, or figureout how to distort it exploitatively, than the benefits they could obtain bydoing so. This argument fails, and the first two challenges are exacerbated,when language enters the picture. At some point in the history of ourcultural corpus, it became encoded it in a semantically rich, combinatoriallanguage.I find persuasive the argument that given what a complex, specialisedadaptation language is, there must have existed a strong need for it (i.e., acomplex cultural corpus worth transmitting) before it was selected for. Re-gardless of your stance on this issue when language entered the picture, itbecame possible to transmit far more complex, subtle cultural information,but also to engage in subtler, more complex deceptions. There are mathe-matical models demonstrating that language makes lying powerful and easy19(Lachmann & Bergstrom, 2004), but I suspect that this evolutionary argu-ment is intuitive, simple and uncontroversial enough that it does not requirethem. When you have language, it is easy to lie.Hopefully these arguments convince you that only a cooperative speciescould be cultural. But how do we know that the cooperative chicken didnot preceded the cultural egg?2.1.2 Why does (human-like) cooperation require culture?Many animals cooperate. For instance, some eusocial insects act as a singlesuper organism by sequestering their reproductive line, just like the cellsin our bodies do (Smith & Szathm?ry, 1997). On a smaller scale, somesocial mammals provision one another with public goods, such as makingself-endangering alarm-calls upon sighting a predator (Seyfarth et al., 1980).However human culture could not have piggy-backed on these simplerforms of cooperation. First, humans are not eusocial, we are distantly re-lated and each reproduce individually. Second, specific genetic adaptationsfor specific cooperative behaviours in specific domains could not sustain cul-ture. Culture changes much faster than genes do. As culture accumulates,cooperative dilemmas arise rapidly in many different domains. Who shouldtake the risks when hunting? How should we divide the spoils? Who shoulddo the hard-labour and who should pray for rain? Who marries whom, whoreproduces when and who contributes which resource to which resource-hungry child, during their long, unproductive juvenile period (Gurven et al.,2006; Kaplan & Robson, 2002; Lancaster et al., 2000; Walker et al., 2006,2002)? By the time our slowly mutating genome develops a mechanism (e.g.,a signal; Boyd et al., 2010) for solving one of these dilemmas, our rapidlychanging cultural corpus has generated countless others.If ancestral humans had genetic adaptations for sustaining cooperationthat supported their cultural accumulation, they must have been domain-general adaptations that could rapidly stabilise arbitrary, new forms of co-operation whenever they arose. I am aware of no purely-genetic mechanismsfor sustaining culture (e.g., kin-selection, limited dispersal, genetic group-20selection, etc.) that can satisfy this criterion.A second class of mechanisms posit institutions for sustaining culture.That is, highly coordinated, sometimes socially enforced, behaviours thatensure that it is in individuals? interest to cooperate in arbitrary domains.A simple example is a police force or ?punishment pool? (Sigmund et al.,2010). Individuals first put resources towards sustaining the existence of animpartial punisher (e.g., paying the sheriff) and then the punisher spendsthose resources on making defectors pay. These institutional mechanismscan ?keep up with culture?. They can sustain arbitrary forms of cooperationwithout requiring slow, specific genetic adaptations. However most insti-tutional mechanisms presuppose a species that is able to dynamically andrapidly coordinate such institutions. For instance, how do people establishpunishment pools? How do they ensure the pool?s resources do effectivelypunish? How do they coordinate what specific behaviours the pool shouldpunish. How do all these moving parts keep up with our rapidly evolvingculture?Even if such institutional mechanisms did not require language and so-phisticated, culturally transmitted concepts, they would at a minimum re-quire that we carefully attended to and trustingly imitated one-another?sbehaviour. While institutional mechanisms undoubtedly played and con-tinue to play a role in sustaining and managing the latter forms of humansociality, they cannot be invoked to explain the emergence of culture withoutstumbling into an explanatory regress.A third class of models posits genetically-selected, individual-level be-havioural strategies that, in aggregate when common-enough, sustain coop-eration in arbitrary domains. The flagship of such promising solutions isreciprocity.?Direct reciprocity? explores how individuals? can benefit by conditioningtheir behaviour towards someone on how that someone treated them in thepast. If enough individuals act this way, pairwise cooperation can thrive(Axelrod & Hamilton, 1981; Boyd & Lorberbaum, 1987; Doebeli & Hauert,2005; Trivers, 1971; van Veelen et al., 2012). However direct reciprocitystruggles to explain cooperation among many individuals simultaneously,21such as provisioning non-excludable goods to an entire group (e.g., sharingvaluable cultural information).?Indirect reciprocity? (Leimar & Hammerstein, 2001; Ohtsuki et al., 2006;Panchanathan & Boyd, 2004; Panchanathan et al., 2003; Sigmund, 2012) orreputation-based cooperation is a better candidate for catalysing the emer-gence of human culture. Models of indirect reciprocity (discussed furtherin chapter 3) explore how individuals can benefit by conditioning their be-haviour towards someone on how that someone treated a third person inthe past. Though most models of indirect reciprocity only consider pairwisecooperation, it is easy to see how they can be extended. Once individualshave reputations, and once those reputations coordinate how others treatthem, selection will favour individuals who do whatever it takes (as longas it is not too costly) to improve their reputation. If provisioning pub-lic goods (e.g., freely sharing knowledge) improves one?s reputation, thenindirect reciprocity can sustain that too.However for indirect reciprocity to hold water as a solution to the coop-erative dilemma of culture, one would need to demonstrate that it can getoff the ground without presupposing the existence of a cultural species. Inchapter three I argue that existing models do make this supposition and sofall prey to the explanatory regress we are trying to avoid. I then go on toprovide an alternative that does not.In short, I am not aware of any existing theories (or formal, evolution-ary models) that show how the kind of domain-arbitrary, rapidly adaptingcooperation necessary for culture could emerge with assuming sophisticatedcultural pre-adaptations.If the cooperative dilemma of culture is a real puzzle of CGC theories,then how can we solve it?2.2 Three kinds of solutionsI am aware of three broad approaches to untangling the cooperative dilemmaof culture.222.2.1 Route 1: The big stepThe first approach supposes that human culture was a fortuitous accident,the convergence of several other mature preadaptations. My impressionis that this is many non-experts? default model, though it has also beenproposed by careful thinkers (e.g., Pinker, 2010).Humans became bipedal, began using tools, developed language includ-ing specialised, localised cognitive adaptations, became quite intelligent anddeveloped many of the specialised cognitive adaptations psychologists aredocumenting today. All this happened, without any substantial accumula-tion of a cultural corpus (certainly not enough to warrant worrying aboutthe cooperative dilemma), during the five million years or so since we splitfrom other primates. Finally, during the last fifty thousand years or so, amature version of human culture emerged all at once, a consequence of theseother adaptations.In this scenario, there is no cooperative dilemma because culture doesn?tcoevolve slowly with genetics. It comes into existence full-blown after hu-mans are, more or less, anatomically and cognitively modern. We need notpuzzle over how selection favoured genes that promoted reliance on culturesimultaneously to selecting cooperative cultural content, because the geneticadaptations happened first and then culture came second. By the time cul-ture was on the table, humans were intelligent enough to reason their waypast the fitness trough of mutual exploitation to the distant fitness peakof mutual cooperation. This route out of the dilemma relies on us beingtoo smart to culturally exploit one another, smart enough to share culturefreely.I can understand why this is most people?s default argument, and whyit seems intuitive. Emerging evidence suggests that humans are intuitivedualists (Chudek et al., forthcoming), we think about our intentional, ra-tional human minds and our animalistic bodies as different (and potentiallyseparable) kinds of things. We intuitively imagine minds as inhabiting bod-ies. This makes it particularly easy to imagine that our gene-built brainsand bodies came first, and then our culture-shaped minds began inhabiting23them, fully-formed, afterwards.While I believe these accounts do not hold up to close inspection, thisdissertation is not the place for that argument (see Boyd et al., 2011b, for re-cent arguments to this end). Notice only that for such big-step explanationsto be plausible, they must specify a suite of adaptive benefits that carvedout a ?cognitive niche? (e.g., Pinker, 2010) and drove our metabolically costlybrain expansion (Aiello & Wheeler, 1995; Kotrschal et al., 2013), but didnot do so for other species in our planet?s long history. They must also ex-plain what favoured our fantastic penchant for language, which children notexposed to a language-community seemingly develop spontaneously (Sen-ghas et al., 2004), without invoking its advantages for the transmission ofcomplex, encoded information.Explanations that put culture first can do this parsimoniously?the adap-tive value of our accumulated cultural knowledge selects for brains ever bet-ter at accessing and using it. If we prefer this parsimony, we must pay forit by confronting the evil teacher problem.2.2.2 Route 2: The arms raceAn alternative is that, though teachers have always been incentivised to ex-ploit, learners have evolved cognitive counter-measures, ways of sifting thecultural wheat from the deceptive chaff. On this account, there is an evolu-tionary arms race between evil-teachers, whom evolution is shaping to sendbiased information, and sceptical learners. Learners may avoid exploitativecultural information by, for instance, preferring information that is backedby action (Henrich, 2009), averaging information between models (Boyd &Richerson, 1988) or imitating others? model choices (Henrich & Gil-White,2001).I find this solution more plausible but have several concerns. First, evenif learners are not manipulable, it is not obvious why teachers would trans-mit cultural information at all (though plausible solutions do exist, such asthe possibility that valued teachers accrue social benefits, termed deference;Henrich & Gil-White, 2001). Solving the second challenge merely exacer-24bates the first. It may be possible that a run-away evolutionary processcould navigate this trade-off. Learners have an advantage, then teachersdo, then learners, and so on. Cognitive adaptations for culturally exploitingand avoiding exploitation escalate quickly enough that it is always worthtransmitting ever more valuable cultural information. However I have notyet seen (or developed) any clear models of whether and how such a processcould work.On the empirical front, if our were species were situated at the cuttingend of such an arms-race, we ought to be suspicious learners and shrewdlymanipulative teachers. Instead, when it comes to sharing cultural informa-tion, we are both credulous and honest. We happily send our children toschools where they diligently learn from often under-paid, selfless strangers.Children (Lyons et al., 2007; Whiten et al., 2009) and even adults (McGuiganet al., 2011) dutifully follow the strange and seemingly pointless instruc-tions of strangers, even strangers who are patently from exotic out-groups(Nielsen & Tomaselli, 2010). Children even go out of their way to enforcerules learned from strangers on others (Rakoczy et al., 2009). Infants delib-erately make patently incorrect choices when strangers suggest them witheven very subtle pedagogical cues (Top?l et al., 2008). We even find it hardto distinguish strangers? subtle suggestions from our own memories (Loftus& Palmer, 1974), and infants trust knowledgeable strangers over their ownmothers (Stenberg, 2009).These high levels of trust contrast starkly with children?s impressiveselectivity in choosing between potential social models. They prefer, forinstance, to learn from their more confident (Birch et al., 2009), previouslyknowledgeable (Birch et al., 2008), self-similar (Buttelmann et al., 2012)and prestigious (Chudek et al., 2012) peers (see Chudek et al., 2013a, for arecent review). This suite of learning behaviour is more likely a product of aselective environment where communities shared knowledge freely, honestlyand cooperatively children?s dilemma was distinguishing the higher qualitymodel, than an arms race between Machiavellian deception and lie-detecting.252.2.3 Route 3: The ratchetA third possibility is that some co-evolutionary process ratcheted up humancooperation and culture simultaneously. This ratchet would start from smallamounts of culture, little and simple enough that its transmission wouldnot generate a cooperative dilemma. This early culture would generateevolutionary dynamics that sustained cooperation in the ways required tosustain more complex culture, which would sustain more cooperation andso on.There are many possible ways that such culture-cooperation ratchetscould have worked. For instance, one could posit that teaching began pri-marily among kin (Fogarty et al., 2011), allowing enough cultural sophistica-tion to accumulate that complex cooperation sustaining mechanisms couldcome into play and allow broader information sharing. Or perhaps earlyteaching was promoted by its utility in improving the return of mutualisticendeavours like hunting (see Sterelny, 2012, for one such account).Alternatively, culture could have given rise to new strategic niches (e.g.,prestigious leaders Henrich & Gil-White, 2001) or new learning strategiessuch as conformism (Boyd & Richerson, 1988; Henrich & Boyd, 2001) and?credibility enhancing displays? (Henrich, 2009) which changed the nature ofthe cooperative dilemmas faced by our ancestors.I am keen to promote a clearer understanding of the cooperative dilemmaof culture, and to see the many plausible solutions proposed, formalised,debated and evaluated. To help drive this process forward, here I will presenta novel proposal which uses reputation to kick off the ratchet by solvingthe cooperative dilemma in a way that selects for conformity to arbitrary,culturally evolving norms.Contemporary interactions between cooperation and cultural learning,especially non-concious behavioural imitation, give prima facie grounds tosuspect that they have a long and intertwined evolutionary history. Peoplewho are imitated by others tend to act more prosocially afterwards, bothadults (for reviews of this extensive literature, see Chartrand & Bargh, 1999;Chartrand & van Baaren, 2009; Lakin et al., 2003) and children (Carpenter26et al., 2013). Also, children are more likely to trust information learnedfrom individuals who have imitated them (Over et al., 2013). Reciprocally,people who are motivated to affiliate with others?for instance, those whohave been ostracised (that is, excluded from the benefits of their peers?sociality)?respond by imitating others more. Again, this is true of bothadults (Chartrand & van Baaren, 2009; Lakin et al., 2008, 2003; Leightonet al., 2010) and children (Over & Carpenter, 2009).In chapter 3, I argue for the plausibility of a previously unconsideredratchet: Negative Indirect Reciprocity (NIR). NIR explains how, if smallamounts of cultural learning brought proto-reputations into existence, com-munities could have coordinated their opportunities to exploit others in away that enforced their shared social norms. The existence of such community-enforced norms, in turn, could support more sophisticated, institution-basedforms of cooperation, facilitating the emergence of a complex, cooperativecultural corpus.In the next chapter I present the mathematical theory that articulatesNIR in a way that is accessible to psychologists. I focus on deriving clear,testable predictions about contemporary psychology. In the subsequentchapters, provide empirical tests of these predictions. In chapter 4, I surveypeople?s actual reputational intuitions and compare them to the predictionsmade by indirect reciprocity theory, including NIR. In chapter 5, I present agenuinely novel test of NIR. I test its fit to a recently observed psychologicalphenomenon that it was not designed to explain.27Chapter 3Negative indirect reciprocityThe puzzling origins of both human cooperation and cultural learning arelikely intertwined. Some aspects of human cooperation are shared with otherspecies and were likely shaped by the same kinds of distal pressures (e.g.,kin-based sociality; Daly & Wilson, 1988; Hamilton, 1964; Park et al., 2008;Smith, 1964; Stewart-Williams, 2007). Other aspects seem peculiar to hu-mans, such as queuing, paying taxes, and sacrificing to all-powerful deities(Norenzayan & Shariff, 2008). Formal models of how such cooperation-sustaining institutions emerge and persist typically presuppose that we area highly cultural species. For instance, important models assume that in-dividuals can readily coordinate their cognitive representations related toidentifying who is a deserving ?recipient? and a responsible ?donor? (Boydet al., 2010; Leimar & Hammerstein, 2001; Panchanathan & Boyd, 2004),that people establish abstract institutions (Sigmund et al., 2010) and knowhow to interpret one-another?s signals of cooperative intention (Boyd et al.,2010).These assumptions imply a deeper puzzle, since cognitive capacities forsophisticated cultural learning themselves pose a cooperative dilemma. Theyimply that natural selection favoured mutations that relaxed our genome?scontrol over its phenotype, and allowed fitness-relevant behaviour to beshaped by cultural information (Richerson & Boyd, 2004) acquired fromunrelated conspecifics. For this to be plausible, something must ensure that28the information acquired from others is, on average, fitness-enhancing, espe-cially if large, metabolically expensive brains are required to access it (Aiello& Wheeler, 1995).But the more influence culturally-learned information has on one individ-ual?s behaviour, the greater the selection pressure for others to exploit thatdependence by distorting the information they transmit to their own ben-efit. At first, many learnable phenotypes (e.g., stone tools use techniques)may have been hard to conceal or distort. However as soon as culturalknow-how became complex enough that it required language or pedagogy(Csibra & Gergely, 2009) for transmission, lying to exploit trusting othersbecame cheap and almost limitless in its potential for lucrative deceptions(Henrich, 2009; Lachmann & Bergstrom, 2004). This ?cooperative dilemmaof culture? is exacerbated by the fact that cultural information changes farmore quickly than genetic information, making it unlikely that genetically-evolved intuitions alone could distinguish useful from exploitative culturalknowledge. Before language and the cognitive capacities for coordinatingcomplex cultural institutions could have emerged, some mechanisms oper-ating on the same timescale as cultural change must have reliably sustainedthe quality of the public good that is our shared corpus of adaptive culturalknow-how.To tackle this puzzle, I ask how mechanisms that rapidly ensure co-operation in arbitrary domains (e.g., hunting, sharing information, trade)can get off the ground without pre-existing capacities for socially coordi-nating complex institutions and cognitive representations. One promisingpossibility are models of ?indirect reciprocity? (IR). Prima facie most IRmodels merely assume that (a) individuals have opinions of one another,(b) that these opinions influence how they treat each other and (c) thatcommunities somehow synchronise these opinions. These synchronised opin-ions are called ?reputations?. Once reputations exist they can catalyse theemergence of more complex forms of cooperation (Chudek & Henrich, 2011;Panchanathan & Boyd, 2004).Since many primates form coalitions with non-kin (Higham & Maestrip-ieri, 2010; Langergraber et al., 2007; Perry & Manson, 2008; Silk, 2002;29Watts, 2002), the first two assumptions typical of IR models are plausiblypreadaptations in our phylogenetic lineage. The third assumption impliessome social coordination and suggests a cooperative dilemma of culture. Onecould accrue many fitness benefits by strategically manipulating reputations.However it is also plausible that early, pre-verbal reputations were trans-mitted by observing others interactions (i.e., rather than gossip) and couldnot be easily distorted. Cognitive biases initially evolved to increase thequality of general social learning (e.g., Boyd & Richerson, 2005; Henrich,2009; Henrich & Gil-White, 2001; Laland, 2004; Nakahashi et al., 2012),may have incidentally also caused the transmission of social opinions (whichare, after all, just another type of cultural information) bringing reputationsinto existence. If early cultural learners closely monitored one-another?s di-etary, foraging or tool-use preferences, they may have picked up preferencesconcerning community members too.However, even if we grant the plausibility of the third assumption, ex-isting IR models implicitly assume an even stronger form of cultural coordi-nation. Framed in the context of reciprocal helping, these models supposethat sometimes someone has an opportunity to help but does not, and thattheir reputation worsens due to their inaction. This seemingly innocuous as-sumption implies that their peers somehow coordinate their representationsof both the abstract opportunity to act, and the significance of inaction.This is a sophisticated cognitive feat. It is especially impressive if we as-sume it emerged early enough to sustain the diverse, cooperative culturalcorpus that allowed an African primate species to spread across the globe.Noting this issue, Leimer and Hammerstein write that IR models assume?a reasonably fair and efficient mechanism of assigning donors and recipi-ents?a well-organised society, with a fair amount of agreement between itsmembers as to which circumstances define the roles of donor and recipi-ent.?(Leimar & Hammerstein, 2001). Subsequent IR models have mirroredthese assumptions without concern for their limitations.Here I fill the gap between the emergence of human culture, cooper-ation and reputations by showing how IR can get off the ground withoutassuming coordinated reactions to ?inaction?. Our model assumes that inac-30tion never changes reputations and demonstrates that even so IR can formthe substrate of more complex forms of cooperation. In fact, our modelis grounded in a relatively modest assumption about early cognition: thatindividuals disliked (i.e., worsen their reputational representation of) thosewho actively and observably exploited someone they liked (i.e., those witha good reputation).For several reason I focus on ?negative cooperative dilemmas? where ?de-fecting? means gainfully exploiting someone and ?cooperating? means seeingsuch an opportunity to exploit someone but passing it up (?doing nothing?).Substantial positive cooperation presupposes negative cooperation:Before more complex forms of mutual aid, defence and helping emerge,the ubiquitous opportunities to exploit each other (particularly, theold, weak and injured) must be brought under control. Otherwise, ex-ploitation and cycles of revenge will undermine positive cooperation.Positive cooperation exacerbates the negative dilemma(but not the reverse):The mutual aid of positive cooperation can create an abundance ofexploitable resources, both tangible (e.g., food caches) and intangible(e.g., trust). If cooperation has not first been stabilized in negativedilemmas, exploitation can quickly sap these benefits, sabotaging theviability of positive cooperation.Escalating returns: Prior to the emergence of complex institutions likemoney and debt, if an individual with a good reputation is helpedmultiple times (i.e., by multiple peers) they typically experience dimin-ishing marginal returns. A little food when you are starving providesa huge benefit; a lot of food when you are full provides only a smallone. On the other hand, repeated exploitation (e.g., stealing someone?sresources) can put victims in an ever more desperate situation withballooning fitness consequences. This suggests negative dilemmas mayhave generated steeper selection gradients earlier in our evolutionaryhistory, and so had more influence on the direction of evolution.31Built-in individual-level motivation: In a positive cooperative dilemmawith unobservable inaction (or lack of sufficient agreement about whatconstitutes ?inaction?), an individual?s reputation can endogenouslyrise (by observably helping) but not fall. Though an individual?s rep-utation might fall accidentally, selection will never favour individualswho take deliberate costly actions to worsen their reputation. Recip-rocally, negative dilemmas generate selective pressure for individualsto take deliberate, costly actions to improve their reputation. Positivedilemmas can?t accomplish this until sufficiently complex cultural in-stitutions or cognitive abilities establish agreement about what consti-tutes ?inaction?. This creates a chicken and egg situation for positivedilemmas, since substantial cooperation is required before sophisti-cated cultural-cognitive abilities can emerge.Relevance to culture: The cooperative dilemma of cultural learning, themain hurdle to more sophisticated institutional forms of cooperation,is a fundamentally negative dilemma. Individuals must pass up op-portunities to gainfully deceive their credulous conspecifics.Preadaptations are more plausible: Negative dilemmas are not sym-metric with positive cooperative dilemmas because they require thatindividuals notice, cognitively represent and respond to opportunitiesto profit by exploiting others, while positive cooperation requires theyrepresent opportunities to pay costs to help others. The former abili-ties were likely better, earlier in the culture-cooperation coevolutionaryprocess, since they yield direct, self-interested gains.Contemporary humans are more sensitive to harm than helping:Harmful or aversive actions, events or stimuli are more likely to haveeffects, and typically have stronger effects on contemporary humansthan their positive or beneficial counterparts (for extensive and influ-ential reviews, see Baumeister et al., 2001; Cacioppo & Berntson, 1994;Rozin & Royzman, 2001). This pattern recurs across the gambit of hu-man cognition, from sensation, to the experience of emotions and their32effect on cognitive processing, to trajectories of learning, to memory,impression formation, and the effect of feedback on self-perception. Ofparticular relevance is that negative information (i.e., about other?sharmful acts) seems to have a far more potent effect on diminish-ing someone?s reputation than positive information has on restoring it(Fiske, 1980; Rozin & Royzman, 2001; Skowronski & Carlston, 1987).Early antecedents of this negativity bias may even be apparent amongthree-month-old infants (Hamlin et al., 2010). In fact, people are morelikely to judge that someone caused, and intended to cause, a negativeoutcome than a corresponding positive one, even if their actual actionwas identical (Knobe, 2003, 2010). The ubiquity of negativity biases incontemporary cognitions suggests they run deep and may have ancientevolutionary roots. If our ancestors were as negativity-biased as weare, the impact of negative cooperative dilemmas may dwarfed positiveones in determining the long-run distribution of their reputations.Contemporary humans are more sensitive to harm by commissionthan harm by omission:Contemporary humans tend to condemn others moral transgressionmore severely (Baron & Ritov, 2004; Cushman et al., 2006; Sprancaet al., 1991) when they are the result of deliberate actions (commis-sions, which play a central role in NIR), than if they are the conse-quence of an equally intentional inactions (omission, which are absentfrom NIR). Correspondingly, people seem less-disposed to transgressby commission than omission (Ritov & Baron, 1999), especially if theymight be punished by others (DeScioli et al., 2011). These effects,which seem peculiar to negative commissions (Spranca et al., 1991)not positive ones, support NIR?s emphasis on negativity cooperationby commission alone.3334.Opportunityto actOpportunityto exploitVictim hasgood rep-utationExploit(Inflict d-amage, earn t-akings,reputation worsens)sDo nothing(Nothing changes)1? sVictim hasbad reputation Exploit(Inflict d-amage, earn t-akings)1??Opportunityto improvereputationCostly chanceto improvereputationvDo nothingPeers disappointed(Reputation worsens)?Peers apathetic(Nothing changes)1??1? v1??Costless exogenous reputation improvement(Reputation improves)??Pay for reputation improvement (volunteering)(Pay costs (k), reputation improves)NIR-P: ? = 1; NIR-V: ? = 0;NIR-M: ? = 1Evolving variables: v,sParameters: ? ,? ,?Consequences: d, t,k35Figure 3.1 (preceding page): The NIR decision tree. The probability of each branch is described byblue parameters and green variables (v: evolving disposition to pay reputationimprovement costs; s: evolving disposition to exploit victims with good and badreputations). Red text at terminal nodes describes the consequences of eachoutcome.ModelsTo help resolve the puzzle of early human cooperation I unpack three stagesof complexity that emerge from a more general model of Negative IndirectReciprocity (NIR). These insights follow from two convergent thought exper-iments. One possibility is a ?discrete strategy? perspective, where we imagineinteractions between very different kinds of individuals such as those whoalways cooperate with well reputed individuals (reputation-conditional coop-erators; RepCoop) and obligate defectors (Exploiter). An alternative isa ?continuous disposition? perspective (based on the successive invasions as-sumptions of Adaptive Dynamics (Geritz et al., 1997; Waxman & Gavrilets,2005)), where we imagine communities of individuals who share very similar,perhaps genetically endowed, dispositions to cooperate. In either case, wecan reason formally about what kinds of individuals would be favoured byselection, and both perspectives lead to similar general conclusions.Here I summarise these key qualitative insights and feature the succinctbut informative discrete strategy invasion criteria: the conditions underwhich obligate defectors (Exploiter) cannot invade a population of obli-gate cooperators (RepCoop). Figure 3.1 depicts the logic of these models.The Mathematical Model section below contains full details.Imagine a single, large population of individuals who each have a ?reputation??a community-wide opinion that influences others? behaviour?which can beeither ?good? or ?bad?. I represent this reputation as a stochastic variablewhose stationary distribution is the probability of being ?good? on average.During their lifetimes these individuals encounter two kinds of oppor-tunities. Sometimes (with frequency 1??) they notice a way to exploit aconspecific, yielding some takings (t) to the exploiter while damaging (d > t)their target. This situation is error-prone: sometimes well-meaning individu-als accidentally exploit (probability ?), and sometimes exploitation attemptsfail (probability ?). By assuming accidental exploitation is vanishingly rare(? ? 0), I present simplified expressions that preserve the model?s essentialinsights as long as ? remain small(? ? 120). The mathematical model sectionbelow provides the full expressions and robustness analyses.36I assume that individuals tend to dislike those who exploit someonethey like. That is, exploiting someone with a good reputation causes one?sown reputation to worsen. However I assume that forgoing opportunitiesto exploit (i.e., cooperating by inaction) carries no consequences. Assumingotherwise would imply that individuals were recognizing that opportunitiesto exploit existed, assessing that another individual noticed them as well,coordinating their reactions to these counterfactuals as a community, andso on.Consequently, not-exploiting badly-reputed individuals only ever yieldscosts and that poorly reputed individuals are always exploited. In this simplemodel, which focuses on active exploitation, unconditional cooperators neverprosper.Individuals also have opportunities to improve their reputation (withprobability ?). My first model?Pure Negative Indirect Reciprocity (NIR-P)?assumes that such improvement is costless and exogenous. Our earliestreputation-using ancestors had no awareness of their own reputation norhow to improve it. However their reputations may still have improved atrandom after some time, perhaps because their peers had limited memoriesand eventually forgot their old gripes or because they stumbled upon a non-excludable food resource that their peers gratefully shared.All the models presented here, including NIR-P, are bistable. They havean uncooperative equilibrium?where selection favours exploiting anyonewhenever the chance arises?and a cooperative equilibrium?where selectiondisfavours exploiting well-reputed individuals. At NIR-P?s cooperative equi-librium, a population of individuals who never exploit well-reputed peers(RepCoop) have, on average, higher fitness than rare individuals who al-ways exploit everyone, so long asExploitation inefficiency????td<RepCoop reputation????1?Exploiter reputation? ?? ?2?1+??? +??The left side of this inequality represents the inefficiency of exploitation;the ratio of benefits gleaned by an exploiter (e.g., thief) to harm caused37to their victim, which is often much greater (e.g., stealing from the weak).NIR is particularly stable in domains where exploitation opportunities arecommon and their consequences dire (e.g., stealing from the weak, sick andold).The right represents the difference between cooperators? average repu-tations (unity at the cooperative equilibrium when ? ? 0) and the invad-ing defector?s reputation. The possibility of accidental exploitation (? > 0)makes this inequality harder to satisfy by making cooperators? reputationsslightly worse on average, but does not change these qualitative insights(the details of this possibility are spelled out in the ?Mathematical Model?section, below).Figure 3.2 shows the population frequency of reputation-respecting RepCoopneeded before selection favours more cooperation. When opportunities forexploitation are far more common that opportunities for reputation improve-ment (? is small) and exploitation is inefficient ( td is small), a few cooper-ators are enough to trigger a cascade of ecological interactions that lead toa world where only poorly reputed individuals are exploited. A convergentresult under a ?continuous disposition? perspective implies that, under thesecircumstances, selection will mould a population only slightly ill-disposed toexploit their better-reputed peers, into highly reputation-sensitive individu-als loathe to exploit those with good reputations.A key challenge for models of cooperation is explaining how a speciesinitially composed of uncooperative individuals could arrive at the cooper-ative equilibrium?s basin of attraction. Here NIR has an easier time thanmost other approaches. It is plausible that preadaptations for friendship,coalition-formation and direct reciprocity gave early reputation-users someproclivity to dislike those who harmed their allies before social learning be-came strong enough to coordinate individual opinions into community-widereputations. That is, this evolving system may have started within its coop-erative basin of attraction, particular if inefficient exploitation opportunities(low td ) were plentiful (low ?).Assume that the reputation-sensitive communities described by NIR-P did emerge. As reputations came to carry great fitness consequences,38.0.0 0.4 0.4 3.2: NIR-P basins of attraction and equilibrium reputations.The location of the internal unstable equilibrium that dividesNIR-P?s cooperative (above the lines) and uncooperative basinsof attraction (left panel); and the equilibrium reputations (rightpanel) of cooperative RepCoop and Miser (higher, red lines)and uncooperative Exploiter (lower, blue lines) for td = 0.1(darkest lines), 0.5 and .0.8 (lightest lines). All errors (?,? ,?)set to 120 . When the proportion of non-exploiters is above thethreshold demarked in the left panel, selection, on average,favours even less exploitation (i.e., more cooperation).39selection could begin to favour individuals who noticed costly opportunitiesto improve their reputation and were disposed to act on them. For instance,they might chose to share a resource they could have kept to themselves tomake their peers? sentiments towards view them more favourable. To modelthis I assume that if an opportunity to improve one?s reputation occurs(?), it is sometimes costless and exogenous (probability ?), and sometimes(1??) requires the individual to pay a deliberate cost (k). For brevity, I callthis ?volunteering?, since the most interesting cases are those in which thesecosts raised reputations by contributing to others? fitness. These error pronevolunteering attempts sometimes fail (probability ?), but are still costly. Icontinue to assume that these early reputation-users could not coordinatereactions to inaction, and so ?not volunteering? carries no consequences forreputations. NIR-P is a special case of this broader model (i.e., where ? = 1).My next model?Voluntary NIR (NIR-V)?extends NIR-P by askingjust how much costly volunteering the threat of reputation-based exploita-tion can sustain. I consider two distinct cooperative strategies; both neverexploit well-reputed peers, but one always volunteers (RepCoop) and theother never does (Miser). In the formal model below I show that RepCoophave an advantage over Miser when exploitation is inefficient ( td small), op-portunities for reputation improvement are relatively rare (? small), costlyopportunities relatively plentiful compared to costless exogenous improve-ment (? small) and these costs are not too great ( kd small). Here I show thecondition for a cooperative, volunteering population (RepCoop) to do bet-ter than a rare, uncooperative, un-volunteering mutant (Exploiter). Thisis easiest to express in terms of the long-run probability that each strategywill be well-reputed (pi):pinirvRepCoop = 1pinirvExploiter =2??2??+(1??)(1??)td < (pinirvRepCoop?nirvExploiter)?kd?1??2(1??)(1??)40The first term on the right again represents the difference between co-operators? and defectors? equilibrium reputations. Now a second term ex-presses the additional burden of costly reputation improvement. In general,this condition can be satisfied so long as the costs of contributing are not toogreat ( kd is small) and opportunities to improve reputations are infrequent(? ,? is small).Continuous disposition perspectives on NIR-V models (see mathematicalmodel, below) suggest that, if intermediate dispositions to contribute arepossible they will be favoured by selection under NIR-V. Selection favourscontribution rates that balance the costs of reputation improvement againstthe benefits of sometimes gainfully exploiting others.To see why NIR-V is important, consider what ?volunteering? represents.Among a reputation-using community, selection favours doing whatever ittakes to improve your reputation, up to a certain cost (k). This could includeresource sharing, grooming, or chasing pests away from shared resources, butalso includes conformity to others? preferred behavioural standards and imi-tation of the best-reputed individuals. This selective pressure for conformityto whatever pleases one?s community could help sustain more sophisticatedforms of socially coordinated cooperation. NIR-V provides a plausible cog-nitive foundation for the emergence of ?social norms? (Chudek & Henrich,2011). Under NIR-V adhering to norms is rewarded, with a higher reputa-tion, but failing to is not punished.NIR-V also describes plausible cognitive and socioecological precondi-tions for the emerge of coordinated responses to inaction.In societies at NIR-V?s cooperative equilibrium?where individuals aredisposed to perform costly to please their peers?as individuals becomes evermore attentive to their own opportunities to volunteer they might also noticeand respond to others? volunteering opportunities. If volunteering typicallypleases peers, these more reputation-savvy communities may (with proba-bility ? ) be disappointed when someone pass up a volunteering opportunity,causing their reputation to worsen.Such reputation loss is not a deliberate attempt to punish deviance, it isan emergent consequence of prior selection for cognitive systems that attend41to volunteering opportunities. Nevertheless, since low reputations lead toexploitation by many peers, this disappointment coordinates community-wide sanctioning of failures to perform commonly expected behaviours.My final model?Mandatory NIR (NIR-M)?asks how interactions changeas volunteering gradually becomes a norm sanctioned by reputation loss(? ? 1). Now NIR-V is a special case (i.e., where ? = 0). At NIR-M?scooperative equilibrium, obligate cooperator-volunteers (RepCoop) resistinvasions by obligate defector-nonvolunteers (Exploiter) whenpinirmRepCoop =1??(1??)1??(1?? )(1??)pinirmExploiter =2??2??+2??(1??)+ (1??(1??))(1??)(1??)1??(1?? )(1??)td <pinirmRepCoop?pinirmExploiterpinirmRepCoop? kd1pinirmRepCoop?(1??)2(1??)1??The conditions for cooperation and volunteering to thrive under NIR-M are similar to NIR-V, however notice that cooperator?s reputations arelower (as ? becomes higher), a consequence of accidental failures to vol-unteer. These lowered reputations inversely weight the burden that volun-teering costs place on well-intentioned cooperators, making NIR-M a morefavourable environment for non-cooperators community expectations of vol-unteering rise ( kd and ? rise). From a continuous disposition perspective, asreputational enforcement of volunteering increases (? ? 1) the equilibriumdisposition to volunteer also increases since not doing so carries weightierreputational consequences.NIR-M, in particular, is capable of sustaining high-levels of costly volun-teering when defectors are rarely given free passes back to good reputations(low ?) as nearly everyone must conform to behavioural standards to sus-tain their reputation (see Figure 3.9). At this equilibrium an observer wouldsee an apparently naturally harmonious society (i.e., with little exploitation)and even high-levels of prosocial volunteering, or norm compliance. Mean-while, hidden from view, such cooperation would be sustain by rare butcostly exploitation of the poorly reputed. Detailed ethnographic work by42Henrich suggests this mechanism may be important in small-scale societies(Henrich & Henrich, forthcoming).DiscussionTogether these models map a path from minimal cognitive perquisites tolarger-scale forms of human cooperation by first suppressing within-groupexploitation?such as theft or rape?and then harnessing it to sustain coop-erative contributions to public goods?such as meat sharing or communaldefence. The logic of this process and testable predictions it implies arepresented in Figure 3.3.NIR-P describes dynamics when reputational systems first emerge: ifcommunity members are sufficiently reluctant to exploit their well-reputedpeers, selective forces will sustain and enhance this reluctance, perpetuatingharmonious (i.e., non-exploiting) communities. This is particularly likelyif exploitation opportunities are common and benefit the perpetrator littlerelative to the harm they cause the victim.Once harmonious communities exist NIR-V can emerge. That is, se-lective pressures can dispose individuals to make some costly, reputation-improving contributions to others welfare (which I have called ?volunteer-ing?). To do this they must develop an awareness of what behaviours wouldplease others on average. This puts a community?s normative behaviouralexpectations on the selective landscape. Ironically, it is the central challengeof NIR?that ?negative cooperation? is typically unobservable and so cannotreliably improve reputations?that leads to pressure for the cognitive abili-ties assumed by models of latter forms of cooperation?that individuals canrecognise and rapidly coordinate on arbitrary shared norms.Once NIR-V dynamics lend potent fitness consequences to shared expec-tations, and cause individuals to sometimes (as long as it is not too costly)do whatever it takes to satisfy those expectations, NIR-M can push commu-nities even closer to full-blown social norms and a psychology for navigatingthem. If individuals come to expect reputation-raising acts and are some-what disappointed by their absence, failure to volunteer can actually lower43one?s reputation. NIR-M shows that this strengthens selective pressures foradherence to community expectations, by providing a larger reputationally-distributed stick for their enforcement.To thrive in the social-ecology described by NIR-M individuals mustbe quick to perceive their community?s norms?the behaviours that pleaseothers on average, which could include generosity in times of plenty, sharingadaptive knowledge or wearing trendy running shoes?and be disposed toadhere to them. Communities meanwhile, come to wield a powerful meansof enforcing compliance to these norms. This distributed mechanism fornorm-enforcement emerges without any individuals intending it; they merelyselfishly exploit friendless, low-status victims when the opportunity arisesbecause they know they can get away with it. Indeed we may still witnessthese dynamics today, as the recurrent emergence of schoolyard bullyingrecapitulates the socio-ecological dynamics of early, pre-institutional humansocieties (Card et al., 2008; Merrell et al., 2008; Rodkin & Berger, 2008).The ?volunteering? norms that emerge under NIR-M could be prosocialacts, but need not be. In fact any arbitrary and even maladaptive commu-nity norm could be sustained by this mechanism. There are two reasons tosuspect that over time such volunteering would become increasingly proso-cial. First, improving others? welfare is particularly likely to raise their opin-ion of you. The creates a what cultural evolutionists have termed a contentbias favouring prosocial reputational content. Second, by making deviationfrom local community expectations costly, NIR-M, favours migrants whoadopt the norms of their new community rather than maintaining their oldbehaviours. This decreases behavioural variability within groups relativeto variation between them, increasing the between-group selective pressuresfor norms that lead to success in intergroup competition, which may includecontributions to defense, raiding, economic productivity, alliance building,trading and information sharing (Chudek & Henrich, 2011).These cognitive and socio-ecological conditions make it far easier for evenmore potent, coordinated or institutional, forms of cooperation to emerge.By reputationally rewarding those who share valuable cultural information,NIR also untangles the cooperative dilemmas that would otherwise prevent44the emergence of a culture-sharing species.The reputational system postulated by NIR imposes minimal cognitivedemands on early reputational cooperators, since they can ignore (1) any-thing that happens to people in bad standing, (2) all ?non-events? (like notexploiting), and (3) the exploiter?s previous reputation.In ancestral human societies, when individuals fell sick, were injured,or faced emergencies requiring them to rapidly leave camp, exploiters hadopportunities to steal food, mating opportunities, allies, beads, and rawmaterials (skins, flint, ochre, and obsidian) with little chance of directretribution?either because the victim could not pinpoint the perpetratoror was in no position to enact revenge. In times of distress (illness or in-juries) exploitation is particularly easy and the loss of valuable resources isparticularly damaging.These exploitative opportunities were likely distributed more or less atrandom, as my models assume. In contrast, opportunities for positive coop-eration likely covary with fitness, with fitter individuals having more chancesto help.The early stages of the emergence of generalised, high-fidelity culturallearning (Henrich & Henrich, 2007) may have provided the cognitive founda-tions for individual friendships, coalitions and opinions of others to transmitsocially, becoming ?reputations?. I hypothesise that these early reputationsinteracted with random exploitation opportunities to shape social cognitionand lay the foundations for the latter evolution of human cooperation andculture.Because the evolutionary story outlined has been carefully mathemati-cally formalised to ensure that it is consistent with the constraints imposedby natural selection. One benefit of such formalisation is that it makes theassumptions and predictions of the model transparent, allowing them to beprecisely tested. In Figure 3.3 I expound the testable claims entailed byeach step in the NIR account. These come in two forms: assumptions andpredictions.NIR is motivated by and grounded in several assumptions about our an-cestors? lives and interactions. For instance, several critical assumptions are45instantiated by NIR?s reputation-assessment rules. Like other IR models,NIR assumes that people track the reputations of those they interact with,that the way they treat those people is influence by their reputation, andthat those people?s reputations are affect by the way they treat others. Ifthese facts were true of our ancestors then (ceteris paribus) they ought to betrue of their modern descendants. If we didn?t observe these kinds of repu-tation dynamics in contemporary humans, we would have strong reason todoubt that NIR and other IR models are good descriptions of our ancestors?interactions.What sets NIR apart from other IR models is its emphasis on exploita-tion, that is: cooperation by deliberate, observable action (and not inaction)in negative cooperative dilemmas. NIR assumes that these kinds of interac-tions played a particularly important role in ancestral reputational and evo-lutionary dynamics. Contemporary populations do show robust negativity-(Baumeister et al., 2001; Cacioppo & Berntson, 1994; Rozin & Royzman,2001) and omission-biases (Baron & Ritov, 2004; Cushman et al., 2006; De-Scioli et al., 2011; Ritov & Baron, 1999; Spranca et al., 1991). That is, theirreputations are particularly sensitive to defection in negative dilemmas.NIR?s testable claims are still more specific. NIR expects these biasesto manifest in second order reputational interactions in particular. That is,the difference in reputational consequences when one interacts with a wellor poorly-reputed target individual ought to be more pronounced when onedefects in negative dilemmas, than in other kinds of cooperative interactions.As far as I know, this more specific claim has not been tested prior to theevidence presented in chapters 4 and 5.If we observe the psychological phenomena assumed by NIR, we canconfidently claim that we have tested NIR. We cannot, however, claim tohave explained the phenomena. NIR does not explain why these conditionsexist, it merely depends on the assumption that they do.On the basis of these assumptions (and others outlined in Figure 3.3)and the logic of natural selection, NIR also makes predictions about how in-dividuals adapted to an NIR-ecology ought to behave. These behaviour andphenomena, if we observe them, are explained by NIR at an ultimate level.46NIR tells us, for instance, why individuals ought to be less inclined to exploitbetter-reputed peers (by showing the plausibility of the ecological conditionsunder which selection minimise the this disposition), pay arbitrary costs toimprove their reputation, or conform to arbitrary community norms. How-ever testing these claim in modern populations is tricky. First, because theyare quite broad. Many other models also predict a world where reputationsmatter and people follow and enforce norms, and it many different processesmay have contributed to this outcome. Second, because NIR provides thebedrock for more complex forms of cooperation and so anticipates that ourcontemporary patterns of cooperation have been altered by many thousandof years of evolving cultural institutions. That is, NIR is a model of howcooperation first began, but not necessarily of how (except in special cases)it is maintained today.The psychology and behaviour of contemporary humans are better suitedto testing NIR?s quite specific assumptions, than its more general predic-tions. That said, NIR could be clearly tested and its contribution to thefoundations of human cooperation disambiguated from other mechanisms?,if we could identify contemporary societies that were not governed by long-evolved institutions and could recapitulate the socio-ecological dynamicsplayed out by our ancestors. The schoolyard societies continually rediscov-ered by generations of contemporary children, foisted by modern educativeinstitutions into the daily company of their equally young and naive peers,my provide just such an opportunity.The remainder of this dissertation kick-starts the more tractable projectof testing NIR?s specific assumptions. In chapter 4, I test whether people?sreputational assessments of others show the negativity- and omission-biasedpatterns of second-order indirect reciprocity predicted by NIR. In chapter5, I assess whether a peculiar, recently-observed anomaly in people?s moralreasoning can be explained by these same assumed conditions.47..Socio-ecology . .Cognitive mechanisms . .Empirical Questions thatTest NIR?s Assumption.Empirical Questions thatTest NIR?s Predictions..Preadaptations:???Personal coalitional affiliationsmake individuals less likelyto exploit their ?friends?, andless likely to befriend someonewho exploits their friend.???.& .Enough cultural learningthat individual affiliationsbecome community-wide reputations.. .? Do people and other primates have friends,allies or coalition-mates?? Are they less likely to exploit their friends?? Are they less likely to befriend those whoexploit their friends?..NIR-P:(? = 1)???Selection favours disposi-tions to only exploit peerswith bad reputations.????Improving one?s reputation?becomes a target of selection..& .Cognitive capacitiesfor noticing own reputa-tion and opportunitiesto improve it (e.g., bymaking costly contribu-tions to others? welfare).. .? If an actor exploits a well-reputed target, dopeople assign the actor a lower reputation?(1st-order reputation-assessment ruleassumed by NIR; Tested in Ch. 4 & 5)? Does an actor exploiting a badly-reputedtarget, suffer less reputation loss than oneexploiting a well-reputed target?(2nd-order reputation-assessment ruleassumed by NIR; Tested in Ch. 4 & 5).? Are people less inclined to exploit well-reputed than badly-reputed targets?(Key behavioural adaptation to NIR;Tested in Ch. 4 & 5)..NIR-V:(? < 1, ? = 0)???Selection favours dispositionsto take arbitrary actionsto please one?s peers whenone?s reputation is bad.????Arbitrary, shared com-munity expectations? be-come a target of selection..& .Cognitive capacities fornoticing other?s oppor-tunities to improve theirreputation (e.g., by someprosocial act) and disap-pointment if they do not.. . .? Do people track their own reputation?? Do they track opportunities to improve it?? Will they incur costs to improve it?? Are people more likely to take (costly, proso-cial) reputation-improving action when theirreputation is bad?.. . . . . .? Do people condemn (and are they more likelyto exploit) those who fail to proactively con-form to community norms?? Are they highly motivated to conform to theirpeers? reputation-relevant expectations?.. . . .NIR-M:(? < 1, ? > 0)???Selection more strongly favours dispositionto attend to and satisfy arbitrary commu-nity expectations. Communities reliably co-ordinate on arbitrary behavioural norms.???The content of community norms becomesthe focus of between-group selection, lead-ing to more sophisticated institutionalmechanisms for sustaining cooperation.Common Assumptions? Do negative cooperative dilemmas effect rep-utations and fitness consequences more thanpositive dilemmas?(Negativity-bias, assumed by NIR;Widely observed, and specifically testedfor 2nd-order reputations in Ch. 4 & 5)? Does cooperation/defection by deliberate ac-tion (commission) effect reputations morethan by deliberate inaction (omission)?(Omission-bias, assumed by NIR;Widely observed, and specifically testedfor 2nd-order reputations in Ch. 4)Figure 3.3: NIR?s logic, testable assumptions and predictions483.1 Mathematical modelIn this section I will once again walk through the logic of NIR. However,this time as I do I will explicate a mathematical model of the ecologicalinteractions it engenders and their long-term evolutionary consequences. Ateach step, I will clearly lay out the mathematical assumptions that underlieeach piece of verbal reasoning and their implications.3.1.1 Context and overviewThe simultaneous emergence of human cooperation and culture is a puzzle.Humans are able to share complex, encoded information about their world.We call this information ?culture?. For somewhere between a few hundredthousand and a few million years, this corpus of inter-generationally trans-mitted know-how has been accumulating and evolving. Today it containsimpressive concepts and skills like ?science? and ?opera?. Simultaneously, hu-mans have begun cooperating with each other on scales rarely seen outsideof eusocial species. We often chose to endure a relative individual cost tobring about a relative benefit for one or many others.The emergence of cooperation is a well-known evolutionary puzzle. Amonghumans, culturally evolved institutions (such as police forces, reputationsand library fines) make it much easier to explain how cooperation canevolve than in non-cultural species. A less well-know puzzle is how cultural-transmission itself emerges. Particularly central to this puzzle is the ?co-operative dilemma of culture?. Members of a cultural species must trustinformation they receive from others. This however makes it easy for othersto exploit them by distorting that transmitted information to their own ben-efit. These benefits should select for ever more cultural exploitation, untilit is no longer adaptive to culturally learn from others. How, before theemergence of cooperation-sustaining cultural institutions, did the quality ofcultural information stay high enough, for long enough, for our sophisticatedcultural learning capabilities to evolve?One step towards untangling this conundrum is establishing how simple,cooperation-sustaining reputational systems can emerge from simpler pri-49mate preadaptations. Several models have demonstrated how ?reputations?can sustain arbitrary forms of cooperation. However, these models implicitlyassume a strong form of cultural learning. Specifically: that communitiesagree on how to interpret one another?s inactions. They assume that some-times an individual faces an opportunity and chose to do nothing and thatthis will prompt a coordinated response from their social peers. This impliesthat others perceive the same opportunity and share a cognitive representa-tion of what ?should? be done in such situations. This is quite a sophisticatedcapacity to ascribe to our early reputation-judging ancestors, and it is notobvious whether it could arise without a ?cooperative dilemma of culture?.In this model I try to establish whether a reputational system for sus-taining cooperation could emerge even when inaction (i.e., not helping inpositive dilemmas, not exploiting in negative ones) does not change reputa-tions.We focus on the negative cooperation context (i.e., cooperating by ?notexploiting? someone, rather than ?helping? them) for the reasons detailed inthe main text.To avoid the potential confusion and ambiguity of talking about negativecosts and benefits, I describe these dilemmas in terms of the damage (d) doneto victims and the takings (t) earned by exploiters.We define Negative Indirect Reciprocity (NIR) as a system of interactingindividuals, with these properties:Assumptions.1. Individuals have regular opportunities for negative cooperation (i.e.,exploitation)2. Individuals have reputations. That is, others assign them a status ofgood or bad.3. Inaction (i.e., cooperating by not-exploiting) does not change reputa-tions.504. Exploiting ?good? individuals worsens one?s reputation.The final assumption seems to be a requirement common to any reputa-tional system capable of sustaining cooperation (Ohtsuki et al., 2006).There are three possible reputational rules we could add to this: ex-ploiting bad individuals could either improve, worsen or not change one?sreputation. Here I consider what I consider the case most plausible givenprimate pre-adaptations: exploiting bad individuals does not change one?sreputation. This is the most plausible starting point since it requires no spe-cial cognitive adaptations, just ignorance or apathy about whatever happensto ?bad guys?.This opens up the possibility of interpreting ?good reputation? as meaning?people you care about?, since all actions towards badly reputed targetsare ignored. This interpretation provides the explanatory bridge betweenreputations and primate cognitive preadaptations for forming coalitions, asdescribed in the main text.Assumptions.5. Exploiting bad individuals does not change reputations.But how do reputations improve? We consider two possibilities (seedecision trees in Figure 3.1). First, reputations could improve at random(with probability ?). Second, they could improve when an individual makesa deliberate, costly effort to improve them (at cost k).3.1.2 Model definitionImagine a large population of individuals; call them a ?community?. Eachindividual is born, interacts many times with their peers, reproduces and51dies. Their interactions are shaped (on average, in the long run) by their(and others?) genetically transmitted behavioural dispositions. These dispo-sitions can be characterised as either discrete strategies (Discrete Strategyapproach, below), or as continuously varying traits, whose average value issubject to natural selection (Adaptive Dynamics approach, below). Bothyield convergent insights.To express the continuous interpretation we use lower case latin variablenames (specifically s,v) to talk about a focal individual?s disposition, andupper case (S,V ) to refer to the population average of that same disposition.Individuals vary in their disposition to a) exploit a well-reputed peer(variables s/S), and b) to take costly actions to make others like them more(i.e., to improve their reputation; variables v/V ). It is easy to show that,within this model, selection will always maximise dispositions to exploitbadly-reputed individuals. Verbal logic suffices: not exploiting a ?bad guy?involves sacrificing a material benefit (t), but never yields any reputationalbenefit. By assuming that badly-reputed individuals are always exploited,our model focuses in on the less immediately obvious dynamics of exploiting?good guys?.To express the discrete interpretation, were merely consider individualswho?s dispositions are set to their boundaries. Specifically, obligate defec-tors (Exploiter: s? 1, v? 0); miserly cooperators (Miser: s? 0, v? 0)who do not exploit others but do not take costly actions to improve theirreputation (i.e., by improving others? welfare) either; and reputation-basedcooperators (RepCoop: s ? 0, v ? 1) who never willingly exploit and al-ways try to improve their reputation. As in the continuous version, we donot consider strategies that cooperate with poorly-reputed individuals sincetheir payoff must be strictly smaller than RepCoop?s.The outcomes of these interactions cumulate to influence the relativefrequency at which individuals genetically transmit these dispositions to thenext generation. Specifically, all else being equal, an individual?s reproduc-tive fitness is proportional to the sum of a) their takings (t) when exploitingothers, b) the damage (d) they incur from others? exploitation and c) thecosts (k) they pay to improve their reputation.52We assume that ?opportunities to improve reputations? and ?opportuni-ties to exploit others? occur with relative frequency ? : (1??). We assumethe relative rate of these two opportunities is an independent parameter inour model, thereby implicitly assuming that the dispositions we are mod-elling do not affect it (e.g., Exploiter is not more likely to face an oppor-tunity to exploit others than RepCoop is).Assumptions.6. Behavioural dispositions that lead to higher fitness become more fre-quent over time.7. Individuals can have a different disposition to exploit peers who they likeand peers who they don?t like. That is, reputation influences cooperativebehaviours.8. The probabilities of having an opportunity to exploit, be exploited orimprove reputations are uncorrelated with anything else in our model.We take as our starting point a world where individuals observe and im-itate one another?s opinions of one another, such that these opinions quicklyconverge on a community-wide consensus. We call this the individual?s ?rep-utation?. An unanswered and interesting question is how a community ofindividuals can evolve reputations in the first place. We do not attempt toanswer that here, but suspect that it involves early adaptations for culturallearning, such as prestige (Henrich & Gil-White, 2001) or conformity (Boyd& Richerson, 1988) biases. These can produce substantial asymmetries inhow likely the most influential individuals? or the majority?s opinions are ofbeing transmitted to others. To keep the present model simple, we considerthe limiting case where there is community consensus on reputations.To keep our model as simple as possible (without sacrificing insight), we53Rates?1?? ? (0,1) Relative frequency of reputation improvementexploitation opportunities.? ? (0,1) Probability of reputations improving costlesslyErrors? ? (0,1) Accidental cooperation (doing nothing)? ? (0,1) Accidental defection (exploiting)? ? (0,1) Failure to improve reputation, despite paying costsCommunity Reactions? ? (0,1) Probability that ?not helping? worsens reputation.Table 3.1: Summary of parametersassume that reputations can be in two states (good or bad). This makessense, because though the actual underlying psychological representation ofan individual?s reputation may be continuous, when individuals face a deci-sion about exploitation (at a given t and d) or costly reputation improvement(at a given k), they must ultimately make a binary choice.By reasoning about who has higher fitness when (i.e., the sum conse-quences of these interactions), we can draw inferences about which disposi-tions are likely to evolve under which circumstances. To do this we need tofigure out what an individual with certain dispositions would experience ina population of individuals with other dispositions.3.1.3 Reputational dynamicsOur first step is to lay out all the circumstances under which an individual?sreputation might change, weighted by their probabilities of occurring. This54is depicted visually in Figure 3.1.First we define the probability, when a reputational change happens, thata focal individual?s reputation will become good (Pg) or become bad (Pb).Then, we can derive the stationary distribution of this two-state stochas-tic process as the relative probability that the focal individual?s reputationbecomes good (G = PgPg+Pb ).To do this, we will need to define parameters that represent all thedifferent things that we?ve said can occur. We have already represented therelative rate at which opportunities to improve reputations arise, relative toopportunities for exploitation ( ?1?? ). We also know the probability that anindividual?s reputation improves for extrinsic reasons, without them needingto take any action at all (?). If their reputation does not improve by suchgood fortune (1??), we have also assumed that the (average) probabilitythat an individual will pay a cost to improve it is a consequence of theirevolved disposition (v). But what are the chances that this costly attempthits it mark? Let?s assume that deliberate attempts to improve reputationsfail with probability (?). Putting this all together, the relative probabilityof one?s reputation improving isPg = ? (? +(1??)v(1? ?))What about worsening? Reputations worsen (in our model) when a focalindividual exploits someone with a good reputation. We know that exploita-tion opportunities arise with probability (1??), but the focal individual willbe the potential exploiter (not the potential victim) in only some of these(half of them, by assumption 8). Next their reputation will only worsenif their potential victim has a good reputation. Again by assumption 8,this will be (on average) the frequency of individuals with good reputationsin the population (for now let?s call this variable G, later we will solve it?svalue). Next, they need to be inclined to exploit their victim, which we haveassumed happens according to their evolving disposition (s). Even if theychose not to exploit this well-reputed target (1?s), there might still be somepossibility (?) they do so by accident. Finally, even though the focal indi-55vidual might want (and even try) to exploit someone, there might be someprobability (?) that they fail entirely and accomplish nothing (including notworsening their reputation).Do individuals who decide not to take a costly action when the oppor-tunity arises (1? v) experience a worsened reputation? This would implythat the community had a coordinated response to their inaction, and thuswell-coordinated expectations about each other?s behaviour. This is the veryassumption we are trying to avoid, so at first we want to model a world wheretheir reputation does not change. As we will see, the resulting socio-ecologyis one that favours an increasingly shared awareness of reputation-improvingopportunities, and so we also want to see how our model changes as thisshared awareness gradually increases. So, we include a parameter (? ) in ourmodel that describes the probability that ?not taking costly actions? (or not?volunteering?, for brevity) actually worsens reputations.Putting this all together we have:Pb = (1??)2 G(s+(1? s)?))(1??)+??(1??)((1? v)+ v?)G (G) = PgPg+Pb= 2?(v(1??)(1??)+?)2?(v(1??)(1?? )(1??)+?+? (1??))+G(s(1??)+?)(1??)(1??)(3.1)We express the stationary distribution as a function of G, the probabilityof interacting with a well reputed individual, whose value we do no knowyet.This general expression makes it easy for us to explore specific cases. Forinstance:? a world before people start deliberately improving their reputations(? ? 1),? a world where failing to meet community expectations does not worsenreputations (? ? 0), and56? a world where it does (? > 0),? domains in which exploitation is easy (? ? 0) or hard (? ? 12), and soon.Another approach to reputationsRather than parametrising the relative rates of opportunities, another ap-proach to modelling change in reputations is to assume that each individ-ual confronts the same ordered sequence of interactions (for instance, seePanchanathan & Boyd, 2004). First they might have an opportunity toimprove their reputation, followed by one or more opportunities to worsenit by exploiting others. This approach implicitly invokes assumption 8 byassuming that there is a constant probability (let?s call it ?) that anotheropportunity to exploit arises before an opportunity to improve one?s repu-tation does. Consequently each subsequent exploitative opportunities? con-sequences must be weighted by its ever smaller chances of occurring. Thegeometric sum of these diminishing probabilities converges to a single num-ber ( 11?? , if the first exploitative event is certain, ?1?? if it also occurs withprobability ?). The modeller can then derives the relative rate (and weightsthe consequences) of reputation improvement to exploitation. Specifically,1 : 11?? or 1 : ?1?? .We feel it?s simpler to merely parametrise this rate and find the sta-tionary distribution. However the solution is identical in both cases. Totranslate between these two approaches, use this map: ? = ??1??2 if the firstexploitation opportunity is certain, or just ? = 1?? if it is not.It is also worth noting that in this model I allow reputations to improveexogenously (i.e., with probability ?), but they only ever fall endogenously(i.e., when someone exploits someone else). This allows me to focus ourmodel more tightly on the dynamics of exploitation and simplifies some ofthe math, at the cost of creating some potentially degenerate reputationalequilibria (i.e., special cases which may not be true in general) at the bound-aries of our system. Specifically, in the absence of accidental exploitation,individuals who refuse to exploit good guys (i.e., s? 0) have perfect repu-57tations. I have explored models where reputations also fall at random, andtheir dynamics and qualitative implications do not differ from those of thesimpler, more tractable model presented here.3.1.4 Behavioural dynamicsNext, we need to say something about what the fitness consequences of afocal individual?s actual behaviour are. Let?s go piece by piece, defining thefitness consequences of improving one?s reputation (W1), exploiting (W2) andbeing exploited (W3), before combining them (W = W1 +W2 +W3). Again,we assume that the probability of encountering an opportunity to exploitsomeone with a good reputation is (G). Now we also need to be explicitabout the probability that a randomly sampled member of the communitywill exploit the focal individual, given that the focal individual has a goodreputation. In our model this is just the mean community disposition toexploit (S).(W1): Given the opportunity to do so (?), an individual so disposed (v)may pay a cost (?k) to attempt to improve their reputation. We assumethey pay this cost even if their attempt fails (1? ?).(W2): Given an opportunity to exploit (which occurs with relative fre-quency 1??2 ) a focal individual could earn some takings (t) and inflict somedamage (d > t) if they meet someone with a bad reputation (1?G), or ifthey meet someone with a good reputation (G) and are disposed to exploitthem (s) or do so accidentally (?), and don?t mess it up (1??).(W3): The same is true of other individuals exploiting the focal indi-vidual, given the focal individual?s reputation. To reason about the averagefitness consequences of these events, we need a way of referring to the aver-age reputational state of a focal individual over their lifetime (let?s call it gfor now).58All together this yields:W1 = (?k)?(1??)vW2 = (+t)1??2 (1??)((1?G)+G(s+(1? s)?))W3 = (?d)1??2 (1??)((1?g)+g(S+(1?S)?))W (g,G) = W1 +W2 +W33.1.5 Combining reputations and behaviourNow we have two dynamics models. One tells us how reputations change andthe how often an individuals will have a good reputation (G ) as a functionof their behavioural dispositions (s,v), their likelihood of interacting with awell reputed peer (G). We also have a model of the fitness consequences thatshape how those behavioural dispositions evolve (W ), which is also dependson the likelihood of meeting a well-reputed community member. Our nextstep is to connect these models, by assuming that the chance of meetingsomeone with a good reputation (G) is well approximated by our model ofhow often your community members have good reputations (G ).Assumptions.9. The long-run probability of encountering someone with a good repu-tation (G) is well-approximated by the mean of the stationary distri-butions of the reputations of the individuals in the population (i.e.,G? G (G)) . This implies that reputational dynamics are much fasterthan the dynamics of behavioural dispositions.Given this assumption, when the population is monomorphic (i.e., ev-59eryone has the same dispositions: V and S), we can easily solve G as:G = G (G)??v=V,s=SG = G (G)Note that there are two quadratic solutions to this equation, but one isalways outside the logical bounds of G ? [0,1]; the other is presented abovein Equation 3.1.How good is this approximation? We can get some traction on this ques-tion by simulating these dynamics. Included with this document is a scriptfor the statistical modelling computer system R. It first creates a popula-tions of n Resident individuals with dispositions S and V , and reputationsset randomly to either good (1) or bad (0). It then randomly selects anindividual, randomly gives them either an opportunity to improve their rep-utation or exploit someone, randomly selects an exploitation target from thesame population and plays out the consequences exactly as described above.It also creates a population of n ?rare? Invaders (with dispositions s andv), and carries out the same process for them, with the one exception thattheir exploitation targets are chosen from the Resident population ratherthan their own. After n iterations, we observe the mean frequency of indi-viduals with good reputations in each population (G and g for Residentsand Invaders respectively). We do this t times, such that (in expectation)each individual has faced t opportunities for reputation change. Using theattached script you can readily and rapidly repeat this for any combinationsof parameters. These simulations show that G and g reliably and rapidlyconverge on G (G) in any reasonably sized community who interact manytimes. A typical simulation run is presented in Figure 3.4.Below we use a similar technique to find G when the population is poly-morphic (comprised of multiple discrete types at known frequencies).60610 100 200 300 400 500TurnCumulative average reputation0. SimulatedAnalytically PredictedResidentInvaderFigure 3.4: Simulations showing the accuracy of analytic predictions about Resident and Invader equilibriumreputations. The theoretically predicted (solid lines) and simulated (dots) cumulative proportionof time that individuals have had a good reputation. Here the population contains 100 commonResidents and the average of 100 different trajectories of rare Invaders who interacted with them.All errors and ? are set to 120 , ? set to 0 (i.e., NIR-V) and ? set to 25 . Residents traits are S= .8,V = .1,while Invaders have s = .5,v = .2. During each ?turn?, as many individuals are randomly selectedto experience a potentially reputation changing events as there are individuals. Consequently, inexpectation, each individuals experienced as many events as there are turns.3.1.6 Discrete strategy approachNow that we have formally described the reputational and behavioural in-teractions, we explore two approaches to reasoning about their long-termoutcomes. The first is the invasion analysis technique of evolutionary gametheory. In this (mathematically aided) thought experiment, we imaginepopulations comprised of the three kinds of individuals described above:Exploiter (s = 1,v = 0), Miser (s = 0,v = 0) and RepCoop (s = 0,v = 1).Since evolution started from uncooperative communities, we first imag-ine a population comprised entirely of Exploiter (the residents) and askwhether a rare cooperator (the invader) could thrive there. We next imag-ine a population entirely comprised of the two kinds of cooperators and askwhether an Exploiter could thrive there. While this approach is useful forgleaning simple and useful insights into these boundary cases, we will alsowant to ask how large a fraction of cooperators are need before they startto have an advantage, on average, over Exploiter.To get started, let?s define the population frequencies of RepCoop (x),and Exploiter (y). This implies the frequency of Miser (1?x?y). Next,let?s express the stationary distribution of the reputation and fitness of arandomly sampled individual practicing each strategy:62GRepCoop(G) = G (G)??v=1,s=0GMiser(G) = G (G)??v=0,s=0GExploiter(G) = G (G)??v=0,s=1Gmixed(G) = xGRepCoop(G)+ yGExploiter(G)+(1? x? y)GMiser(G)Wmixed(g) = W (g,Gmixed(G))??V=x,S=yWRepCoop = Wmixed(GRepCoop(G))??v=1,s=0WMiser = Wmixed(GMiser(G))??v=0,s=0WExploiter = Wmixed(GExploiter(G))??v=0,s=1Again, by assuming that the stationary distribution of reputations is ac-tually a good approximation to the long-run average probability of meetingsomeone with a good reputation, we can readily solve for G.G = Gmixed(G)Again, we can use a similar simulation (now with a mix if three strate-gies rather than an Invader and Resident) to check the accuracy of thisapproximation. Again, we find it very accurate. A sample simulation runis presented in Figure 3.5 and the included simulation code allows you toreadily explore any parameters you please.63640 100 200 300 400 500TurnsCumulative average reputation0. PredictedEntire PopulationCOOPMISERALLDFigure 3.5: Simulations showing the accuracy of analytic predictions about equilibrium reputations of threediscrete strategies. The theoretically predicted (solid lines) and simulated (dots) cumulative pro-portion of time that individuals of three discrete types (distinguished by color) have had a goodreputation, and the population average (black). Here a population comprised of 33 individuals ofeach type was simulated, with all errors and ? set to 120 , ? set to 0 (i.e., NIR-V) and ? set to 25 .During each ?turn?, as many individuals are randomly selected to experience a potentially reputationchanging events as there are individuals. Consequently, in expectation, each individuals experiencedas many events as there are turns. Here ALLD refers to ExploiterWhile we have been able to derive a precise expression for this quan-tity, it is long, complicated and does little to hone our insight on its own.Instead, we pursue two techniques for extracting simpler intuitions aboutthis evolving system. First, we can numerically map out these expression tovisually glean their implications (presented and discussed further below).Second, we can gain useful insights by analytically exploring the condi-tions under which a member of a monomorphic population (the Resident)has higher fitness that a rare mutant of another type (the Invader). To dothis we assume populations are large enough that we can assume Invadersnever interact with their own kind without distorting our conclusions.Assumptions.10. Populations are large enough that we can accurately approximate dy-namics by ignoring the impact of rare mutants on plentiful residents.Invasion Analysis: ExploiterOur model shares a common (and intuitively correct) feature of all mod-els of the evolution cooperation. Cooperative strategies (i.e., Miser andRepCoop) cannot (at least deterministically) invade a population of un-remitting defectors (Exploiter). Their kindness is costly but never re-warded (not even indirectly) when they are rare.To see that this is the case in all the possible worlds we are modelling,we inspect the conditions for these two strategies to invade a population ofExploiter.WMiser??x=0,y=1 >WExploiter??x=0,y=1Which simplifies to the inequality?((1?? )? +? ) >??2((1?? )? +? )2 +2?(1??)?(1??)65Since the RHS is at least as large as??2((1?? )? +? )2, with an ad-ditional term added, this inequality is strictly false. Thus, Miser cannotinvade Exploiter. Notice however that this additional term is propor-tional to ?. That is, if ? is zero, a special degenerate case results whereMiser has the same fitness as Exploiter at the Exploiter monomorphicequilibrium. This is because without random reputational improvementExploiter has a zero reputation at equilibrium and so Miser never for-goes an opportunity to exploit Exploiter. If Exploiter?s reputation isperturbed even slightly from zero, Miser sometimes considers them goodand refrains from exploiting them, but their mercy goes unrewarded in anExploiter population.Next, consider RepCoop invading Exploiter:WRepCoop??x=0,y=1 >WExploiter??x=0,y=1We can simplify this inequality as??2 (? 2 +(1?? )2?2)+2??((1?? )?? +(1??)(1??)) < ?((1?? )? +? )and again readily see that it is never true.So, a population of Exploiter is stable against invading cooperators.Invasion Analysis: Miser and RepCoopMiser is able to invade RepCoop whendk <2?(1??(1?? ))(1? ? (a2)?(1??)(1??))(1??)(1??)(1??)((a2)?(1??)(1??)?2??2?(? (1??)+?)+a2)a2 =??2(1? ?(1?? )(1??))2 +2??(1? ?(1??))(1??) (1??)??(1? ?(1?? )(1??))Meanwhile RepCoop is able to invade Miser when66dk >2?(1??(1?? ))(1? ? (a3?a4)?(1??)(1??))(1??)(1??)??(a4?a3)(1??)?(1??)(1??) +2?a4+?((1??)(1??)?????? (1??))a3?a4(1?2?)+2(1??)???a3 =?? ((? +????)2? +2??(?1+?)(?1+?))a4 = ?(? (1??)+?)These inequalities define four regimes of interactions between Miser andRepCoop strategies. Either (a & b) one and not the other monomorphicpopulation is stable, (c) both are stable (in which case an internal unstableequilibrium must exist between them) or (d) both are unstable (in whichcase an internal, stable, polymorphic equilibrium exists, though this onlyoccurs in a very small region of the parameter space). Figure 3.6 provides avisual intuition into when these four regimes occur in terms of parameters?, ? and ? .To summarise, Miser has a relative advantage when opportunities forreputation improvement are common relative to opportunities for reputa-tion loss (? is high), reputations often rise exogenously (? is high) andcommunities do not yet expect reputation-improving acts (? is low), and soMiser individuals suffer little reputation loss for failing to volunteer. Therelationships of the payoffs can readily be seen on the LHS of the inequali-ties. Miser thrives when exploitation causes little damage (d is small) andreputation improvement is costly (k is large).6768? = 0 (NIR-V) ? = 12 ? = 1 (NIR-M)RepCoop(blue) & Miser(red) resistExploiter0.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 Miser?(blue) and?Miser resistsRepCoop? (red)0.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 stable(blue), Miserstable (red) andboth unstable(yellow)0.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 3.6 (preceding page): Invasability plots for the three discrete strategies in NIR. Shaded regionsshow whether monomorphic populations can resist invasion by rare mutants. Thetop row shows when RepCoop (blue) and Miser (red) resist Exploiter in-vaders. The middle row shows when these two strategies resist invasion againsteach other, and indicates regions of bistability (i.e., overlapping regions) andstable mixed populations (i.e., unshaded regions). By superimposing the uppertwo regions, the bottom row shows when RepCoop and Miser resist invasionby both others strategies, and when Exploiter invades both (yellow). In thesmall unshaded regions in the bottom row, strategies resist. In each sub-figure,the horizontal axis shows the relative frequency of opportunities for reputationimprovement (?) and the vertical axis shows, when reputations improve, the prob-ability that the improve exogenously for free (?), rather than endogenously forcost (k). For these figure I assume that t = 14 ,d = k = 1, and all errors (?,? ,? ,? )are 120 .Invasion Analysis: Exploiter invadesThe condition for Exploiter to invade Miser isa5 =?? ((? +????)2? +2??(?1+?)(?1+?))a6 = ?(? (1??)+?)td > 2(1??)???(1??)(1??)?(? (1??)+?)(a5?a6)(a5?a6)(a5?(1?2?)a6)The condition for Exploiter to invade RepCoop isa7 =?(1? ?(1?? )(1??))2?2 +2?(1? ?(1??))(1??)?(1??)a8 = (1? ?(1?? )(1??))?d?tk <dk?a7?a82?(?1+?)?(?1+?)a7?a8?+2(?+????)??2? 1??(1?? )(1??)(?(1??)(1??) ??a7?a8)It is worth noting that as ? goes to zero, the expression ?a7?a8 approaches1??(1?? )(1??)(1??(1??))(1??)(1??) . This allows us to readily derive the key simplificationsin the main text.Though this precise expression is fairly complex to puzzle over symbol-ically, a visual presentation makes its implications clear. The top row ofFigure 3.6 shows the regions where RepCoop and Miser can resist inva-sion for parameters ? , ? and ? . Both strategies do better when ? is low,since reputations are more valuable and costs paid to maintain them arebetter recouped.When not-volunteering carries no negative consequence (? = 0), Miseris quite robust. However as ? increases, Miser struggles if they cannot relyon reputations improving extrinsically (? low), while RepCoop can hold itsground.Alternatively, we can unpack these expressions by considering special,70theoretically interesting cases of the parameters. Here we examine the sta-bility conditions for RepCoop in particular, since it is the more theoreticallyinteresting of the two.Per-scenario breakdown of RepCoop stabilityNIR-PWhen the simplest, earliest reputations first emerged among our ances-tors, we can assume that individuals were not aware of their own reputationsand rarely took deliberate action to improve them. Instead, reputations im-proved exogenously with some probability (tracked here by ?). For instance,they may have just gotten better over time as others forgot the past, or in-dividuals may have accidentally performed actions that improved them. Toinvestigate this situation we evaluate our inequality when reputations onlyimprove at random(? ? 1) and community don?t care if you fail to volunteer(? ? 0). After some simplification, we gettd< 1? 2?1+??? +??Which, interestingly, is (given the same case of the parameters)td < GRepCoop ?GExploitert < d(GRepCoop ?GExploiter)Put simply, for Exploiter to thrive the advantages of exploitation needto exceed to additional damage incurred by having a bad reputation.NIR-VThis is the key case in the emergence of cooperation by NIR. NIR-Pshows us that a society can emerge, and be stable in the face of mutationand natural selection, in which individuals gain a fitness advantage frommaintaining a high reputation. We now have in hand a plausible selectionpressure favouring mutations which cause individuals to take costly actionsthat improve their reputation. But just how costly can these actions be,71and do these deliberate efforts at reputation improvement change NIR?sdynamics?To investigate this case we relax our constraints on ?, assuming only that? = 0. Here we show the simplified case where ? ? 0. The supplementalMathematica file contains complete expressions.The condition for cooperators to resist invasion can again be simplifiedin terms of the reputational equilibria (given this case of the parameters).t + k ?1??2(1??)(1??) < d(1? 2??2??+(1??)(1??))t + k ?1??2(1??)(1??) < d(GRepCoop ?GExploiter)td < (GRepCoop ?GExploiter)?kd?1??2(1??)(1??)Now we can see precisely the degree of extra burden that the cost ofdeliberate reputation improvement places upon cooperators. The damagespared them by their good reputation (GRepCoop?GExploiter) must not onlyexceed the benefits of exploitation they?re forfeiting ( td ) but also the coststhey pay to maintain it ( kd?1??2(1??)(1??) ).If RepCoop individuals prosper, NIR-V can sustain both non-exploitationand some degree of arbitrary costly reputation improvement, as long as dam-age is great relative to takings and opportunities for reputation improvementare relatively scarce compared to opportunities for exploitation.If reputations sometimes improve costlessly (?), Exploiter have anadvantage. In the limiting case where reputations cannot improve withoutdeliberate effort, our inequality simplifies to:d? tk> ?1??21??NIR-MThe earlier models give us some reason to think that natural selectioncould have favoured costly cognitive adaptations for a) attending to one?sreputation, b) attending to what actions would raise it, and c) being moti-72vated to take those actions (even if they?re costly). Eventually, this couldlead to communities of individuals who very deliberately and carefully tryto improve others opinions of them.Since in such a community, individuals would be carefully attendingto their own opportunities to improve their reputation by contributing toothers? welfare. This could equip them with all the cognitive prerequisitesthey?d need to begin noticing others? opportunities for reputation improve-ment by acting in their interests, and be disappointed if they didn?t. Suchdisappointment could the be the foundation for coordinated reputationalpunishment of those who don?t meet community expectations. Since suchspontaneous, widespread disapproval would actually motivate other-fitness-enhancing reputation-reparation, it would be self reinforcing, eventuallydriving up the probability that failure volunteer worsens reputations (? ? 1).Could such a society still sustain cooperation? Just how costly a levelof volunteering could it sustain?t(1??(1??)1??(1?? )(1??))+2k(1??) ?(1??)(1??)< d(1??(1??)1??(1?? )(1??) ?2??2??+2??(1??)+ (1??(1??))(1??)(1??)1??(1?? )(1??))td <GRepCoop?GExploiterGRepCoop ?kd1GRepCoop?(1??)2(1??)1??A simple way to understand what this inequality is telling us is to com-pare it to NIR-V?s condition. Now the reputational advantage that coop-erators achieve is greater. That is, the term (GRepCoop ?GExploiter) is nowweighted inversely by (GRepCoop), which is smaller than one, making thewhole term larger. Meanwhile the costs they pay are more onerous too,similarly weighted by 1GRepCoop .Overall, except in cases when volunteering costs are very low (k is smallrelative to td ), RepCoop will be less stable under NIR-M than NIR-V. Ifcommunity demands begin to escalate, individuals who simply ignore themeventually come to thrive.As we?ve seen above, these additional costs buy RepCoop an advantage73against Miser, and as we?ll see below, they also result in a higher level ofequilibrium volunteering.Numerical MapsBy analysing the stability conditions for monomorphic populations we?vegained some insight into when different strategies can thrive, and seenthat mixed populations will sometimes also contain an unstable equilib-rium. These unstable mixed-equilibria define the basins of attraction of themonomorphic boundary equilibria, offering insight into how likely each strat-egy is to emerge. We?ve also seen that stable mixed population of RepCoopand Miser is also sometimes possible.To offer some traction of the locations of these internal equilibria, wehave mapped out their locations in Figure 3.7 for several different parameterregions. Note that unlike our assumption-testing simulations above, thisis not a simulation. It is merely a different way of interpreting analyticexpressions. In general their dynamics are identical to the insights gleanedfrom our invasion analyses. Cooperative strategies thrive when exploitationis inefficient ( td is small), bad reputations stick (? is low), RepCoop thriveswhen the cost of volunteering are low (k is small) and individuals are skilledat navigating reputations (? high).The attached Mathematica file present an interactive barycentric plot, bymeans of which you can readily explore the dynamics of this three-strategysystem across the full scope of parameter values. This file also includes is thefull expression for GMixed in all its massive glory, and all the code requiredto re-derive the full model.7475NIR-P (? = 1) NIR-V (? = 0) NIR-M (? = 1)? ? 110 ,210 . . .910C DMC DM0.4C DM0.3t ? 110 ,210 . . .910C DMC DM0.3 C DM0.3k ? 12 ,1 . . .92C DMC DM2C DM76Figure 3.7 (preceding page): Meta-barycentric plots of NIR boundary equilibria. First-order barycentricplots depict the evolution of the relative frequency of discrete NIR strategies.Corners of the triangle represent monomorphic populations of RepCoop (left),Miser (top) and Exploiter (right) and spaces between them represent mixedpopulation in the corresponding proportion. Along their exterior edges barycen-tric plots may contain ?boundary equilibria??relative frequencies at which the twostrategies at the abutting corners have equal fitness. This meta-plot shows howthe location of those boundary equilibria changes as three key parameters change(outer rows, ?, t and k). When the row-parameter is at its minimum, the locationsof boundary equilibria are depicted with a star (stable) or filled circle (unstable).Green stars show edges where fitness is always equal. Increasingly lighter arrows(stable: dashed, unstable: solid) show where the equilibria move to as the cor-responding parameter changes. If an equilibrium does not shift substantially atthe first increment in the parameter sequence (e.g. as ? changes from 110 to 210),the first parameter value that makes a substantial difference is printed beside thecorresponding arrow (e.g., the stable internal equilibrium moves dramatically as? goes from 310 to 410 in the top-right plot). Since each edge contains a single inter-nal equilibrium, the stability of monomorphic populations at the correspondingcorners can easily be inferred. Corners abutting edges with stable equilibria arethemselves unstable, and vice-versa. Unless otherwise indicated by column or rowlabels, parameters are set to: k = d = 1; t = 14 ;? = ? = 110 ;? = ? = ? = 120 .3.1.7 Evolving continuous traits interpretationThere is a second way we could reason about the long term evolution ofthese dispositions. Rather than assuming that change happens by a dra-matic mutation (e.g., a population of Exploiter (S = 1,V = 0) in which asingle cooperator (s= 0,v= 1) arises, we can assume it happens by small mu-tations which spread to fixation before another mutation arises. Using thisapproach, we assume the members of the resident population all have ap-proximately the same dispositions (0 ? S,V ? 1). These dispositions evolvegradually over time by small genetic mutations arising that change the dis-position slightly. Such mutations typically either spread to fixation or vanish(depending on whether they provide a fitness advantage or disadvantage)before another mutation arises.To formally reason about the long-run consequences of such a process,we ask what would happen to a mutant whose dispositions varied from thepopulation average by some small amount (s= S+?s,v=V +?v). How smallan amount? Small enough that a linear approximation of the differencebetween the fitness of the resident and that of the invader gives an accurateapproximation of the evolution of the system.In evolving systems where small dispositional changes yield fitness differ-ences that are approximately linear,1 these approximations can yield accu-rate inferences even if large mutations sometimes occur. For systems whosedynamics are highly chaotic or just very non-linear, this approximation onlyholds in the limit of very small mutations. Numerical inspection of NIR sug-gest that it?s fitness difference are typically monotonic and approximatelylinear, giving us confidence in the insights of this adaptive dynamics analy-sis (for a discussion of other limitations of the assumptions underlying this?adaptive dynamics? approach, see Waxman & Gavrilets, 2005).1for instance, if having a greater disposition to do something improved your fitness by,on average x fitness units, then having a disposition doubly as great would yield close to2x units improvement77Assumptions.11. Everyone?s similar enough, and12. Mutations are small enough, and13. Tend to spread to fixation before new mutations arise.14. Mutations that successfully invade will typically spread to fixation.Using these assumption, we can solve for the probability that an averageresident has a good reputation at any point in time (G?), and also a rareinvader in that same population (g?).G? = G (G?)??s?S,v?VG? =?2(1??)?(1??)(?+(1??)S)(?+(1??)V (1??))+?2a20??a0(1??)(1??)(?+(1??)S)a0 = ? (1??)+? +(1?? )(1??)V (1? ?)g? = G (G?)Using this we can readily express the average fitness of residents andrare invaders in any given population, and the difference between them as:WInvader = W (g?, G?)WResident = W (G?, G?)|s?S,v?VWdi f f = WInvader?WResidentWith this difference in hand we can readily infer, for any given resident78population (i.e., S,V ), whether a rare mutant would have an advantage byexamining the ?Selection Gradients?:~SG(V,S) =??????dWdi f fdvdWdi f fds????????????????s=S,v=VNotice that we evaluate these derivatives at the point where the invader?straits are identical to the resident?s. Effectively we are taking a first orderTaylor series approximation of the evolutionary dynamics at the vicinity ofthe resident?s traits. That is, we are linear approximating the effect of smallmutations in S and V on fitness.These selection gradients are a powerful tool and let us readily map outthe dynamics of the system for any given parameter combination. In thesupplemental Mathematica file we provide interactive plots of these selectionlandscapes, where they can be interactively observed as parameters change.We can also glean useful insights by analysing the selection gradients an-alytically. We have found that to make this analytically tractable we needto a) considering the limiting cases of NIRV (? = 0) and NIRM (? = 1) inturn and b) consider a simplified system where errors do not occur (i.e.,? = ? = ? = 0). We have numerically compared this simplified approxima-tion to dynamics with reasonable errors (? ? ? ? ? ? 120) and found thatthey generally describe a qualitatively identical system, where equilibriumlocations and null clines are only slightly perturbed by errors. There is oneexception. Assuming away accidental defection (? = 0) causes a special de-generate case when residents are entirely undisposed to exploit (S = 0). Inthis case nothing ever worsens residents? reputations and so there is no rea-son to ever volunteer (V? ? 0). However even very small rates of accidentaldefection can lead to substantial rates of equilibrium volunteering.To tackle this we proceed by analysing internal dynamics in the specialcase of no errors and to show that the internal equilibrium is unstable anddynamics will eventually drive dispositions to their boundaries. We then79consider each boundary in turn and explore the (S = 0) with making theseassumptions.Unstable Internal EquilibriaThe first step is to find this system?s equilibria, dispositions that to not leadto evolutionary change (V? , S?), for NIRV and NIRM respectively, by solving:~SG(V? , S?) = (0,0)T~SG(V?NIRV , S?NIRV ) = ~SG(V? , S?)???=0=???????d(2?k?(k+t)?(1??)t2)+2??t(k+t)22(1??)?(k+t)(dk+t(k+t))dkdk+t(k+t)??????~SG(V?NIRM, S?NIRM) = ~SG(V? , S?)???=1=??????? ?1??d(1??)?2k?d(1??)??????or??????d(1??)t2?2k?(?k(d+t)?dt)2(1??)k2?(d+t)d(1??)?2k?(1??)(d+t)??????While NIRV contains a single equilibrium, NIRM contains two. However,notice that one equilibrium lies strictly outside the bounds of the system (at? ?1?? ), and so regardless of its stability, dynamics around this equilibriumwill push the evolving variable to their boundaries rather than some internalpoint.To make this inference about the other equilibria we can assess their ?con-vergent stability?. That is, we ask whether selection will drive dispositionstowards or away from these points. We do this by assessing the Jacobianmatrix of these gradients. That is, if ~SGV and ~SGS are the components of theselection gradients that correspond to selection in the V and S dimensions80respectively:J =???????~SGV?Vd ~SGVdS?~SGS?Vd ~SGSdS??????By evaluating the Jacobian at an equilibrium we express the rate ofchange in the fitness difference as resident populations approach this pointin two-dimensional space.The Jacobians evaluated at the NIRV and second NIRM equilibria are:J???=0,V=V?NIRV ,S=S?NIRV =??????? t(dk+t(k+t))(2t(k+t)+d(2k+t))(??1)2d(k+t)2(2k+t)(dk+t(k+t))(dk(2k+t)+t(k2?t2))(??1)?dt(k+t)(2k+t)? t(dk+t(k+t))(??1)?d(2k+t)2k(k+t)(k+2t)(dk+t(k+t))(??1)2?2dt2(2k+t)(??1)??????J???=1,V=V?NIRM ,S=S?NIRM =??????t2(d+t)2(??1)2(??t+t+k?)2k(??t+t+2k?)(d(??t+t+k?)?kt?) ?k(d+t)(??1)(??1)?(2d(??t+t+k?)?t2(??1))2(t(??1)?2k?)(kt?+d(t(??1)?k?))? k2t(d+t)2(??1)(??1)?2d(??t+t+2k?)(d(??t+t+k?)?kt?)k3(d+t)(??1)2?2((??1)d+2k?)d(??t+t+2k?)(d(??t+t+k?)?kt?)??????For NIR?s equilibria to be convergently stable all the eigenvalues of thismatrix must be negative (i.e., dynamics must drive the system back towardsthis point in all dimensions). This will only be true if the determinant of theJacobian (i.e., J11J22? J12J21) is positive (see Routh-Hurwitz Conditions inOtto & Day, 2011, chapter 8). However, for both NIRM and NIRV, a littlealgebra shows that this is never true. That is, NIR?s internal equilibriumis never stable. Evolutionary dynamics will always drive populations awayfrom this point and towards the boundaries (i.e., where S = 1, S = 0, V = 181or V = 0) of this system. Our careful numerical investigations of manyparameter combinations suggest that this is also always true for intermediatevalues of ? and reasonable error rates.To get a more intuitive sense of these dynamics, we provide interactivephase-space plots in the Mathematica supplement and static versions fortheoretically interesting parameter combinations in Figure 3.8.Boundary Dynamics: The S-null-cline is always unstable.First a little algebra shows that, just as under a discrete strategy interpre-tation, a world of complete defection must always be stable. Specifically:~SGS??S=1 > 0If no-one respects reputations, selection will never favour a shift awayfrom this situation. Next we find the location of the system?s null-cline inthe S-dimension. That is, the curve at which ~SGS = 0. This expression tellsus, for a given V , the value of S at which selection gradients are flat. See theMathematica supplement for the full expression and Figure 3.8 below for avisualisation of this curve.We find a single S-dimension null-cline (see Mathematica supplement).Given this, and the fact that selection must push upwards at the S = 1 edge,we know that when this null-cline is within the bounds of the system (i.e.,0 < S < 1) it must be always unstable, or dynamics would not continue topush the system against its upper S-bound. Equivalently, we know it mustbe stable when it is located at S > 1.8283NIR-P(? = 1)NIR-V(? = 0)NIR-M(? = 1)? =1100.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 = 120.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 3.8 (preceding page): Selection gradients for three version of NIR (columns) in two different pa-rameter regions (rows), depicting how selection would shape continuously evolvingdispositions to exploit well-reputed peers (S, vertical dimension) and volunteer toimprove your reputation (V , horizontal dimension). Orange lines depict the ap-proximate (when errors are zero) null-cline in the S dimension; dispositions to ex-ploit above this line tend to get greater, dispositions below (in the shaded region)get smaller. The red dots indicate the precise location of stable V equilibria alongthe S= 0 and S= 1 edges; dispositions to Volunteer tend to approach these points.Parameters are set to: k = d = 1; t = 14 ;? = 110 ;? = ? = ? = 120 . Here one can seethat though NIR-M sustains higher rates of volunteering in cooperation-favouringparameter regions (e.g., top row), NIR-V continues to suppress exploitation evenwhen costly-reputation improvement does not pay (e.g., bottom row).Boundary Dynamics: The evolution of V.Since we know the system has a single unstable null-cline running throughit?s S-dimension, we know dynamics will eventually push it to S = 0 or S = 1.What will happen to V on these edges?When S = 1, the selection gradient for volunteering ( ~SGV ) is strictlynegative, specifically it is (1? ?)(?k)?. This makes sense. If reputationdoes nothing to stop others exploiting you, why would you pay to maintainit? Consequently when exploitation is rife, volunteering will be eliminated.When S = 0, we can find the location of the V equilibrium (call it V?S=0)by solving for when the selection gradients are zero. While we do findprecise analytic solutions for these points, they are again long and unwieldyexpressions that offer little insight on their own. Instead, here we present thelimiting cases of NIRV and NIRM, and below we show the consequences ofthis equilibrium on actual volunteering rates (i.e., weighted by ? and 1??).Equilibrium selectionWe?ve seen three different ways that early reputation-using societies?those inwhich individuals? opinions of one another were somehow coordinated?mayhave structured their interactions: NIR-P, NIR-V and NIR-M. It is naturalto ask, if these societies coexisted, what would be their consequence of theirinteractions? Which would thrive and which would be outcompeted? Wehave two pieces of information that could help us answer this question.The first is the reputational equilibrium (i.e., stationary distribution)at the cooperator behavioural equilibrium. Those societies that have theleast internal exploitation may have a competitive advantage. For societiesat their cooperative equilibria ?least exploitation? means ?fewest individualswith bad reputations?. Here, NIR-M is transparently outcompeted by NIR-Vand NIR-P, since it includes an additional mechanism for worsening repu-tations. The relative standing of the two deliberate-improvement models toNIR-P depends on just how many more costly opportunities there are forreputation improvement than costless ones (i.e., ?), which is an empiricalquestion.85There is another feature of NIR that can inform us about long runbetween-society interactions. It is not implausible that (at least some of)the reputation-boosting acts we modelled are actually instances of positivecooperation. That is, individuals bearing costs to improve the fitness of theirpeers.What is the relative rate of volunteering (e.g., gleaning favour by proso-cial giving) between NIR-M and NIR-V? Here we can use the equilibriumvolunteering rates at the non-exploitation boundary V?S=0 to gain some in-sight. This is presented in Figure 3.9. As opportunities for prosociality in-crease, the incentive to give declines in NIR-V as individuals less often findthemselves with bad reputations, but continues to increase in NIR-M. How-ever as volunteering costs (weighted by opportunities) become too onerous,NIR-M can suddenly collapse to it?s full-defection equilibrium, while NIR-Vdoes not.In general, this suggests that as long as exploitation is common andinefficient, NIR-M systems should sustain higher rates of volunteering thanNIR-V systems, and so be better candidates for victory in between-groupcompetition.3.2 Additional supplemental materialsAdditional supplemental materials are provided as computer files.They include:? R-code for running reputation simulations and numerically approxi-mating equilibrium locations in the discrete-strategy interpretation.? A Mathematica file (proprietary symbolic algebra software), by meansof which the main results can be readily rederived.86.0.0 0.1 0.2 0.3 0.4 S=0???(1??)Figure 3.9: Relative volunteering rates for NIR-V and NIR-M. Rel-ative rates of costly reputation-raising contributions to others?welfare maintained by NIR-V (red lines) and NIR-M (blue lines)at their respective cooperative equilibria. Lines extend as longas the cooperative equilibrium is stable (i.e., ~SGS??S=0 < 0), whenkd is 2 (unbroken lines), 1 (dashed lines) and 12 (dotted lines).Here d = 1, t = 14 ,? = 110 and all errors (? ,? ,?) are 120 . Noticethat if costs become too onerous, cooperation under NIR-M col-lapses while NIR-V continues to sustain cooperate as volunteer-ing gradually declines.87Chapter 4Surveying indirectreciprocity: how do peopleassign reputations?Humans can be impressive cooperators. Our metropolises are crowded withanonymous passers-by whom we could swindle, waylay, defraud, pickpocket,intimidate or otherwise exploit. Yet, we don?t. Instead we hold the bus doorfor them, offer them directions, give them alms, and sometimes even risk ourlives to save theirs. Of course, it hasn?t always been so, and some contempo-rary societies are still fraught with bandits, thieves, charlatans, rapists andcorrupt officials. Recently considerable scholarly attention has been appliedto explaining how human cooperation arose, targeting questions like: how iscooperation sustained, why do our patterns of cooperation differ from otherspecies, and why do they vary so dramatically between societies? Why dosome groups thrive in mutual aid and cooperation, while others collapseunder the weight of distrust and mutual exploitation?Any complete answer to these questions will likely feature contemporarycultural institutions. For instance: police forces, laws, courts, debt, taxa-tion, intellectual property, democratic elections, labour unions and parkingfines. However such institutions cannot be the root of the explanation, sincethey themselves depend on many cooperative interactions. Police officers,88judges and vote-counters usually stand to gain far more than the averageperson from corruption and deceit. If sufficiently many people ignored them,laws, debts, taxes and parking fines would be unenforceable. Our institu-tions seem to hold one another aloft. Together they free us from worryingabout whether each stranger, shop keeper or new friend will betray us. Butif each institutions depends on some or many individuals already being co-operatively inclined, how did whole flotilla become buoyant in the first place,and why have other animals not followed similar trajectories?Reputations, and reputation-based cooperation, play a central explana-tory role the three main types of explanations that interested thinkers haveposited.The first kind places causal primacy on our extraordinary intelligence(e.g., Pinker, 2010). First we became smart, then we figured out how to co-operate, conceived of and designed clever institutions, and so on.1 Howeverclever ancestral humans may have been, these individual-intelligence-basedexplanations must be coupled with population-level accounts of how earlyhuman groups first escaped uncooperative equilibria (situations in whichevery individual benefits most by being uncooperative). Intelligence alonecannot sidestep these population-level questions; after all, cleverer-still con-temporary humans also struggle with such dilemmas. One simple gambitis to claim that language allowed our ancestors to talk and, especially, togossip. Language and intelligence allowed them to coordinate community-wide response to uncooperative individuals, making cooperation an individ-ually fitness-enhancing, smart choice. Since formal models of this processbranched off from models of ?direct? reciprocation between pairs of individ-uals (e.g., Axelrod & Hamilton, 1981; Boyd & Lorberbaum, 1987; Doebeli& Hauert, 2005; Trivers, 1971; van Veelen et al., 2012, more generally, theseare models of viable strategies for repeated interaction between pairs of in-1These intelligence-based arguments, though they are perhaps the most common folk-explanation, face many unsolved theoretical challenges. For instance, how could selectionpressures produce extraordinary intelligence, bearing the concomitant costs of big brainsand smaller guts (Aiello & Wheeler, 1995; Kotrschal et al., 2013), unless people werealready cooperative enough that they freely shared the cultural knowledge that makes bigbrains worth having?89dividuals), this population-level reputation-coordination process is usuallycalled Indirect Reciprocity (IR).A second category of explanations of human cooperation give casual pri-macy to cooperation-sustaining mechanisms continuous with other species,especially kin-selection and direct reciprocity. These are sometimes called?misfire accounts?, because they posit that we, effectively, mistake strangersfor kin or long-term interaction partners. Our psychology, these accountsposit, is calibrated to an ancestral past where most our interaction part-ners were kin or stuck around a long time. They have yet recalibrate tothe anonymous cities and transcontinental trade networks of the Holocene2.These misfiring cooperative dispositions provided the foundation for moresophisticated cooperative institutions including, possibly, the deception-freesharing of valuable knowledge that allowed our ancestors (and us) to be-come so smart in the first place. However, kin-selection struggles to sustaincooperation as phylogenetic distance increases, and direct reciprocity is onlyeffective in dyads or small groups(Chudek et al., 2013b). These explanationsmust also be paired with an account of how, even in a cooperatively misfir-ing species, large-scale cooperative institutions first arose. Again, reputationprovides a parsimonious bridge from individual psychological dispositions toself-sustaining population-level dynamics.A third kind of explanation shifts the focus from just the evolution of ourpsychology to the evolution of our institutions themselves. Such culture-genecoevolutionary accounts start by explaining the genetic adaptations thatmay have led humans to transmit cultural information in the first place; andto do so well enough that culture began accumulating and evolving acrossgenerations into technologies and institutions that no human could devise ontheir own. These accounts then model how our evolved psychology biases theevolution of our cultural knowledge, and how that in turn biases the geneticpressures on our psychology. They derive predictions about contemporarypsychology, behaviour and institutions based on the long-term consequencesof these two interacting evolutionary processes. Since reputations?which2Misfire arguments, though common among evolutionary psychologists, also face theirshare of theoretical and empirical challenges (Chudek et al., 2013b; Fehr & Henrich, 2003)90socially convey information about one?s peers?are themselves a kind of cul-tural information, the question of how reputations can sustain cooperativecultural knowledge sharing is particularly relevant to culture-gene coevolu-tionary theorists.While the first two kinds of explanations merely assume that reputa-tional information is freely and accurately transmitted among individuals,culture-gene coevolutionary theorists have a long history of confronting thecooperative dilemmas inherent in cultural transmission. They have proposedseveral mechanisms that could sustain generalised cooperation, includingthe quality of reputational information, among a cultural species; including:conformity (Henrich & Boyd, 2001), deference (Henrich & Gil-White, 2001),and credibility enhancing displays (Henrich, 2009).However you prefer to explain the foundations of human cooperation,reputation is likely to play a central role. Next I review the main theoreticalinsights about the population processes by which reputation brings aboutcooperation (Indirect Reciprocity, IR). I then turn to testing these predic-tions against contemporary people?s actual intuitions. While I synthesise IRtheory from a culture-gene coevolutionary perspective, the empirical testsare relevant to all perspectives on human cooperation.Reputation, in theoryAt first glance, reputations seem a simple route to cooperation. If you do badthings to others, your community will eventually find out and start doingbad things to you. In the long run, the reputational costs of exploitative oruncooperative acts outweigh the immediate payoffs and good guys prosper.But formal explorations of reputations? population-level dynamics quicklybecome quite complicated. To help unfamiliar readers keep a handle on thiscomplexity, I provide a glossary of important terms in Table 4.1.First, there are many different ways that reputations could influencebehaviour?different action-rules. That is, there are many different be-havioural strategies people could employ, given their own reputation andthat of the potential target of their actions. Imagine that your commu-91CooperativeDilemmaA situation where an Actor could chose between a be-haviour that advantages them, and one that advantagesone or more others more.Prisoners?dilemmaA cooperative dilemmas with only two individuals in-volved.PositiveDilemmaA cooperative dilemma where the active Action yields acost for the actor and larger benefit for the Target(s),relative to doing nothing.NegativeDilemmaA cooperative dilemma where the active Action yields abenefit for the actor and larger cost for the Target(s),relative to doing nothing.Actor The individual who makes a choice in the focal coopera-tive dilemma.Target The individual affected by the Actor?s action.Action What the actor actually does, cooperating or defecting.Cooperating Choosing the other-advantaging choice in a cooperativedilemma.Defecting Choosing the self-advantaging choice.Helping Cooperating in a positive dilemma (cf. passively defect-ing).Exploiting Defecting in a negative dilemma (cf. passively cooperat-ing).Reputation A cognitive representation of an individual which influ-ences how one treats them in a cooperative dilemma.Judge An individual assessing a cooperative dilemma, and re-vising their representation of the Actor?s reputation.Action-rule A strategy for how to Act in cooperative dilemmas, giventhe reputation?s involved.Assessment-ruleA consistent system for judging the Actor in a cooperativedilemma, given their action (1st-order judgement), theirTarget?s reputation (2nd-order judgement) and their ownprevious reputation (3rd-order judgement).Table 4.1: Glossary92nity is particularly gossip prone and reputations really matter. Should youalways cooperate with others? Would you be better off only cooperatingwhen your reputation is bad? Or perhaps only when the potential target ofyour act is good? The consequences of these choices depends on how yourcommunity judges these acts?their assessment-rules.Imagine that one of your well-reputed peers had helped a poorly-reputedtarget (e.g., loaned them some money). Would you (and your community)lessen your opinion of them for consorting with scum, or glory them as aGood Samaritan? Say they had exploited a bad person (by stealing from athief, for instance), would you think them as virtuous as Robin Hood or as avillain for their crime? It is not obvious a priori how individuals and commu-nities should reputationally judge one-another?s behaviour, let alone how asocio-ecology populated by many different action-rules and assessment-ruleswould unfold.Early formal IR theory considered a simple ?first order? assessment-rulecalled ?Image Scoring? (Boyd & Richerson, 1989; Nowak & Sigmund, 1998).Under Scoring, any cooperative act improves reputations and any defection(i.e., uncooperative act) worsens them, regardless the Actor?s (i.e., individualacting) and Target?s (i.e., individual acted upon) prior reputations. Scoring?saction-rule is to only cooperate with Targets whose reputation is sufficientlygood.However further investigations revealed that Scoring is not evolutionar-ily stable (i.e., it will be out-competed by other strategies) if people some-times make mistakes (Leimar & Hammerstein, 2001; Panchanathan et al.,2003). That is, if sometimes individuals intend to cooperate but don?t man-age to (?I really did mean to return your shovel, and I really did misplaceit?), a Scoring-based community would exploit one-another often enoughthat unremitting defectors (individuals who never cooperate with anyone)could prosper in their midst. This happens because Scoring-communitiesreputationally-punish anyone who defects even on unequivocal scoundrels.This fact inspired theoretical work on more sophisticated ?Standing? strate-gies. More recent work has described more sophisticated variants of Scoring(Berger, 2011; Rankin & Eggimann, 2009), or specific kinds of interactions93(Suzuki & Akiyama, 2007a,b), or ecologies (Brandt & Sigmund, 2006) whereScoring strategies can persist.Standing assessment-rules are 2nd-order, they are sensitive to Targets?reputations. The earliest Standing models (e.g., Panchanathan & Boyd,2004) assumed that cooperative acts always improve reputations, while onlydefecting on good people worsens them; defecting on bad people leaves rep-utations unchanged. Standing?s action-rule is just like Scoring?s. However,unlike Scoring, even if people occasionally defect when they intended to co-operate, Standing-individuals in a Standing-community will do better in thelong run than unremitting defectors.Ohtsuki and Iwasa (Ohtsuki & Iwasa, 2004) surveyed all 4064 possiblecombinations of action-rules and assessment-rules, when reputations are bi-nary (individuals can be good or bad, nothing in between) and individualscan either cooperate or defect, no other reputation-relevant events occur.They found that eight of those stood out as particularly evolutionarily vi-able (they could not be invaded by other strategies, and maintained highlevels of cooperation) and called them the ?leading eight?.3 Like Standing, theleading eight like cooperation with good Targets (it improves reputations)and dislike defection on good Targets (it worsens reputations). What tellsthe leading eight apart is how they judge actions (cooperation or defection)on bad Targets.The mathematical techniques available to IR-theoreticians offer clear in-sights into long-term consequences when entire groups use just one assessment-rule, even if there are also unremitting cooperators or defectors present, orif sometimes individuals make genuine mistakes. However the complexitiesthat result from many different assessment-rules interacting with a singlecommunity quickly become analytically intractable (cf. Uchida & Sigmund,2010, for a single exception).Two other important developments in IR-theory bear mention. Oncereputations exist, they can be linked to any arbitrary behaviour (not just3These strategies have been dubbed by an ad hoc assortment of, sometimes inconsistent,names. For brevity and clarity I will refer to them all as Standing-strategies, since theyare all (at least) second-order IR reputational systems.94cooperative dilemmas) to prop up its evolutionary stability (Panchanathan& Boyd, 2004). Imagine for example a reputation-based community at aprimary school, where individuals with particularly bad reputations (losers)are picked on and exploited by their peers, while individuals with partic-ularly good reputations (cool kids) receive disproportionately preferentialtreatment. Being mean to cool kids worsens one?s reputation, while beingmean to losers does not. These ingredients are enough, in theory, for thereputational system to persist at the school, even as generations of chil-dren who instantiate it come and go. Now imagine that one generationdecides, by whatever dynamic processes such decisions emerge, that any-one who likes the television programme ?Sesame Street? is a loser. Givenits reputational consequences, this distaste for Sesame Street could persistacross many generations of children at the school, even though it is neither acooperative dilemma itself, nor is it directly connected to reputationally in-fluenced cooperative and exploitative acts in this community. Analogously,reputations could support the persistence in historical time of maladaptivecultural traits that are otherwise difficult to explain (such as footbindingor female infibulation). More importantly, while IR-itself only stabilises co-operation among dyads, this potential for linking reputations to arbitrarybehaviour provides an avenue for IR to also prop up larger scale cooperativeventures, such as selfless sacrifice in war (e.g., Mathew & Boyd, 2011), orpublic maintenance of more sophisticate cooperation-sustaining institutions(e.g., Sigmund et al., 2010).Second, recent work has drawn attention to an important difference be-tween positive cooperation (cooperating by helping others, or defecting bydoing nothing) and negative cooperation (defecting by exploiting others, orcooperating by doing nothing). For brevity, I call this the valence of co-operation. While early work implicitly treated these situations as identical(or, at least, does not explicitly draw a distinction between them),Chapter3 highlights important differences. While it is fair to assume that explicitaction (i.e., actively cooperating by giving a gift or exploiting by stealing)could carry reputational consequences, the same does not hold for passiveinaction (e.g., cooperating by not stealing someone?s unguarded valuables,95or passively defecting by not warning them about an encroach danger youforesee). Such inaction is likely to go unnoticed (and so unpunished or unre-warded) by anyone other than the potential actor, unless a community haswell coordinated expectations about what opportunities exist, when, andwhat the normative responses are to them. Since such coordination of socialnorms to non-events is unlikely early in the coemergence of cooperation andculture, Chudek & Henrich conclude that negative cooperation in particu-lar may have played a key, early role predict that our psychology may beparticularly attuned to noticing and responding to it.Reputation, empiricallyMost empirical tests of IR have been done in the context of experimen-tal games. Participants typically compete in computer-mediated repeated,positively-valenced prisoners? dilemmas (i.e., they can pay a cost to create alarger benefit for someone) with anonymous strangers for small amounts ofmoney. In these settings, investigators give participants access to differentkinds of potentially reputation-relevant information (such as their target?shistory of cooperating or defecting with others) and measure what effect ithas on their choices.There are several reasons that these only studies provide limited insightinto whether and how IR structures interactions in contemporary popula-tions, let alone how it may have influenced our ancestors whether it left asignature on contemporary cognition.Incomplete Assessment The majority of studies (e.g., Charness et al.,2011; Engelmann & Fischbacher, 2009; Seinen & Schram, 2006; Simp-son & Willer, 2008; Sommerfeld et al., 2008, 2007; Wedekind, 2000;Wedekind & Braithwaite, 2002) only provide participants with 1st-order reputational information (the actions of the Actor), and so can-not infer whether participants use 2nd- or higher-order assessment-rules. A few studies have provided participants information on Tar-gets? reputations, usually as a list of cooperations or defections by eachTarget, sometimes at a cost (Ule et al., 2009). Though early work96found no evidence of 2nd-order Standing-like assessment, subsequentwork (Bolton et al., 2005) found the opposite?that Standing assess-ment is common enough that it can substantially skew Actors? out-comes. While these 2nd-order studies are valuable, they still typicallyonly investigate one or two of the many possible 2nd-order assessmentrules described by theory.Mono-cultural These studies are mono-cultural and typically use the glob-ally peculiar sample of Western university undergraduates (Henrichet al., 2010). These leaves them poorly positioned to draw inferencesabout how universal reputation-assessing intuitions are, and by exten-sion, how IR may have influenced early human evolution.Exclusively positive dilemmas Perhaps due to tradition, these studiesframe games exclusively as positive dilemmas?participants can helpeach other make more money. Where negative interactions are consid-ered (e.g., G?chter & Herrmann, 2009; Milinski & Rockenbach, 2012;Ule et al., 2009), they appear as costly punishment (where Actors paya cost to disadvantage Targets) rather than as negative cooperativedilemmas. Existing studies are not in a position to judge whetherdilemma valence matters for IR.Competitive, artificial context People?s reputational intuitions may beculturally transmitted (e.g., by imitation of parents? or peers? actionsand assessments). If so, they would take the form of context-specificbehavioural rules and response, and may differ between, say, the work-place, the bar and the family home. Measurements such context-specific variation are valuable for evaluating cultural-evolution inter-pretations of IR theory. Existing studies put participants in an un-usual, highly artificial context where they know they are expected tocompete with strangers for small sums of money while observed byscientists. Such contexts could evoke their distinct behavioural strate-gies and normative responses, concealing patterns of IR that governsparticipants? lives outside the lab.97Anonymity A substantial drawback of experimental games is a conse-quence of one of their greatest strengths. It is standard practice inexperimental games to ensure participants? anonymity. This helpsassure investigators that experimental variables are influencing par-ticipants? responses, rather than concerns about what others mightthink of them after the study. However IR theory presupposes a socio-ecology entirely unlike this. IR emerges among communities of indi-viduals who know each other well, interact for large portions of theirlives, are interested in accumulating gossip and detailed informationabout one-another?s behaviours in many domains, and have detailedcognitive representations of each other that inform their behaviouralchoices. Consequently, representations of the specific target and actormay be needed to evoke participants? IR-relevant cognitions.Writers, story-tellers and movie-makers are masters of evoking emotions,judgements and sympathies in their audiences. To create sentiments of righ-teous approval when a hero vanquishes a villain, or abandons them to theirdemise instead of helping, they first create a relationship between the audi-ence and the villain (e.g., by showing the villain delightedly spraying pepperin the eyes of yelping puppies). The audience?s subsequent approval whenthe hero?s actions may depend implicitly on their own vitriolic cognitiverepresentations of the villainous Target. Since IR is inherently a system oflong-term interactions with well known individuals, merely reading a list ofanonymous others? defections, even a second order list, may be insufficientto evoke participants? IR reputation assessment-rules.One way to side-step these concerns is to measure reputation changein naturalistic settings, or better, natural experiments. However this ap-proach can only evaluate the subset of assessments that correspond to whatis happening ?in the wild?. One straightforward solution is to systematicallydescribe the full set of situations relevant to IR, and simply ask people howthey would respond to them. The effectiveness of this approach depends onwhether people can, and are willing to, accurately simulate and report theirreputational assessments of a verbally described situation. Even if these98self-reports are imperfect, they provide a valuable complement to existingexperimental-game-based tests of IR.Of course, ideally we would like to strengthen our inferences by findingways to conduct experimental games that overcome the limitations above.However before attempting this costly and difficult feat, it behoves us (andis our ethical due to the tax-payers who fund our investigations) to initiallysurvey this landscape using the most cheap and direct method available tous: simply asking people. That is what I have undertaken here.Present studyBy asking people for their reputational judgements in precisely those situa-tions described by IR-theory, I aim to address these questions:Scoring or Standing? Are people?s reputational judgments informed by2nd-order information (e.g., defecting on bad guys is ok), or exclusively1st-order (e.g., defecting is always bad)?Distribution of reputational systems How are people?s reputational judg-ments distributed across the space of all possible 1st-order and 2nd-order reputational systems?Does the valence matter? Do reputational judgments differ between pos-itive cooperative dilemmas (where someone can help others or do noth-ing) than they do in negative ones (where they can exploit others or donothing)? How, specifically? Is there covariance with the valence ofpeople?s responses (e.g., do people prefer negative responses (stealing)to defection in negative dilemmas)?Are there cross-cultural differences? Do people from phylogeneticallydistinct cultures use different reputational systems?What are the covariates of reputational judgments? Do people?s rep-utational judgments systematically covary with the source of their in-formation (gossip or direct observation), that actual consequences forthe target independent of the intended act (a good outcome, a bad99one, or ) and participants? individual differences (including age, sex,religion, education and time taken to make decisions)?Genes, culture or individual learning? Which of these three mecha-nisms were more likely to have produced the patterns of variabilitywe observe in people?s IR strategies?4.1 MethodsTo infer each participant?s reputational intuitions, I solicited their judg-ments about two scenarios (available in the supplemental materials). Ineach scenario I asked participants to imagine arriving in a new situationand meeting four new people, either a new workplace or a new group offriends. First I introduced them to a ?Target? character, who was cued ei-ther ?good? or ?bad?. I did this by describing their behaviour (cooperating ordefecting) in three dilemmas: an n-person public goods dilemma, a pairwiseasynchronous prisoner?s dilemma, and an ?information dilemma?, where theTarget possess information that they could benefit by keeping secret or theycould share to the benefit of others (see Table 4.2).100101Setting Type Good Target Bad TargetWorkplace Public GoodsDilemmaShe is often the one who cleans the small, sharedkitchenette at work, even though she?s rarely, ifever, the one who creates a mess.She often makes a mess in the small, shared kitch-enette at work, but you?ve rarely, if ever, seen hercleaning it.PairwisePrisoner?sDilemmaWhen she works on a project with another work-mate, she always makes sure to share the creditif it goes well and willingly shares the blame if itgoes badly.When she works on a project with another work-mate, she tries to take all the credit if it goes well,even though she is rarely willing to share in theblame if it goes badly.InformationDilemmaWhenever she meets a valuable client she putsthem in contact with other relevant team mem-bers, even though she could get an advantage bybeing their exclusive point of contact.Whenever she meets a valuable client, she triesto prevent them contacting other relevant teammembers, so she can gain an advantage by beingtheir exclusive point of contact.Friends Public GoodsDilemmaA few weeks ago she offered to buy tickets toan event for several of your new friends. Lateryou found out that the tickets had actually costmore than she had asked your friends to give her.She had lost money so that the tickets would becheaper for her friends.A few weeks ago she offered to buy tickets to anevent for several of your new friends. Later youfound out that the tickets had actually cost lessthan she had asked your friends to give her. Shehad made money by having her friends pay morefor the tickets than they originally cost.PairwisePrisoner?sDilemmaWhenever someone is moving, she is almost alwaysthere to help them pack and move their furnitureto their new house, even though she has nevermoved herself.She has moved several times and had other friendscome and help her pack and move her furniture toher new house, even though she has always beentoo busy to help anyone else when they were mov-ing.InformationDilemmaShe loves shopping. Whenever she hears about agreat sale, she always tells all her friends aboutit too, even though doing so makes it more likelythat the sale items will run out before she findswhat she wants.She loves shopping. When she hears about a greatsale, she never tells all her friends about it untilshe has had a chance to buy everything she wants,even though doing so makes it more likely that thesale items will run out before her friends find whatthey want.Table 4.2: The cues used to establish the Target?s reputation, classified by type and scenario. Here I referto the Target, whose name which varied between conditions, as ?she?.Next, participants were introduced to three Actor characters. Each Actorfaced an identical cooperative dilemma, either positively valenced (oppor-tunity to help the Target) or negatively (opportunity to exploit the Tar-get). One Actor (the Cooperator) helped (positive valence) or didn?t exploit(negative valence) the Target. Another (the Defector) did the reverse. Theorder in which participants read about the Cooperator and Defector wasrandomised for each participant, but always reversed in the second scenariothey read. A third individual (the Neutral) always faced the dilemma afterthe Cooperator and Defector, but was interrupted by the Target before theycould act.Participants then were asked to make four judgements about each ofthese four characters (our dependent measures). Each judgement was madeby moving a slider (initially at its mid-point) or along a horizontal linebetween two boundary labels. Slider position was encoded as an integerbetween zero and one-hundred, with higher numbers representing morefavourable (?better?) judgements, which was then rescaled to the unit in-terval to facilitate beta-regression (Cribari-Neto & Zeileis, 2010; Ferrari &Cribari-Neto, 2004; Smithson & Verkuilen, 2006).Our dependent measures were:Good How good a person they believe each individual is. Boundary labels:?Good? and ?Bad?.Like How well they think they would like each individual. Boundary labels:?Like? and ?Dislike?.Favour How likely they would be to do a slightly inconvenient favour forthe individual. Boundary labels: ?Do Favour? and ?Don?t do Favour?.Money How likely they would be to return ten dollars that the individualhad dropped. Boundary labels: ?Return? and ?Keep?.The first two question were intended to tap cognitive representations ofreputation. First, by directly asking how ?good? each individual is. Second,since terms like ?good? may evoke normative answers that map imperfectly to102participants? implicit sentiments, by asking about a vaguer sense of ?liking?.The third and fourth questions were designed, to measure the behaviouralconsequences of reputation. The third solicited participant?s behaviour inthe positive dilemma of doing someone a favour. Soliciting participant?sbehaviour in negative dilemmas (i.e., opportunities to actively exploit some-one) was challenging, since such behaviour is strongly proscribed in mostcontemporary societies. Based on pretesting, I settled on ?not returningmoney? as the closest approximation to the prototypical negative dilemmaof ?stealing? which still generated at least some variability in people?s an-swers.At the end of the study, I gave people a limitless text box and askedthem to explain to us their rational for the choices they had made. Theseare available in their entirety in the supplemental materials.I systematically varied the following aspects of this core design:Random between scenariosThe following variables were randomly assigned to each participant in theirfirst scenario and were different in the second scenario:Names The names of the Target and Actors (picked from one of two fixedsets: Tracy, Ashley, Betty, Cassie; and Stephanie, Judy, Katie, Lucy).Setting The setting in which the events took place (among friends or inthe workplace).DV Order The order in which each of the four questions was presented.First Actor Whether the first Actor to face a dilemma is the Cooperatoror Defector. The second Actor is always the other of this pair, whilethe third is always Neutral.Manipulated between participantsThe following variables were explicitly manipulated between conditions:103Target Reputation Order The reputation of the Target (either good orbad) in each scenario. All participants saw one scenario with a goodTarget and one with a bad Target. Which they saw first systematicallyvaried between conditions.Source The ostensive source of the information the participant is reading:their own experience, or gossip from a reliable source.Consequences The final consequences for the target: none mentioned, bad(a social event fails) or good (a social event succeeds).Valence The valence (positive or negative) of the dilemma faced by Actors.In positive dilemmas, the Actors could pay a cost to provide a benefitto the Target. In negative dilemmas, Actors could extract a benefitby imposing a cost on the Target.Here is an example scenario. In this instance, the Target (Stephanie)is cued as having a good reputation, the Source of the information isgossip and the Consequences are bad for Stephanie. The three Actorsare Judy, Katie and Lucy. The difference between positively and negativelyvalenced variants of the dilemma are indicated with square brackets andbolded text. In the positive variant, Judy cooperates by doing something,in the negatively valenced version she defects by doing something. Katiedoes the opposite, but achieves it by inaction. Which of these (action orinaction, cooperation or defection) happened first was assigned at randomand controlled for in regressions.In both settings, positively and negatively valenced versions of the dilemmadiffered only in whether someone considered adding or taking pieces of paperfrom an unattended pile.Imagine that you?ve recently started a new job and are getting toknow your workmates. You?ve again met several people and learned alittle about them.One of the first people you met was Stephanie. You?ve noticedseveral things about Stephanie.104Stephanie is often the one who cleans the small, shared kitchenetteat work, even though she?s rarely, if ever, the one who creates a mess.When Stephanie works on a project with another workmate, shealways makes sure to share the credit if it goes well and willingly sharesthe blame if it goes badly.Whenever Stephanie meets a valuable client she puts them in con-tact with other relevant team members, even though she could get anadvantage by being their exclusive point of contact.Recently you witnessed some very interesting interactions involvingStephanie and three other individuals: Judy, Katie and Lucy. Youheard about these events from a friend who managed to witness it allunseen. You trust this friend and are confident that this is all true.One day at work Stephanie wanted to finish early because she washosting a social function that evening that she needed time to organise.However Stephanie also had a very large stack of paperwork that sheneeded to finish before she could leave for the day. In fact, everyonein the office has to do identical paperwork from time to time, butStephanie happened to have a lot due that day. Everyone could seethat Stephanie was very anxious because if she didn?t have enoughtime to prepare for her social function it would likely be a failure.Even though each employee?s paperwork is indistinguishable, it is strictcompany policy that each employee must do their own, so Stephaniecould not ask others for help.In about the middle of the afternoon, Stephanie had to go to ameeting. She left her stack of paperwork on her desk unattended.First, Judy came by and noticed the paperwork. She looked aroundand thought that no-one was watching. It was obvious that she could[take some of Stephanie?s paperwork and do it herself / addsome of her own paperwork to Stephanie?s pile] and, if no-onesaw her, then no-one would ever know. Judy stood and looked atthe paperwork pile for quite a while, clearly considering her options.Eventually she decided to [quickly take some of the paperworkfrom Stephanie?s large pile. Now she had more paperwork todo herself, but Stephanie had less. / quickly add some of herown paperwork into Stephanie?s large pile. Now she had lesspaperwork to do herself, but Stephanie had more.]Not long after Judy had left, Katie came by and also saw the pile.105She also looked around and didn?t see anyone watching and then stoodand looked at it for a quite while, considering the same choice as Judy.Unlike Judy, Katie eventually decided to do nothing, she merely walkedaway.Finally, after Katie had left, Lucy came by and also saw Stephanie?sunattended paperwork. She also looked around and then looked at itfor a while, but before she could reach a decision, Stephanie came back.Lucy just ended up chatting with Stephanie briefly.In the end, Stephanie did not finish the paperwork in time and hersocial event was a disaster.Operationalising reputation: dichotomous theory tocontinuous empiricismA major challenge of operationalising IR-theory is that formal models al-most exclusively consider binary reputations (you?re either good or bad,there?s nothing in between), or sometimes discrete, small-integer reputations(Leimar & Hammerstein, 2001; Nowak & Sigmund, 1998). Real people seemto, prima facie, represent reputations continuously (you can be a little bitbetter reputed than me, a little less well than my sister), or at least readilyoffer continuous answers to reputational questions.Theorising about binary reputations makes sense. Though I may rep-resent your reputation continuously, when it comes to actually using yourreputation to make a cooperative decision (e.g., to help you or walk away),I often have to make a binary choice. Either your reputation will be abovea critical threshold, or not. By simplifying reputations as ?above or belowthis threshold?, IR-models become more transparent and mathematicallytractable, without sacrificing too much generality. In binary models, thework of representing continuous changes that do not cross the threshold iscarried, implicitly, by the magnitude of various probabilistic scaling param-eters, such as the relative frequencies with which events occur.I translated models of dichotomous reputations into predictions aboutcontinuous reputational change by using the map in table 4.3. For instance,when a dichotomous theory predicts that an event would cause an individ-ual to keep a good reputation but change their bad reputation to good, I106considered this equivalent to predicting that continuous reputations would?improve? following the same event.One particular challenging translation is posed by dichotomous assessment-rules that specify that, after an event, ?good individuals become bad butbad individuals become good?. Two of the leading eight reputational sys-tems specify such reputational ?flipping?. However I have chosen to excludethis possibility from this investigation for both a theoretical and a practicalreason.Theoretically, flipping makes little sense for continuous reputations. Thereare two ways one could interpret it. First, it might imply that reputationsalways converge towards a neutral mid-point. In the long run this would, ce-teris parabis, eliminate the reputational differences that sustain cooperationand cause the system to collapse. Alternatively they might over-shoot in theopposite direction, with good individuals becoming even more bad and badindividuals more good4, causing ever more dramatic swings in continuousreputation. It is hard to imagine how this could be sustained in practice.Though both these alternatives could be ?patched-up? by additional assump-tions (such as non-linear interaction probabilities), I know of no theoreticalwork that has attempted this. I suspect such systems are unlikely to play alarge role in human cooperation, and in any case consider them beyond thescope of early empirical work.Practically, ignoring such 3rd-order reputational flipping makes the em-pirical challenge of testing IR far more tractable. By restricting the informa-tion participants have to only 2nd-order (the Target?s reputation and Actor?saction, but not the Actor?s reputation), I reduce the complexity and lengthof scenarios they read, reduce their memory load and chances for confusion,and greatly rein in the combinatorial explosion of possible interpretations.Thus, I assess only 2nd-order5 reputational systems, the ?sensible six?4So, for instance, by performing the same action twice (e.g., stealing your sandwich),I would be considered a horrible person and then be considered even better than I wasbefore I?d done anything.5Note that these are 2nd-order assessment-rules under my continuous translation (butnot necessarily under a dichotomous interpretation). That is, an Actor?s previous reputa-tion does not matter when assessing how their Action changes their reputation.107Dichotomous Theory Continuous Interpretation(G? G,B? G) Improvement (?)(G? G,B? B) No Change (?)(G? B,B? B) Worsening (?)(G? B,B? G) N/ATable 4.3: A map from dichotomous reputations reputation changes(e.g., from good (G) to (?) bad (B)) to their continuous inter-pretations (e.g., continuous representations of the reputation willexperience improvement).of the ?leading eight?. This restriction makes this project simpler, moretractable and likely more accurate, at the cost of neglecting the least plausi-ble theoretical possibilities. My results could suggest that people are in factusing more complex strategies however, as you will see, they do not.Continuous challenges: Boundaries and magnitudesMoving to continuous interpretations of reputations entails confronting thelimitations of measuring continuous quantities. I considered three alterna-tives. The first was magnitude-free measures, such as rank ordering eachcharacter on each question (liking, goodness, etc.). This has the advantageof offering unambiguous information about participants? relative rating ofeach character, but offers no information about the magnitude of differencesbetween them. For instance, a participant who judged a Defector on a badTarget as far better a Defector on a good Target but both worse than aNeutral or Cooperator, would be indistinguishable from a participant whodisliked both defectors equally.A second alternative is an unbounded continuous scale. For instance,participants could be asked to assign a number from negative to positiveinfinity describing their rating of each subject. Though this transparentlycaptures rank-ordering and some index of magnitude, magnitude differencescan be dramatically different between subjects and without extensive cali-bration for each subject, they are very difficult to compare.The alternative I chose was a bounded continuous scale. Participants an-108swered each question by moving a slider between two boundary labels. Whilethis makes it easy to interpret magnitudes and compare participants, it in-troduces rank-ordering ambiguity at the boundaries. For instance, imaginean individual who sets the sliders for both Cooperator and Neutral max-imally to ?Return? on the Money measure. Do they consider these twoindividual?s reputations as equal, or one as higher but both high enough toscore maximally on this particular question? Consequently my analysis mayover-estimate the frequency of no-change (?) reputational systems.ParticipantsSurveys were administered online, and participants were recruited via Ama-zon?s Mechanical Turk. This online sample that has been shown to givereasonably reliable answers to simple questions (Rand, 2012).To detect possible cultural differences, I recruited participants from thetwo best-represented populations on Mechanical Turk: North Americans(35.4% male; mean (sd) age: 34.5(13.4)) and Indians (53.5% male; mean (sd)age: 31.1(10)). Both populations were presented with identical materialsin English. 144 participants were recruited from each population. Theywere assigned to each condition in semi-random order: each condition wasassigned once in random order, then each a second time, and so on.Some participants were then excluded on the basis of three criteria. First,I logged all participants? IP-addresses and any repeat addresses had all theirinstances excluded. Second, the bulk of participants took between eight andeighteen minutes to complete the study, however a small number managedit in under five minutes. I excluded these individuals as very likely to havebeen inattentive, since my pilot-subjects were not able to achieve the samewhen instructed to complete the survey as fast as they could. Third, par-ticipants were asked to type two specific, ordinally specified letters from theEnglish alphabet; anyone failing this question was excluded. Excluded par-ticipants (41 Indians and 24 Americans) were replaced from their respectivepopulations in a random order.Participants were asked for the following demographic information: age109in years, sex, languages spoken, education level, self-reported ethnicity, andreligious affiliation.This study was initially designed purely for the American sample, andtheir data was gathered first. When it became clear that their judgementsshowed interesting patterns (detailed below), I extend the study to Indianparticipants. This was largely a convenience sample, since a large popu-lation of Indian participants are available on Amazon?s Mechanical Turk,the recruitment platform on which the American version had already beendeployed. This is not an ideal cross-cultural comparison, since characternames were all in English and the situations described were invented byEnglish-speaking North American resident graduate student (myself). How-ever, it does provide good preliminary insight into whether, approximately,reputational judgements are similar or strikingly different across cultures.AnalysesOur design allows for two different kinds of analyses.Continuous AnalysesFirst, I modeled participant?s judgement as beta-distributed (the naturaldistribution for bounded continuous measures), and constructed linear re-gression models to estimate the relationship of other variables to the mean(?, logit-linked) and precision (? , log-linked) of this distribution (Cribari-Neto & Zeileis, 2010; Ferrari & Cribari-Neto, 2004; Smithson & Verkuilen,2006).I was particularly interested in the effect of four main predictors andtheir interactions:Target-Good A dichotomous indicator of whether the Target was cueda ?good? (vis-?-vis bad) in the scenario in which the judgement wasmade.Actor-Cooperator Whether the individual being rated was the Coopera-tor (vis-?-vis defector).110Sample-Indian Whether the participant who made the rating was re-cruited from the Indian (vis-?-vis U.S.A.) sample.Valence-Negative Whether the participant had read about a negatively(vis-?-vis positively) valenced dilemma.I also controlled for a set of covariates in every model. Specifically:participants? age, sex and how long it took them to complete the study,whether they saw the good Target at work or among friends, which scenariothey saw first, which setting the particular data-point being regressed wasjudged in, the source of information (direct or gossip) and the ultimateconsequences for the target (good, bad or not reported).To calibrate each individual?s judgements in each scenario to their ownbaseline, I also included their rating of the Neutral character on the samemeasure as a covariate in ?-models.To maximise statistical power, I analysed each judgement (not some ag-gregate of them) and modelled the within-participant covariance in answersby computing clustered robust confidence intervals.Discrete AnalysesThe continuous analysis attempts to infer population parameters (e.g., thoseof the beta-distribution that best describes participants? responses). How-ever this experimental design also allows me to, for each individual, to pickout precisely which of the 81 possible continuous, 2nd-order reputationalsystems their choices instantiated. We can ask whether each individual, forinstance, considered a someone who defects who on a well reputed target asworse then someone who defects on a poorly reputed target.To calibrate each individual?s judgement to their own within-scenariobaseline, I first subtract each judgement from their rating of the Neutralcharacter. This makes inferences robust to participants who were inclinedto rate all characters better in one scenario (i.e., if they always favouredfriends over workmates).To discretely classifying a judgement difference as ?improving? (?), ?wors-ening? (?) or ?unchanged? (?) one must specify a minimum threshold for a111difference. Just how close must two judgements be using a 100-point mouse-based slider, before we assume the participant is trying to convey an identicaljudgment and attribute the difference to an unsteady hand, poor vision orlaziness? While my research assistants could readily set two sliders exactlyequivalent when asked to, I wanted to give participants a greater marginof error. Here I report results using a 2-point threshold. The supplemen-tal materials also present results for 0-,2- and 10-point thresholds (they areclearly labelled in the directory: /Chapter 4/discrete_strategy_tables/).4.2 ResultsThese data are rich and amenable to many interesting analyses. Here Ireport those I found most pertinent to answering the questions outlinedabove. Table 4.5 summarises the key likelihood-maximising parameter esti-mates that underlie my conclusions. Full quantitative details are availablein the supplemental materials (in the file /Chapter 4/regression_tables.pdf ).Did dependent measures capture reputational judgements?Table 4.4 presents the correlations between my four dependent measures.I designed Linking and Good to measure the same underlying cognitiverepresentation of reputation, and indeed they were highly correlated. Theyalso correlated highly with judgments about doing the characters a Favour,but all three measures were more distinct from judgments about whetherto return some lost Money to a character. This final measure was excep-tional in that almost 45% of the 2304 judgments I recorded (one for each offour characters, in two scenarios, for two samples of 144 participants) wereset to at their maximum boundary (participants would certainly return themoney), as opposed to merely approximately 15% on other measures. Isuspect this reflects compliance to social norms about how people shouldbehave in these situations. Though this restricted variability dampens thesensitivity of this measure, it makes the 55% of deviations particularly in-teresting. If taking the money involves violating a normative proscription, itis even more plausibly an index of participant?s behavioural dispositions in112Liking Good FavourGood 0.86Favour 0.82 0.8Money 0.38 0.4 0.42Table 4.4: Correlations between our four measures of participants?reputational judgmentsnegative dilemmas (where a deliberate action is required to extract a benefitat others? expense).I am confident that these measures index participants? representation ofcharacters? reputation.113114Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive DilemmasLiking Good Favour Money?? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas?? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas?? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas? ? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????115Figure 4.1 (preceding page): American Participants? absolute judgements of Targets and relative judge-ments of Actors. Judgements are divided by dilemma Valence (rows) and depen-dent measure (columns). Horizontal lines represent participants? mean absolutejudgements (left axis) and their 95% confidence intervals (obliquely shaded zones,estimate from beta-models) of Neutrals (red lines), good Targets (green, leftmostlines) and bad Targets (black, rightmost lines). Symbols represent the mean (pin-pointed by a red dot) of the difference (right axis) in each individual participant?sjudgements of Cooperators (angelic faces, ?) and Defectors (crossed swords, ?)relative to their own Neutral judgement in the same scenario. Since symbols rep-resent the difference between two beta-distributed variables, their 95% confidenceintervals (transparently shaded zones) are asymptotically normal approximations.116Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive DilemmasLiking Good Favour Money? ? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas? ? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas? ? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas? ? ? ?Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas????Coop. Defect Coop. Defect0. Target Bad TargetNegative DilemmasPositive Dilemmas? ? ? ?Figure 4.2: Indian Participants? absolute judgements of Targets and relative judgements of Actors. Thesame visual representation as Figure 4.1, for Indian participants.117Sample (?) Measure (?) U.S.A. IndiaModel (?) Valence (?) + - + -Manipulation(TargetG)Liking 4.14(0.29)** 3.57(0.30)** 1.71(0.26)** 1.84(0.22)**Good 3.49(0.29)** 3.44(0.29)** 1.50(0.22)** 1.55(0.22)**Favour 3.72(0.28)** 3.55(0.35)** 1.46(0.21)** 1.76(0.21)**Money 1.85(0.45)** 3.24(0.43)** 0.62(0.19)** 0.70(0.20)**1st Order(ActorC)Liking 0.02(0.19) 2.45(0.22)** 0.36(0.20)^ 1.25(0.17)**Good 0.23(0.17) 2.51(0.19)** 0.19(0.15) 1.00(0.17)**Favour 0.26(0.21) 2.30(0.19)** 0.30(0.15)* 1.15(0.16)**Money ?0.24(0.23) 1.38(0.19)** 0.21(0.15) 0.66(0.13)**2nd Order Cooperators(TargetG)Liking 0.78(0.27)** ?0.10(0.18) 0.16(0.19) ?0.04(0.18)Good 0.43(0.22)^ ?0.10(0.13) 0.37(0.19)^ ?0.08(0.16)Favour 0.52(0.26)* ?0.12(0.15) 0.19(0.19) ?0.10(0.15)Money ?0.23(0.32) 0.10(0.12) ?0.02(0.17) 0.03(0.17)2nd Order Defectors(TargetG)Liking ?0.33(0.14)* ?0.62(0.18)** ?0.10(0.20) ?0.48(0.17)**Good ?0.27(0.12)* ?0.47(0.16)** ?0.22(0.15) ?0.25(0.13)^Favour ?0.51(0.18)** ?0.41(0.18)* ?0.21(0.16) 0.01(0.13)Money 0.09(0.26) ?0.42(0.16)** ?0.05(0.15) 0.05(0.15)2nd Order Difference(ActorC * TargetG)Liking 0.77(0.42)^ 0.63(0.33)^ 0.25(0.25) 0.51(0.20)*Good 0.67(0.31)* 0.45(0.25)^ 0.56(0.26)* ?0.11(0.18)Favour 1.08(0.41)** 0.37(0.24) 0.40(0.23)^ ?0.02(0.18)Money ?0.12(0.42) 0.56(0.17)** 0.18(0.20) ?0.03(0.18)118Table 4.5 (preceding page): ?? : p< .01,? : p< .05, ^: p< .1Summary of key results for Chapter 4.Coefficients and (standard errors) describing the relationship likelihood-maximisingrelationship between participant?s beta-distributed judgments and ?whether the ac-tor being judged is a Cooperator (vis-?-vis Defector)? (ActorC), ?whether the Targetof their action has a Good (vis-?-vis Bad) reputation?(TargetG), and their interac-tion,controlling for all manipulated variables (see methods), and participants? age,sex and how long they took to finish. Full models with all covariate parameterestimates and precision estimates are available in the supplemental materials.Divided by sample (U.S.A. or India, columns) and dilemma-valence (positive [+],or negative [?]) and four measures of participant?s reputational judgments (innerrows). Outer rows describe five analyses, showing the relationships between: Ma-nipulation, Target judgements and our manipulation of the target?s reputation;1st Order, Actor judgements and whether the actor cooperated or defected; 2ndOrder Cooperator/Defector, judgements of each type of Actor, and whethertheir Target was cued as Good; 2nd Order Difference, the magnitude of the dif-ference between judgments of Cooperators and Defectors and whether the Targetwas cued as Good.Notice that (1) our manipulation appears effective; (2) 1st-order reciprocity is onlyevident in negative dilemmas; (3) we only see evidence for 2nd-order reciprocityin our US sample, though our Indian sample trends in the same direction; and(4) among our US sample 2nd-order differences are most evident in ?behaviouralresponses in kind?: the Money measure responds to negative dilemmas while theFavour measure responds to positive dilemmas.Was the Target?s reputation manipulated successfully?Across all measures, in all samples and valences, judgments of Targets werejudged worse in ?Bad scenarios? (one?s where the manipulation attemptedto worsen the Target?s reputation) than in Good scenarios. This is evidentwhen regressing Target judgments on the reputation manipulation in eachscenario (Target-Good), controlling for all covariates (see SupplementalTables for a full exposition, and the first row of Table 4.5 for a quantitativesummary). I also saw an interesting cultural interaction: the differencebetween Target judgements in Good and Bad scenarios was smaller amongmy Indian (mean judgements: .82 and .53) than among my U.S.A. sample(.94 and .39). While Target judgements still significantly covaried with themanipulation in India, it seems to have been less effective there.Figures 4.1 and 4.2 make these patterns visually apparent by showingmean Target ratings in Good (green horizontal bar) and Bad scenarios (blackhorizontal bar) in both samples.Did participants dislike defection, irrespective of Targetreputation (i.e., first-order IR)?In other words, were Cooperators rated better than Defectors? Yes, butmuch more so in negatively-valenced than positively-valenced dilemmas.I regressed participants? judgements on whether the Actor they werejudging was a Cooperator or Defector (Actor-Cooperator), interactedwith participants? culture (Sample-Indian) and dilemma valence (Valence-Negative), controlling for all manipulated variables (see methods), andparticipants? age, sex and how long they took to finish. These analyses re-vealed a three-way interaction between these predictors, across all dependentmeasures. Consistently, Cooperators were judged better than Defectors (in-dependent of Target reputation) in negative dilemmas, while effects in pos-itive dilemmas merely trended non-significantly in the same direction (seeTable 4.5, 1st Order row). This negativity bias was also consistently weaker(though still statistically significant) among my Indian sample. These re-sults are depicted in figures 4.1 and 4.2 by the relative heights of Cooperator119symbols (?) to Defector symbols (?).Were participants? judgments contingent on the Target?sreputation (i.e., second-order IR)?The strictest translation of dichotomous theories into continuous realityyields two predictions. Under Scoring, the magnitude of reputation changeshould be independent of the Target?s reputation. Under Standing, defectingon bad Targets should not reduce reputations at all, while defecting on goodtargets should. My data are inconsistent with both these strict predictions.The ?2nd order? rows in Table 4.5, and the relative heights of green-shaded (good Target actions) and black-shaded (bad Target actions) symbolsin Figures 4.1 and 4.2 show that, in contrast to the strict Scoring prediction,participants? judgments did sometimes depend on Targets? reputations.The distance between the black-shaded region around defection symbols(?), representing a 95% confidence interval around the mean judgment ofdefections against bad Targets) and the red line and its shaded 95% confi-dence interval (representing judgments of the Neutral character) shows that,in contrast to the strict Standing prediction, participants did condemn thosewho defected on Bad targets.A more relaxed interpretation of Standing is that reputation change willbe moderated continuously by a Target?s reputation. That is, defectingon bad Targets may still entail reputation loss, but less than defecting onless good Targets. My data supports this relaxed interpretation among myAmerican sample. Among my Indian sample this was only a trend.There are two ways to assess 2nd-order IR. First, we can ask how judge-ments of Defectors were related to the reputation of their target (Target-Good, Table 4.5, 2nd Order Defectors row). Across all measures, judgmentswere influenced by an interaction between valence and culture. AmongAmericans, across all measures and both valences (except Money in posi-tive dilemmas) those who defected against Good Targets were judged worsethan those who defected against Bad Targets. Indian participants tendedto trend in the same direction, but only achieved statistical significance onthe Liking measure in negative dilemmas.120However when we ask the same question about Cooperators (Table 4.5,2nd Order Cooperators row), we see no evidence that those who Cooper-ated with Good targets were judged better than those who cooperated withBad targets in negative dilemmas. In positive dilemmas, only among Amer-icans were those who Cooperated with good Targets preferred to those whocooperated with bad Targets, and even then only on the cognitive (Goodand Liking) and positive behavioural (Favour) measures, not for negativebehaviour (Money).We can ask another 2nd-order question: ?is the magnitude to which co-operators were preferred to defectors (i.e., the 1st-order effect) greater whenthe Target is Good?? To answer this, I examined the interaction between tar-get reputation (Target-Good) and Actor identity (Actor-Cooperator;Table 4.5, 2nd Order Difference row). Again my data reveal robust inter-actions between valence and sample across all measures. Americans showedsubstantial effects on the positive behavioural measure (Favour) when judg-ing actors in positive dilemmas, and reciprocally, a strong response on thenegative behavioural measure (Money) when judging actors in negativedilemmas. In both cases, the difference between Cooperator and Defec-tor ratings was greater when the Target was good. My cognitive measures(Good and Liking), trended in the same directions as these effects, ap-proaching statistical significance independently and achieving it if pooled.Among my Indian sample, positive dilemmas trended similarly to Amer-icans? (achieving significance only for Good ratings), while negative dilem-mas showed no evidence of a relationship, except for Liking, which regis-tered a significant effect of similar magnitude to Americans?.To summarise, Americans showed the clearest evidence of 2nd-order in-direct reciprocity, especially when responding behaviourally ?in kind?. Thatis, participants? 2nd-order responses were strongest when judging whetherthey would do a favour for someone they?d seen interacting in a positivedilemma (taking action to provide costly help, or refusing to do so), orwhen judging whether to keep someone?s misplaced money who had just in-teracted in a negative dilemma (taking action to gainfully exploit someone,or refusing to). As in other cases, my Indian sample tended to show ame-121liorated trends in the same direction, though effects here were particularlyweak and inconclusive.Covariates and precisionThe supplementary regression tables specify detailed estimates of the re-lationships of the covariates to the models reported above. To gain moresuccinct insight into their overall impact, I regressed participant?s judge-ments on to these covariates across all measures.The order in which scenarios were presented did not substantially af-fect participants? judgements. There was no evidence of rating differencesbetween the two kinds of scenarios (friends or work), though Indian partic-ipants? tended to give higher ratings overall if they read about the ?goodTarget? in the context of work scenario.Older participants tended to give higher ratings in all circumstance, es-pecially in positive dilemmas. Males, on the other hand, tended to givelower ratings in positive dilemmas.Those participants who took the most time to answers also gave morevariable answers, and among American participants, tended to give higherratings. Participants gave less variable ratings (and these ratings tended tobe higher, but not statistically significantly) when the ultimate consequencesfor the target were mentioned, regardless of whether those consequences weregood or bad. The fact that both ?good? and ?bad consequences? caused ashift in the same direction relative to ?no consequences? suggests that themere reminder that the Actors? actions had tangible consequences for theTarget influenced their judgements.There was no evidence that the ostensive source of the information (di-rectly witnessed or heard from a reliable source) had any appreciable influ-ence on judgements. Though, of course, the source in all cases was my onlinestudy and participants were merely being asked to imagine the scenarios, sothis lack of effect should not be over-interpreted.122Discrete AnalysisTo discretely test for successful manipulation of the Target?s reputation, Isubtracted each participant?s Target judgement from their judgement of theNeutral character in the same scenario. I then categorised each participantaccording to whether, relative to Neutrals, they judged Targets who?d beencued as good better than those cued as bad (G > B), the reverse (G < B) orapproximately equally (G?B) at various thresholds. Here I report results fora 2-point threshold (i.e., judgements within 2-points of each other are consid-ered equal), other breakdowns are presented in the supplemental materials.I used a ?2 goodness of fit test to assess whether the number of partici-pants in each category differed from expectation when randomly samplingfrom truly indiscriminate participants. I also used a simple non-parametricbinomial test to assess whether the proportion of individuals favouring theGood target was higher than those favouring the Bad target. Participants?judgements unequivocally responded to the manipulation. These results arepresented in Table 4.6.I performed parallel analyses for Cooperators (Table 4.7) and Defectors(Table 4.8). Unsurprisingly this reveals similar patterns to those gleanedfrom my continuous analyses. Those who cooperate with Good Targets aremore often rated as better than those who cooperate with Bad Targets, es-pecially in positive dilemmas. Those who defect on Bad Targets are ratedbetter than those who defect on Good Targets, especially in negative dilem-mas. These patterns are clearest among my American sample while Indianresponses are noisier.123124Negative Dilemma Positive DilemmaU.S.A.Liking Good Favour MoneyG> B 67 67 67 29G? B 4 2 4 41G< B 1 3 1 2Pr(?2)? 0 0 0 0Pr(Binom)? 0 0 0 0Liking Good Favour MoneyG> B 68 70 68 31G? B 2 2 3 39G< B 2 0 1 2Pr(?2)? 0 0 0 0Pr(Binom)? 0 0 0 0IndiaLiking Good Favour MoneyG> B 51 54 53 34G? B 5 4 1 23G< B 16 14 18 15Pr(?2)? 0 0 0 0.02Pr(Binom)? 0 0 0 0.01Liking Good Favour MoneyG> B 51 53 53 32G? B 6 6 5 25G< B 15 13 14 15Pr(?2)? 0 0 0 0.05Pr(Binom)? 0 0 0 0.02Table 4.6: Frequencies of discretely categorised judgements of Targets. The raw number of individuals whorated Targets in Target-Good scenarios, relative to those in Target-Bad scenarios as better (G > B),worse (G < B) or approximately equally, at threshold 2, on each of our four measures, broken downby dilemma valence and sample. For each measure we include the p-value from a ?2 test of whetherthe proportions of individuals in each of the three categories is equal, and a binomial test of whetherthe proportion who judged Targets better when the Target was good differed from those judging thembetted when the Target was bad.125Negative Dilemma Positive DilemmaU.S.A.Liking Good Favour MoneyG> B 19 19 23 12G? B 21 22 23 49G< B 32 31 26 11Pr(?2)? 0.13 0.2 0.88 0Pr(Binom)? 0.09 0.12 0.78 1Liking Good Favour MoneyG> B 35 35 38 16G? B 12 14 18 48G< B 25 23 16 8Pr(?2)? 0 0.01 0 0Pr(Binom)? 0.25 0.15 0 0.15IndiaLiking Good Favour MoneyG> B 28 28 31 24G? B 12 11 8 25G< B 32 33 33 23Pr(?2)? 0.01 0 0 0.96Pr(Binom)? 0.7 0.61 0.9 1Liking Good Favour MoneyG> B 25 36 33 21G? B 17 14 13 28G< B 30 22 26 23Pr(?2)? 0.17 0.01 0.01 0.58Pr(Binom)? 0.59 0.09 0.43 0.88Table 4.7: Frequencies of discretely categorised judgements of Cooperators. The raw number of individualswho rated Cooperators in Target-Good scenarios, relative to those in Target-Bad scenarios as better(G> B), worse (G< B) or approximately equally, at threshold 2, on each of our four measures, brokendown by dilemma valence and sample. For each measure we include the p-value from a ?2 test ofwhether the proportions of individuals in each of the three categories is equal, and a binomial test ofwhether the proportion who judged Cooperators better when the Target was good differed from thosejudging them betted when the Target was bad.126Negative Dilemma Positive DilemmaU.S.A.Liking Good Favour MoneyG> B 26 24 23 14G? B 9 8 13 43G< B 37 40 36 15Pr(?2)? 0 0 0 0Pr(Binom)? 0.21 0.06 0.12 1Liking Good Favour MoneyG> B 18 18 21 6G? B 19 22 23 52G< B 35 32 28 14Pr(?2)? 0.02 0.11 0.58 0Pr(Binom)? 0.03 0.06 0.39 0.12IndiaLiking Good Favour MoneyG> B 19 28 34 21G? B 13 10 10 25G< B 40 34 28 26Pr(?2)? 0 0 0 0.75Pr(Binom)? 0.01 0.53 0.53 0.56Liking Good Favour MoneyG> B 25 24 26 17G? B 11 25 19 31G< B 36 23 27 24Pr(?2)? 0 0.96 0.45 0.13Pr(Binom)? 0.2 1 1 0.35Table 4.8: Frequencies of discretely categorised judgements of Defectors. The raw number of individuals whorated Defectors in Target-Good scenarios, relative to those in Target-Bad scenarios as better (G> B),worse (G< B) or approximately equally, at threshold 2, on each of our four measures, broken down bydilemma valence and sample. For each measure we include the p-value from a ?2 test of whether theproportions of individuals in each of the three categories is equal, and a binomial test of whether theproportion who judged Defectors better when the Target was good differed from those judging thembetted when the Target was bad.It bears mention that since these tests are non-parametric, since thesecategories are necessarily approximate and since they were constructed bydiscarding all information about judgment magnitude, these discrete an-layses are dramatically underpowered relative to the continuous analysesreported above. Though the continuous analyses are superior for assessingtheoretically central questions about the environment a potential defector(or cooperator) faces, discrete analyses open up the possibility of examiningthe distribution of individual IR strategies among those judging behaviour,and provide a cleaner match to existing theory.Our design allows us to classify each individual into exactly one of the81 (32?2) potential 2nd-order reciprocity reputational systems. That is, theycould have made three possible judgements (reputations improving, ?; wors-ening, ?; or not changing, ?) about two kinds of actions (Cooperate of De-fect) on two kinds of Targets (Good or Bad). The supplemental materialsinclude large tables that presents the proportion of individuals in each ofthese category, overall and split by sample and valence.For simpler insights, these proportions can be averaged along their mar-gins. The supplemental materials include tables that show how cooperatorsalone (and defectors alone) are judged, depending on their Target?s reputa-tion and how those who interact with good Targets are judged, dependingon whether they cooperated or defected. Here I present the analysis mostpertinent to distinguishing Standing from Scoring, and for distinguishingamong the ?sensible six? strictly 2nd-order leading eight Standing strategies:how actions on a Bad Target are judged. Table 4.9 shows the proportion ofpeople (at a 2-point threshold) who judged Cooperating and Defecting onbad Targets each of the three possible ways (rows and columns), on each ofmy four measures (four filled circles and proportions in each cell).First, as discussed above, the Money measure was unique in that judg-ing everyone equally was by far the dominant response. Second there isa great deal of heterogeneity in people?s reported reputational reactions.Though participants favoured some reputational systems on average, nosingle strategy was instantiated by the majority and almost the full gambitof possible responses received some representation. Third, the plurality of127participants (around 15%) instantiated a strict 1st-order Scoring strategy,while only around 10% were spread among the strictly defined Standingstrategies (see full breakdown in supplemental figures). The marginal dis-tributions of judgments about interactions with Bad targets (Table 4.9)are particularly informative. While in positive dilemmas Scoring is aboutequally represented as the two cooperator-favouring Standing strategies (Ta-ble 4.9, top row), in negative dilemmas it is clearly the dominant reputa-tional system. However, this seems to be an epiphenomenon of an evenstronger signal. In positive dilemmas (where cooperating involves delib-erate action), most participants judge Cooperators as better than Neutralindividuals and vary substantially in their opinion of Defectors, while theopposite is true in Negative dilemmas. Indeed, in negative dilemmas thoughthe majority of participants rated Defectors as worse reputed that Neutrals,a substantial fraction of individuals rated Cooperators as equal to or worsethan Neutrals?assessment-rules that theory has dismissed as evolutionarilyunsustainable.Overall, besides providing the first quantitative estimates of the distri-bution of reputational systems in modern populations, my discrete analysishighlights the considerable heterogeneity in reputational judgements, andthe substantial impact that dilemma valence has on individuals? judgements.It also underscores a key pattern from my continuous analysis: people moststrongly and consistently respond to defections in negative dilemmas.128129Positive Dilemmas Negative DilemmasDefectC oo pe ra te? ? ??.19%.11%.13%.10%.17%.22%.18%.6%.15%.12%.13%.9%?.7%.4%.1%.1%.14%.14%.19%.56%.3%.8%.7%.1%?.12%.15%.10%.6%.7%.6%.8%.4%.7%.9%.9%.6%DefectC oo pe ra te? ? ??.8%.4%.8%.4%.3%.4%.6%.3%.44%.48%.39%.22%?.2%.2%.3%.3%.8%.6%.6%.44%.17%.16%.18%.10%?.4%.3%.3%.3%.1%.3%.3% ?.13%.15%.15%.11%130Table 4.9 (preceding page): A discrete partition of participants? reputational judgements, aggregated byhow Cooperators and Defectors are judged when they act upon Bad Targets. Judge-ments of individuals who either cooperated with (rows) or defected on (columns)badly reputed targets are sorted into those who considered them better (?) thanNeutral individuals (who were in an identical situation, but did not have an op-portunity to act), worse (?) or approximately equal (?, within 2% of the width ofthe mouse-based slider participants used). The proportion of individuals in eachcategory is listed and shown as the proportion of the circle?s area filled, for ourfour measures: whether you like the actor (top left circle in each cell), deem them?good? (top right), would return their misplaced money (bottom left) or would dothem a slightly inconvenient favour (bottom right). Theoretically predicted strate-gies are coloured: members of the ?leading eight? 2nd-order Standing strategies areshaded orange, while 1st-order Image Scoring is in green. Participants are dividedby outer columns into those who read about positive and negative dilemmas.4.3 DiscussionTo the best of my knowledge, this is the first attempt to systematicallymap people?s reputational intuitions onto the space of possibilities outlinedby Indirect Reciprocity (IR) theory by simply asking them. It has severallimitations. First and foremost, I did not directly witness either the trans-mission of reputations nor people?s reputation-based reactions to others realbehaviour. Instead, I merely asked people to report how they would respondto a variety of theoretically relevant situations. In so far as people are ableto simulate hypothetical situations and have accurate introspective access totheir reactions in those situations, this method should yield accurate insightsand lay the way for more complex behavioural studies.However, I do not believe we should put too much faith in introspection,nor do I advocate for the exclusive use of survey methods. Rather, thissurvey is a cheap and informative first pass at measuring people?s patternsof reputational intuitions where cognitive intuitions of specific individualshave been cued. Given the results here, we have good reason to attemptexperimental investigations which somehow preserve the actual anonymityof our participants, while still cuing the person-specific cognitive mechanismsthat underlie our reputational judgements.Second, our materials were deployed on over the internet. This allowedus to tap a larger, more diverse, cross-cultural sample and draw inferencesabout a broader population. However this came at the cost of sacrificingsome control over our study population. I do not know who my participantswere, except by trusting the information they reported (see Rand, 2012, fora justification of such trust). I also cannot be sure their responses were gen-uine and they were not just thoughtlessly ?clicking through? to earn smallamounts of money. Though I used several measures to counteract such be-haviour (IP tracking, timing, dummy questions), it is likely that inattentiveor apathetic participants added noise to our data. However such mindlessclicking should have varied randomly with respect to difference in our sce-narios and conditions, so our statistical techniques should still be able topick out the signal of genuinely-answering participants. If anything, the131noisiness of sample should give us more confidence in the patterns that wereclearly detected. Furthermore, many participants? qualitative accounts oftheir decisions show evidence of careful consideration of the scenarios andgenuine judgement of the characters.Third, to keep our design tractable and because for the reasons outlinedabove they are less plausible, I omitted the possibility of assessing 3rd-order reputational judgements (i.e., I cued Target?s previous reputations,but provided no information on Actors? previous reputations).Finally, this first foray used scenarios set in a workplace and amongfriends. It is quite plausible that the distribution of IR systems variesacross domains?people may have different rules for business and family,for instance?and I am so far only able to assess a small component of thisvariability.Despite these limitations, our investigation has yielded several usefulinsights.Diversity of strategiesTheoretical models of IR typically describe systems of interacting individu-als that are attracted to monomorphic equilibria?that is, they predict thateveryone will eventually use the same assessment- and action-rules, or justassume that they always do. In stark contrast, participants in our surveyinstantiated a large variety of reputational systems. Even on our Moneymeasure, which seemed strongly biased towards judging everyone equally,no single assessment-rules ever described the majority of participants? judge-ments.Though some of this heterogeneity is attributable to the noisiness ofour sample and of people?s simulated judgements about imagined scenarios,there seems to be more going on. When one system did hold the plurality, ittended to be first-order Image Scoring, followed by members of the ?leadingeight? Standing strategies. Rather than dominance by one clear equilibriumresident, the landscape of contemporary IR seems to be complex ecology ofmany different interacting strategies, assessment- and action-rules. There132is also some evidence that strategies vary between cultures (see the fullcatalogue of per-culture discrete strategy breakdowns in the supplementalmaterials, in the directory: /Chapter 4/discrete_strategy_tables/), but asdiscussed above, my convenience-based Indian sample does not allow me toconfidently draw this conclusion.It is also possible that, with the advent of more sophisticated institutionslike formal laws, police-forces, written contracts and so on, IR no longer playsa major role in contemporary societies. That is, theoretical model of IR maybe correct, but these systems may currently be far out of equilibrium becauseextrinsic institutions are exerting far stronger pressures on behaviour. Ifthis is correct, then the proper domain of IR is small-scale historical andcontemporary societies. A testable prediction follows from this logic: thedistribution of IR judgements should be far tighter in small scale societieswith fewer formal cooperation-sustaining institutions.Dilemma valence mattersIR theories have almost exclusively considered the dilemma of helping some-one or doing nothing (cf. Chapter three). Many theorists assume that thesesame models also represent negative dilemmas, after all on paper the relativepayoffs are the same and one payoff matrix can be transformed into the otherby merely subtracting a constant. However in the real world these dilem-mas are very different, as are people?s reputational responses to them. Theyvary in whether cooperating (or defecting) involves deliberately, detectablytaking an action or merely ignoring an opportunity. Actually acting makesit plain to any observers that an opportunity to act existed (whereas it mayotherwise slip by unnoticed), that the actor was aware of the opportunity(which could otherwise be denied) and what choice the actor made.Our participants? strongest reputational responses were always to ?active?choices. That is, to cooperation in positive dilemmas and defection in neg-ative dilemmas. However of these, defection in negative dilemmas triggeredby far the strongest reputational consequences, across both samples and set-tings, of any variable in our design. This was the strongest, most consistent133signal in our data. These patterns are consistent with the negativity-biases(Baumeister et al., 2001; Cacioppo & Berntson, 1994; Rozin & Royzman,2001) and omission-biases (Baron & Ritov, 2004; Cushman et al., 2006; De-Scioli et al., 2011; Ritov & Baron, 1999; Spranca et al., 1991) that have beenregularly observed in other domains of human behaviour. As far as I know,this study constitutes the first direct evidence that such biases also pervadesecond-order indirect reciprocity.Furthermore among our American sample, who were more responsive toour manipulation in general, behavioural responses tended to match the va-lence of Actors? actions. That is, when actors deliberately, actively defected(i.e., in negative dilemmas) participant?s responded by not returning somemoney they had misplaced (an active deviation from the social norm evi-dent in our data), while if they active cooperated (i.e., in positive dilemmas)participant?s were more willing to go our of their way to do them a favour(a positive dilemma for participants).Overall, our data suggest the IR theory should be more attentive todilemma valence, and focus on defection in negative dilemmas in particular.This supports the key assumptions that set Negative Indirect Reciprocity(NIR, Chapter 3) apart from other IR models.Genes, culture or individual learning?IR theory is often used to explain how humans first started cooperating, pro-viding an explanatory foundation for subsequent evolution of human brainsand institutions. Under such accounts, if IR played an important enoughrole in human evolution, it may have left its signature on our IR-relevant in-tuitions and judgements. Specifically, we should see consistent reputationalintuitions among people everywhere.Our data only registered such a signal in one case: reputations fall whenpeople actively defect in negative dilemmas. This happened consistently andstrongly across both our samples. If humans share reliably developing, ge-netically encoded IR-relevant intuitions, our data suggest that they concerndefection in negative dilemmas.134An alternative is that IR evolves culturally. That is, each individual?sintuitions emerge developmentally but are shaped by the distribution ofothers? behavioural and reputational judgements in their society. Society-wide equilibria can emerge from these dynamic interactions, and persistover many generations. The interactions of societies at different equilibriacould even lead to a form of cultural group selection. If this were so, weshould see highly convergent judgements within groups and large differencesbetween them. However, we did not. We saw a great deal of variabilitywithin-cultures and Indians? responses were typically attenuated analoguesof Americans? responses. One exception is that while my American sampleshowed clear evidence of valenced reactions ?in kind?, my Indian sample didnot.It is worth noting that our ability to draw cross-cultural inferences islimited since our Indian participants read materials in English, describingthe interactions of English-named characters. While this first survey of themost conveniently accessible samples is useful, a robust answer to whetherIR-intuitions vary culturally will require surveying a more phylogeneticallydistinct sample of cultures (America and India both inherited culture fromthe British), especially smaller-scale societies with few formal institutions.Overall, the heterogeneous ecology of different reputational judgementsI observed is most consistent with a) people developing their reputationalintuitions by individual learning and b) doing so in a world were dynamicequilibria do not strongly attract them to any one system.Overall conclusionIR is potentially a very powerful tool for understanding both the originsand evolution of human behaviour and institutions. However the theoreticalpicture of IR is both complex and, on its own, inconclusive. Here I haveattempted to wed this theory with data on the reputational judgements ofindividuals from two contemporary populations. I believe a productive nextstep would be to ask similar questions of individuals in small scale societies.1354.4 SupplementalAdditional supplemental materials are provided as computer files.They include:? A list of all qualitative explanations that participants gave of theirchoices.? Discrete partitions of participants into the IR-strategy types, for var-ious threshold levels.? Full regression tables for all analyses.? The data file used to generate scenarios, from which any variant of thescenarios can (with a little cross-referencing) be reconstructed.136Chapter 5It is better to profiteer onthe guilty: is moralcondemnation sensitive toreputation?Moral psychologists continue to demonstrate that people?s moral intuitionscan differ dramatically from what classic moral theories predict (e.g., Gra-ham et al., 2009; Greene et al., 2009; Knobe, 2010). The last two decadeshave also seen rapid growth in the sophistication and scope of evolutionarymodels of human cooperation. These models explore how natural selectioncould produce a species with moral intuitions in the first place. For selec-tion to have favoured contemporary individuals with pro-social moral in-tuitions, ancestral socio-ecological dynamics must have somehow protectedcooperatively-disposed individuals from selfish ones. Formal evolutionarymodels specify how these mechanisms could work and imply predictionsabout how they would shape our psychology (recent examples include: Boydet al., 2010; Panchanathan & Boyd, 2004; Sigmund et al., 2010, and chapter3 of this dissertation).These models have moved beyond simple kinship- or reciprocity-based137cooperation (Axelrod & Hamilton, 1981; Boyd & Lorberbaum, 1987; Hamil-ton, 1964; Smith, 1964; Trivers, 1971), which have broad applicability acrossspecies but struggle to explain many features of human cooperation in par-ticular (Chudek et al., 2013b). Newer models probe the culture-gene coevo-lution of cooperative, cultural species like ours by emphasizing the role ofsocially and culturally transmitted reputations, norms and institutions (fora review, see Chudek & Henrich, 2011).These two scholarly traditions have seen unfortunately little explicitoverlap despite great potential for cross-fertilisation (Mesoudi, 2007; Mesoudiet al., 2006). There is a straightforward logical bridge. If humans evolvedtheir moral intuitions in the selective environments described by these evo-lutionary models, ceteris parabis, moral psychologists ought to observe con-temporary moral judgements and choices adapted to those environments.Three good things can come of testing evolutionary models of coopera-tion with the tools of experimental moral psychology. First, psychologicalevidence can help reconstruct the picture of our evolutionary history, ben-efiting scholars across disciplines. Second, moral psychologist can groundtheir ultimate-level explanations in formal evolutionary theory, avoiding thehighly speculative, untestable evolutionary story-telling that is too often in-voked to justify moral intuitions. Third, formal evolutionary theory candirect our attention to moral phenomena we might not have otherwise con-sidered.Here I test a prediction about moral reasoning that follows from recentmodels of Negative Indirect Reciprocity (NIR, Chapter 3). These mod-els describe a sequence of evolutionary developments that transformed earlycognitive adaptations for forming friendships and coalitions (behaviour com-mon to primates (Higham & Maestripieri, 2010; Langergraber et al., 2007;Perry & Manson, 2008; Seyfarth et al., 2012; Silk, 2002; Watts, 2002)) intoa decentralised mechanism for enforcing individual adherence to arbitrarycommunity-wide behavioural norms. To do so they assume our ancestors?moral evaluations followed characteristic patterns, which ought to persisttoday.To briefly summarise the logic of NIR, imagine that our ancestors regu-138larly encountered opportunities to exploit one another?to achieve some gainat one another?s expense?for instance by stealing food, dominating matingopportunities or taking advantage of another?s illness. Imagine also thatthey began socially-coordinating their individual opinions of one anotherinto community-wide reputations. This might have been a consequence ofearly cognitive adaptations for generalised cultural learning (Boyd & Rich-erson, 2005; Mesoudi et al., 2006; Richerson & Boyd, 2004). Imagine thatthese individuals were less likely to exploit someone they liked (i.e., peerswith a good reputation, their friends) than those they didn?t. Finally, imag-ine that individuals tended to dislike those who exploited their friends.Since exploitation opportunities are often invisible unless they are actedupon (i.e., you would not know that I could have stolen your food cache,unless I do), an individual?s reputation can suffer if they exploit someone,but if they cooperative by inaction, it is likely to go unnoticed. This createsselective pressure for individuals to notice and act upon opportunities toraise their peers? opinions of them (i.e., which, ultimately, causes others toexploit them less). NIR demonstrates how this pressure can eventually leadto a norm-psychology (Chudek & Henrich, 2011)?cognitive adaptations forbeing aware of, adhering to, and reacting negatively to the violation of socialnorms?which can act as the foundation for more sophisticated cooperation-sustaining mechanisms (e.g., Boyd et al., 2010; Panchanathan & Boyd, 2004;Sigmund et al., 2010).NIR models suggest that humans are particularly likely to have genetically-adapted, reputation-based moral intuitions in negative cooperative dilemmas(opportunities to exploit), where defections are easily observed and conse-quences escalate when someone is exploited repeatedly. Meanwhile, in thesocio-ecology described by NIR, positive cooperation (opportunities to help)are more likely to be sustained by norms and institutions, since they re-quire communities to coordinate on shared expectations about when andhow individuals should act. So, for example, NIR anticipates that acrosscultures and history, people?s reactions to someone stealing someone else?sresources ought to be consistent. Specifically, the thief?s reputation shouldfall, though in inverse proportion to the victim?s reputation. On the other139hand, NIR anticipates substantial, norm-driven cultural variability in peo-ple?s reputaitonal judgements of someone who donates resources to someoneelse.Some recent empirical work (Engelmann & Fischbacher, 2009; Milinskiet al., 2001, but cf. Bolton et al., 2005 and chapter 4) has suggested thatmodern humans assess reputations using first-order ?scoring? strategies (theyconsider any defection bad; Brandt & Sigmund, 2005; Nowak & Sigmund,1998), rather than the theoretically more plausible second-order ?standing?strategies (they do not care if someone defects on a ?bad guy?; Brandt et al.,2004; Engelmann & Fischbacher, 2009; Panchanathan et al., 2003). Howeverthis work has exclusively tested cooperation in positive (helping) dilemmas.If NIR correctly describes ancestral social dynamics, then modern humansought to be more disposed to second-order reputational intuitions in negativedilemmas. That is, they should dislike people who exploit others, but notmind as much when the victims are ?bad guys?.A recent empirical discovery by moral psychologists provides a ripe op-portunity to test this prediction. Inbar, Pizarro, & Cushman (2012) docu-mented what I will call ?the lucky profiteer effect?: people morally condemnthose who profit from others loss, even if it is clear they did not contributeto causing it. These authors were motivated to challenge the prevailing as-sumption that individuals are considered blameworthy only if they causedharm or intended to. They presented their participants with vignettes de-scribing an individual (lets call him the profiteer) who makes a profit (e.g.,on the stock market) from the misfortune of others (e.g., a devastatingearthquake striking a third world country). Their participants rated boththe moral blameworthiness of the profiteers action and the turpitude of hischaracter. Across three studies participants disapproved of both the prof-iteer and his actions, even though he clearly did not cause the earthquake(in a fourth study, they observed similar patterns when the profit or lossof a company were at stake instead). Inbar et al. showed that this effectpersists even when the profiteer merely positions themselves to benefit froma misfortune that does not ultimately occur, but disappears if the profiteerhad other reasons for his actions (e.g., insuring his own assets in the third140world country). They conclude that people morally condemn the profiteerbecause they infer his wicked desires.This may be just the pattern of moral condemnation assumed by NIR.NIR describes a socio-ecology where our ancestors condemned one another?sunjustified exploitation, however it is agnostic to how they identified suchexploitation. Causal relationships are notoriously hard to identify, even forscientists armed with statistics. It is plausible that our ancestors recognizedexploitation by merely noticing the coincidence of one individuals? gain withanother?s loss, without needing a clear causal link.The mere coincidence of one individuals? gain with others? loss is stillthe driver of social discord, reputational damage and violent retribution inmany societies (e.g., witchcraft accusations and the ?evil eye?; Adinkrah,2004; Elworty, 1895; Evans-Pritchard, 1976). While NIR does not predictthis disregard for causality, it does make clear predictions about the patternof judgements and reputations that ensue after causes have been disregarded.Specifically, it predicts that the lucky profiteer effect should be amelioratedwhen profiteer?s victim has a bad reputation.This chapter tests this prediction.5.1 Methods, study oneI recruited 105 students from around the University of British Columbiacampus, aged 18 to 25 years (mean= 19.6, SD= 1.7). To avoid confounds incomparing results to Inbar et al.s, I drew a similar sample by including onlyadults who had grown up in North America and spoke English as their mainlanguage. Participants were randomly sorted in to one of two conditions thatdiffered only in whether they read a scenario about criminals or refugees.The scenarios read (variants in square brackets):Floret is a tiny, previously unpopulated island in the pacificthat has recently become a haven for [refugees/criminals] fleeingfrom [persecution/justice] in their home countries. In fact, theisland is now entirely peopled by [refugees/criminals]. Andrew,a stock-market trader, invested in special ?disaster bonds? that141would only make a profit if a disaster hit Floret. A few monthslater a terrible hurricane hit Floret; many of the [refugees/criminals]were killed and the rest were left homeless and starving. Andrewmade a large profit on his bonds and is now quite wealthy.For brevity, I?ll refer to the two kinds of characters in these scenarios as?residents? of Floret, and the ?profiteer? who, by luck alone, makes a profitfrom their suffering.Each participant received a small candy bar for reading their scenarioand rating their agreement with the following three statements (variablenames in parenthesis).(Good) I think Andrew is probably a good person.(Approve) I disapprove of Andrew?s investment.(Friends) I could probably be friends with Andrew.I will refer to these questions by the variable names above. Each questionwas followed by a 10cm horizontal line with vertical dashes every 2.5cm,labelled ?Strongly Disagree? at the extreme left, ?Strongly Agree? at theextreme right and ?Neither? at the centre. Using a ruler, coders convertedparticipants? marks on this line to a 100 point metric. This was scaled tothe open unit interval using the method described in (Smithson & Verkuilen,2006).The second question (Approve) was reverse coded so that higher valuesalways imply a more positive evaluation of Andrew and his actions.Since participants? answered on a bounded scale, I considered modelthese answers as a beta-distributed random variable (Cribari-Neto & Zeileis,2010; Ferrari & Cribari-Neto, 2004; Smithson & Verkuilen, 2006). Thisprovided better fits, but qualitatively identical findings, as the standardtechnique of assuming an underlying normal distribution1.1That is, beta-models always had lower deviance than normal-models?they fit the databetter. Deviance is the only relevant part of AIC, BIC and many other metrics of modelcomparison when both models are fit by estimating the same number of parameters. Thenormal and beta distributions both have two parameters. However the beta is supportedon the bounded unit interval, while the normal, in theory, is boundless.142However, since my analyses were simple (i.e., no complex interactions likeChapter 4), involved only inferences about differences in central tendencynot variability, and since the means of the variables I measured tended to benear the center of the interval, I found that assuming a normal distributionprovided a very good approximation. To illustrate this, I have provided beta-regression confidence intervals beside normal-regression intervals in figures5.1, 5.2 and 5.3. Below I provide the normal-regression parameter estimates,since they yield the same qualitatively conclusions but a) are likely to bemore familiar to most readers, b) are easier to interpret (no link functions,beta-regression parameters are typically logit- and log-linked) and c) do notrequire an explicit regression model of variability (though do implicitly as-sume one), making it easier to communicate insights into differences betweenmeans.5.2 Results, study one5.2.1 Do these questions index the same underlyingconstruct?To answer this question we assessed the correlations between participants?answers.Participants who thought the profiteer was not a good person were morelikely to disapprove of his actions (r = 0.5), and thought themselves lesslikely to be friends with him (r = 0.61). Participants who disapproved ofthe profiteer?s actions were less likely to be friends with him (r = 0.47).Given these intermediate correlations, it is unclear whether these variablesare all indexing a single underlying psychological disposition towards theprofiteer or distinct but related judgements. As such, we report results foreach variable and for an aggregated constructed by taking their arithmeticmean within participants.1435.2.2 Was condemnation of the lucky profiteer sensitive tothe victims? reputation?Participants who saw the refugee scenario condemned the profiteer and hisactions, and did so more than participants who saw the criminal scenario,who did not differ from a neutral response.Including participants age and sex as covariates in regression modelsnever produced even marginally significant predictors (p > .5), never im-proved model fit (p > .9) and did not qualitatively change the direction orsignificance levels of other model parameters.The mean response of participants who saw the refugee scenario did dif-fer significantly differed from ?Neither agreement nor disagreement? for eachquestion and their aggregate. However, participants who saw the criminalcondition did not differ significantly from responding ?Neither?. Moreover,experimental condition was a significant predictor of the mean of partici-pants? ratings on all three measures and their aggregate (see Table 5.1 andfigure 5.1).Unlike Inbar et al., we did not include a contrast condition in study1 where in which the profiteer benefited from the disaster not occurring.However, to the extent that the lucky profiteer effect requires participantsto judged an agent more strongly than merely ?neither agreeing nor disagree-ing?, we only saw this effect when victims had good (refugees) but not bad(criminals) reputations.5.3 Methods, study twoStudy one suggests that the lucky profiteer effect is be ameliorated when thevictim is poorly reputed. However, it has several limitations. First, I merelyassumed that criminals were worse reputed than refugees, but did not mea-sure it. Second, I used a WEIRD sample (Henrich et al., 2010): Westernundergraduate students. Third, while I tested the reputational implicationsof NIR, I did not test its other key prediction: that people should havestrong intuitions about negative cooperation in particular. That is, Andrewonly had an opportunity to gain from others? loss (a negative cooperative144.Rating-0.2-0.100.1CriminalsRefugeesGood Approve(reverse coded) FriendFigure 5.1: Likelihood maximising estimates of the means of partic-ipants? ratings (bar heights), across three measures (columns),across two conditions (bar colors) with 95% confidence internals.Beta-regression estimates of these same parameters are shownin grey; their mean is a circle. Ratings represent the propor-tion of the distance between ?Strongly Disagree? and ?StronglyAgree? that participants made their mark, scaled to [-0.5, 0.5]to correspond to regression models.145Good Approve Friend Aggregate(Intercept) 0.01 (0.04) -0.00 (0.04) ?0.03 (0.04) ?0.01 (0.03)Residents: Refugees ?0.11 (0.05)* ?0.11 (0.06)^ ?0.10 (0.05)* ?0.11 (0.04)*Male? 0.03 (0.05) ?0.02 (0.06) 0.03 (0.06) 0.02 (0.05)Age (years, centered) 0.00 (0.01) 0.00 (0.02) ?0.01 (0.01) -0.00 (0.01)Table 5.1: Likelihood-maximising linear regression parameters ofmodels describing participants? normally-distributed ratingsof the lucky profiteer, on our three measures and their ag-gregate (columns). Rows show the linear relationships of thelisted predictors to these ratings. The dependent variable isthe proportion of the distance along the rating line at whichparticipant?s made their mark, scaled to [-0.5, 0.5] to facilitatestatistical comparisons with the neutral mid-point.**: p < .01, *: p < .05dilemma), not to suffer a loss so that others could gain (a positive coopera-tive dilemma). This second study is designed to address these limitations.I recruited 180 North Americans (48% male; aged 18 to 74, mean 36years, std.dev. 13.9 years) via an online service (Rand, 2012, Amazon Me-chanical Turk). I tracked both participants? IP-addresses and Amazon Me-chanical Turk ?workerIDs? to help ensure each was unique. I included a singlequestion to test whether participants? were paying attention and excludedanyone who failed it (11 individuals). I also asked participants to write ashort paragraph explaining their decisions. All included participants wrotecoherent paragraphs which agreed with their ratings.Participants again read a paragraph about the island of Floret and ?Alex?a stock market trader (the profiteer), with several new systematic differences.The residents of Floret were now one of:(Criminals) War criminals fleeing justice during a brutal civil war in anearby nation. These former warlords caused great suffering to inno-cent civilians in their efforts to accrue power and wealth for themselves.(Refugees) Refugees fleeing persecution during a brutal civil war in a146nearby nation. These refugees experienced great suffering when theywere caught in the crossfire of local warlords? fight for wealth andpower.(Volunteers) Volunteer humanitarian aid workers, organising relief in anearby nation during a brutal civil war. These volunteers take greatpersonal risks to ease the suffering of innocent civilians, caught in thecrossfire of local warlords? fight for wealth and powerVolunteers (who provide costly help to others, relative to Refugees) wereadded to investigate whether the reputational effects in study one are sen-sitive to any change in reputation, or are merely a response to very lowreputations (Criminals). In study two I asked participants to judge boththe residents of Floret and Alex, to measure whether this reputational ma-nipulation was effective.The profiteer now made one of two kinds of investments (emphasis didnot appear in the originals):(Miracle; negative cooperation) Alex deliberately and carefully constructeda complex portfolio of stocks, options and insurance policies whichwould only yield a large profit in the unlikely circumstance that some-thing wonderful happened on Floret, something which brought wealthand prosperity to Floret?s residents. Otherwise, if something wonderfuldid not happen on Floret, the investment would yield a small loss.(Disaster; positive cooperation) Alex deliberately and carefully constructeda complex portfolio of stocks, options and insurance policies whichwould only yield a large loss in the unlikely circumstance that some-thing terrible happened on Floret, something which brought sufferingand ruin to Floret?s residents. Otherwise, if something terrible did nothappen on Floret, the investment would yield a small profit.The addition of a corresponding opportunity for ?lucky? positive coop-eration let me evaluate the sensitivity of the lucky profiteer effect to thevalence of cooperation.147Finally, I now also varied the consequences for Floret between three sce-narios: Intention Only, Lucky and Causal. I have further separated theseby investment type and floret resident type. I have highlighted subtle dif-ferences in italics. These italics were not in the originals.(Intention Only, Disaster) In the end nothing terrible happened on Flo-ret, life went on as normal for Floret?s residents. So, as expected,Alex?s investment did not result in a large profit, instead it yielded asmall loss.(Intention Only, Miracle) In the end nothing wonderful happened onFloret, life went on as normal for Floret?s residents. So, as expected,Alex?s investment did not result in a large loss, instead it yielded asmall profit.(Lucky, Disaster) A few months later something terrible did happen onFloret that changed its residents? lives. A terrible hurricane hit Floret.Many of the [criminals/refugees/volunteers] were killed and the restwere left homeless and starving.(Lucky, Miracle) A few months later something wonderful did happen onFloret that changed its residents? lives. Gold was miraculously foundon Floret. Using this new found wealth, many of the? criminals were able to bribe their ways out of prosecution andbegan undertaking even more ambitious crimes.? refugees were able to successfully start new lives in peaceful coun-tries.? volunteers were able to successfully carry out their humanitarianprojects and ease the suffering of others.(Causal, Disaster) [(Lucky), followed by:] It also turns out that, by astroke of luck, Alex found out about this impending hurricane shortlybefore it occurred, but long after Alex?s investment was set in stone.Alex may have been able to warn the people of Floret so they could148evacuate. This may have prevented Floret?s residents from sufferingthe disaster and stopped Alex from making a profit; but Alex chose todo nothing.(Causal, Miracle) [(Lucky), followed by:] It also turns out that, by astroke of luck, Alex found out about this impending gold discoveryshortly before it occurred, but long after Alex?s investment was set instone. Alex may have been able to inform various governments thatcould stake a claim on the gold. This may have prevented Floret?sresidents from benefiting from the discovery and stopped Alex frommaking a loss; but Alex chose to do nothing.NIR predicts, and previous work in other domains has found (Baron &Ritov, 2004; Cushman et al., 2006; Spranca et al., 1991), that Alex?s actionsshould have greater reputational consequences if they involve commission(i.e., making a phone call), rather than omission (i.e., not making a call eventhough he could have). To avoid conflating this effect with the mere additionof causality, Alex?s choice in ?Causal? scenarios is frame as an omission.In this study I measured reputation judgements using the same mea-sures described in Chapter 4. This facilitates more direct comparison be-tween the two studies, and allows us to assess a proxy for participants? ownbehaviour in both positive (Favour measure) and negative (Money mea-sure)dilemmas.As a reminder, these measures are:Good How good a person they believe each individual is. Boundary labels:?Good? and ?Bad?.Like How well they think they would like each individual. Boundary labels:?Like? and ?Dislike?.Favour How likely they would be to do a slightly inconvenient favour forthe individual. Boundary labels: ?Do Favour? and ?Don?t do Favour?.Money How likely they would be to return ten dollars that the individualhad dropped. Boundary labels: ?Return? and ?Keep?.149Good Liking Favour MoneyGood 0.87 0.71 0.45Liking 0.83 0.75 0.46Favour 0.52 0.54 0.59Money 0.41 0.41 0.5Table 5.2: Correlations among study two dependant variables. Corre-lations between ratings of Floret residents are above the diagonal,correlations between ratings of the profiteer are below.Participants answered by moving a slider between the two boundarypositions, which was initially set at the exact center (0.5 in figures andregression models).5.4 Results, study two5.4.1 Do these questions index the same underlyingconstruct?Table 5.2 presents the correlations among our dependent variables. Since ourtwo measures of participants? cognitive representations of reputation (Goodand Liking) were highly correlated (.83 for profiteer ratings, .87 for Floretresident ratings), we simplified analyses by taking their average as an indexof participants? cognitive representations of an individuals Reputation.5.4.2 Was the manipulation effective?In the case of criminals, yes. Participants consistently gave criminals lowerratings than refugees and volunteers across all measures (see Figure 5.2 andTable 5.3).In the case of volunteers, less so. Though participants? Reputationratings of volunteers were significantly higher than their ratings of refugeesthis effect was about six times smaller than the difference between refugeesand criminals, and was not reproduced by other measures.Our scenarios seem to have more effectively caused participants to en-150.IntentionOnly-0.50- MiracleCriminalsRefugeesVolunteersFigure 5.2: Means (bar heights) of participants? ratings of Floret res-idents (coloured columns), divided by the two kinds of cooper-ative dilemmas implied by the profiteer?s investment (negative:Disaster, and positive: Miracle), and the three kinds of con-sequences (outer rows), with 95% confidence internals inferredfrom normal (black) and beta (grey, means shown by circles)distributions. Lower bars imply that participants had a loweropinion of the residents being judged.151Reputation Favour Money(Intercept) 0.26 (0.04)** 0.40 (0.05)** 0.52 (0.05)**Age (years, centered) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00)*Male? 0.01 (0.03) 0.00 (0.04) 0.00 (0.04)Residents: Criminals ?0.50 (0.04)** ?0.47 (0.04)** ?0.27 (0.05)**Residents: Volunteers 0.08 (0.04)* 0.05 (0.05) ?0.03 (0.05)Scenario: Causal 0.01 (0.04) ?0.03 (0.05) ?0.11 (0.05)*Scenario: Intention Only ?0.03 (0.04) 0.00 (0.05) ?0.10 (0.05)^Investment Type: Miracle 0.01 (0.03) ?0.12 (0.04)** ?0.08 (0.04)^Table 5.3: Participants? ratings of Floret residents in Study 2.Likelihood maximising parameters for regression models ofparticipants? normally distributed ratings of Floret?s residents(dummy variable reference level: refugees), across our threemeasures (columns). The dependent variable corresponds to theposition of participants? sliders between the boundaries, scaledto [-0.5,0.5] to facilitate comparisons with the slider?s initialmidpoint position.**: p < .01, *: p < .05tertain more negative reputational impressions of someone for exploitingothers, than more positive opinions of someone sacrificing to help others.5.4.3 Did participants condemn the profiteer less forprofiting on the suffering of the wicked?Yes.We can see evidence of the simple lucky profiteer effect in our partici-pants? tendency to move the slider from its initial mid-point (coded as 0.5)towards negative boundary label (coded as 0). The mean Reputation rat-ing of a profiteer who profited from a disaster was 0.32 with a standard errorof 0.03 (for full model details, see Table 5.4. It is unlikely that these datawere sampled from a normal distribution with a mean of 0.5 (p < .001).However, this effect was contingent on the person being judged (see Fig-ure 5.3). Participants judged the profiteers? Reputation more positivelywhen those who suffered for their profit were criminals rather than refugees.152Table 5.4 shows the likelihood maximising regression parameters that de-scribe this relationships. Whether the profit was blind luck (Lucky scenario,? = .23, p = 0.04) or causal inaction (Causal scenario, ? = .22, p = 0.01),a profiteer benefiting from harm to criminals was judged better than onebenefiting from harm to refugees. However this reputation-contingent dif-ference did not obtain when profiteer made their investment but no disasteroccurred (Intention Only scenario, ? = .01, p = 0.93).While mere intention does not produce a second-order difference in rat-ings (i.e., between criminals and refugees/volunteers), the first order profi-teer effect still persists. Though the mean of participant?s ratings it is notquite statistically significantly different from the neutral mid-point (coded0 in Figure 5.3 and Table 5.4), it also does not significantly differ from theirratings in the Lucky scenario (refugees: p = .33; volunteers: p = 0.41). Thissecond test is also the strategy that Inbar et al.?s used to arrive at their con-vergent conclusion: that the profiteer effect persist even when the disasterdoes not occur.Participant?s were also less willing to pocket the lost money of a profiteerwho had exploited criminals in Lucky scenarios (i.e., they were more willingto defect in a negative dilemma on someone who had (luckily, indirectly)benefited from the suffering of ?good guys?), however their willingness to dothem a favour (a positive cooperative dilemma) did not covary with resi-dent reputation in these negatively-valenced ?disaster investment? dilemmascenarios.5.4.4 Did participants reward the trader for taking a loss inmiracle scenarios?That is, did they judge Alex the stock market Trader more positively whenAlex made investments that would yield a large loss when others fortuitouslygained? Did their judgements depend on who these others were, and whetherAlex causally influenced these events?The only clear effect in miracle conditions is that traders who sufferedfor the good of criminals were slightly less-well liked than those who sufferedfor volunteers or refugees (see Figure 5.3 and Table 5.5). In these cases, the153ReputationIntention Only Lucky Causal(Intercept) ?0.16 (0.10)^ ?0.20 (0.08)* ?0.42 (0.07)**Age (years, centered) 0.00 (0.00) 0.00 (0.00) -0.00 (0.00)Male? 0.10 (0.09) 0.07 (0.09) 0.08 (0.07)Residents: Criminals 0.01 (0.12) 0.23 (0.11)* 0.22 (0.08)**Residents: Volunteers -0.00 (0.11) ?0.04 (0.11) ?0.02 (0.08)FavourIntention Only Lucky Causal(Intercept) ?0.09 (0.12) ?0.06 (0.13) ?0.09 (0.13)Age (years, centered) 0.00 (0.00) 0.01 (0.01) 0.00 (0.00)Male? 0.18 (0.12) 0.17 (0.14) 0.01 (0.13)Residents: Criminals ?0.04 (0.15) 0.13 (0.17) ?0.01 (0.16)Residents: Volunteers ?0.07 (0.14) ?0.10 (0.18) ?0.17 (0.16)MoneyIntention Only Lucky Causal(Intercept) 0.09 (0.14) 0.02 (0.12) 0.23 (0.15)Age (years, centered) 0.00 (0.00) 0.01 (0.01)^ 0.01 (0.01)Male? 0.26 (0.14)^ ?0.09 (0.14) ?0.14 (0.16)Residents: Criminals ?0.01 (0.17) 0.40 (0.17)* 0.11 (0.19)Residents: Volunteers 0.03 (0.17) ?0.19 (0.17) ?0.04 (0.18)Table 5.4: Participants? ratings of the lucky profiteer in disaster sce-narios. Likelihood maximising parameter estimates for regressionmodels of participants? ratings of the Trader in ?disaster? scenar-ios (a positively valenced dilemma), for the three measures (outerrows) and three levels of causation (columns). The dependentvariable corresponds to the position of participants? sliders be-tween the boundaries, scaled to [-0.5,0.5] to facilitate comparisonswith the slider?s initial midpoint position.mean of participants? ratings for in refugee conditions were significantly morepositive than the slider?s initial half-way point, but the not in others. Thiswas particularly evident on the Favour question, a measure of participants?positively-valenced dilemma responses to an individual who they had seencooperating in a positive dilemma.Furthermore, what average differences did exist were driven by judge-ments in the ?Causal? condition. That is, we saw almost no evidence that ac-cidentally incurring costs to help someone incurs reputational consequencesthey way that accidentally profiting from their loss does.154.IntentionOnly-0.50- MiracleCriminalsRefugeesVolunteersFigure 5.3: Means (bar heights) of participants? ratings of the profi-teer (coloured columns), divided by the two kinds of coopera-tive dilemmas implied by the profiteer?s investment (negative:Disaster, and positive: Miracle), and the three kinds of con-sequences (outer rows), with 95% confidence internals inferredfrom normal (black) and beta (grey, means shown by circles)distributions. Lower bars imply that participants had a loweropinion of the profiteer.155ReputationIntention Only Lucky Causal(Intercept) ?0.12 (0.10) 0.03 (0.09) 0.26 (0.12)*Age (years, centered) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00)Male? 0.15 (0.10) 0.08 (0.10) ?0.08 (0.12)Residents: Criminals ?0.01 (0.13) ?0.13 (0.11) ?0.25 (0.13)^Residents: Volunteers 0.13 (0.12) ?0.05 (0.12) ?0.03 (0.13)FavourIntention Only Lucky Causal(Intercept) ?0.04 (0.13) 0.01 (0.10) 0.29 (0.14)*Age (years, centered) 0.00 (0.01) 0.00 (0.00) -0.00 (0.01)Male? 0.04 (0.13) 0.08 (0.12) ?0.21 (0.14)Residents: Criminals 0.11 (0.16) ?0.02 (0.13) ?0.36 (0.16)*Residents: Volunteers 0.03 (0.16) 0.06 (0.14) ?0.02 (0.15)MoneyIntention Only Lucky Causal(Intercept) 0.10 (0.15) 0.29 (0.11)* 0.41 (0.15)**Age (years, centered) 0.00 (0.01) 0.01 (0.00)* 0.00 (0.01)Male? ?0.02 (0.16) ?0.12 (0.14) ?0.11 (0.15)Residents: Criminals 0.17 (0.20) 0.07 (0.15) ?0.12 (0.17)Residents: Volunteers ?0.03 (0.19) ?0.03 (0.16) ?0.03 (0.16)Table 5.5: Participants? ratings of the lucky profiteer in miracle sce-narios. Likelihood maximising parameter estimates for regressionmodels of participants? ratings of the Trader in ?miracle? scenar-ios (a positively valenced dilemma), for the three measures (outerrows) and three levels of causation (columns). The dependentvariable corresponds to the position of participants? sliders be-tween the boundaries, scaled to [-0.5,0.5] to facilitate comparisonswith the slider?s initial midpoint position.5.5 DiscussionIn two studies we detected a previously unobserved nuance in Inbar et al.?sLucky Profiteer effect. Our participants only condemned those who profitedon others? misfortune, even if only by luck, when the victims were well-reputed (refugees fleeing persecution, or volunteers in a war-torn country),not ill-reputed (criminals fleeing persecution). This is the pattern of moralintuitions that NIR models assume shaped our ancestral socio-ecology.I also observed a clear negativity-bias (Baumeister et al., 2001; Cacioppo156& Berntson, 1994; Rozin & Royzman, 2001). The Lucky Profiteer effect wasonly clearly present in negative cooperative dilemmas (where the profiteermakes a profit from another?s large loss), not positive ones (where theysuffer so others may profit). Also, describing of an island of people whoprofited from other?s misery (war criminals) caused participants? esteemto fall far more than describing people who made great sacrifices to helpothers (volunteer aid workers) caused them to rise. These unambiguousnegativity biases in reputational evaluation match the ancestral patterns ofreputational evaluation assumed by NIR.In addition, I observed that though causal influence is not essential forpeople to condemn the coincidence of one person?s profit from another?s loss,adding an opportunity for causal influence greatly exacerbates the effect.Finally, as in chapter 4, these data suggests that people tend to respond?in kind? to valenced dilemmas. That is, our positively-valenced Favourshowed stronger reactions to positive dilemmas and our negatively-valencedMoney measure responded more strongly in negative dilemma scenarios.This supports NIR?s novel emphasis on the valence of cooperative dilemmas,both now and historically.From an ultimate explanatory perspective, NIR is agnostic to the prox-imate logic of why our moral intuitions follow these pattern. It may be, asInbar et al. suggest, that we infer the profiteer?s wicked desires. Thoughthis account does not prima facie explain why these effects should dependon the victim?s reputation, or why this should cease to be the case if no con-sequences follow. Or perhaps it is connected to feelings of fictive-affiliationor friendship with well-reputed victims but not ill-reputed ones, or it maysimply be a raw, irrational, genetically-encoded moral intuition. Ultimate,evolutionary models like NIR are concerned with teasing out the long-run,dynamic evolutionary consequences of behaviour, whatever the proximatemotivations that cause it.While our study supports NIR as a model of the social dynamics thatshaped our ancestors? moral intuition, no single feature of contemporarypsychology can, in isolation, prove an ultimate, evolutionary hypothesis.Any human behaviour we measure today is shaped by genetically evolved157dispositions, individual learning and development and, crucially, our richcorpus of culturally transmitted knowledge. There will always be manyproximate explanations that fit contemporary psychological facts better thana distal hypothesis, whose predictions are necessarily less precise (though thetwo levels of explanation often support one another).Rather, formally specified evolutionary theories ought to generate abroad range of predictions across many levels of explanation, including psy-chological phenomena, sociological and cross-cultural distributions of be-haviour, patterns of historical change, and so on. The relevance of an ul-timate theory to understanding our moral intuitions is arbitrated by theempirical fit of the breadth of its explanatory scope?it?s ability to connecta broad range of otherwise seemingly disconnected findings from across psy-chology and other disciplines?rather than its specificity for any particularphenomenon. Testing ultimate theories is necessarily an interdisciplinaryeffort, to which this study is a contribution.Here we take a first step by testing one central cognitive prediction ofNIR in one contemporary social group. Conclusively testing whether thelucky profiteer effect is consistent with NIR?s assumptions requires a farmore ambitious project; for instance, systematic cross-cultural studies (e.g.,Henrich et al., 2005) or comparisons of cultures whose phylogenetic relation-ships are known (e.g., Currie et al., 2010; Mace & Jordan, 2011).In the meantime, we can ask whether it is even plausible that our moralintuitions were shaped by ancestral social dynamics at all? We believe it is.Our genome bears the signature of recent and ongoing positive selection (La-land et al., 2010; Nielsen et al., 2007) and some digestive adaptations (e.g.,lactase persistence: Ingram et al., 2009; alcohol metabolism: Peng et al.,2010) have emerged and spread through human populations since the ad-vent of agriculture (i.e., on time-scales of thousands of years). It is plausiblethat cognitive adaptations that make our moral intuitions sensitive to repu-tation have done the same, so long as we can be confident that reputationscarried potent fitness consequences for similar, or longer, timescales.While reputations do not directly leave archaeological footprints, thereis some indirect evidence of their presence. Anatomically modern humans?158whose skulls closely resemble ours, and who were at least physiologicallycapable of gossip?began rapidly migrating across the world?s diverse ecolo-gies by 50 thousand years ago (Endicott et al., 2009; Hudjashov et al., 2007).Our less cognitively-modern hominids cousins achieved a similar feat over amillion and a half years earlier. Convergent evidence from multiple far-flungsites makes a convincing case that hominids had mastered the controlled useof fire between 200-300 thousand years ago (e.g., Karkanas et al., 2007; Roe-broeks & Villa, 2011), and plausible but more controversial case that fireuse may have been present up to almost 2 million years ago (e.g., Clark &Harris, 1985; Gowlett et al., 1981; Wrangham, 2009). Together these sug-gest that up to two million years ago genus homo may have coordinatednovel forms of cooperation (e.g., cooperative hunting) in new ecologies andtransmitted complex cultural information (e.g., fire-use techniques, foragingand hunting techniques, ecological-lore, etc.). Transmission of complex cul-ture is itself a cooperative dilemma, since dependence on culturally learnedknowledge requires that individuals trust others, making them susceptibleto exploitative deception. Since reputations are one of the most plausiblemechanisms our early ancestors had for rapidly generating and sustainingcooperation in arbitrary domains, they have likely shaped our intuitions fortens of thousands of years (since our recent, rapid migration from Africa),and plausibly for hundreds of thousands (since our ancestors were anatomi-cally modern) or millions (since the first migrations from Africa, the adventstone tools and fire use); long enough to leave their mark on our genomeand our cognition.If, as our study suggests, further investigation concluded that NIR dy-namics did shape human moral intuitions?such that we are readily inclinedto exploit those with bad reputations, and to condemn others for doingso?we would be in possession of a powerful explanatory tool. It could, forinstance, help make sense of many innocuous every day phenomena, suchas why many of us only feel excited when an action movie hero kills many?bad guys?, but gnash our teeth when a villain in the same movie threatensa single innocent individual. More practically, this evolved moral intuitioncould help explain and intervene in the recurrent emergence of bullying in159schoolyards across the world; where some low status children are systemat-ically degraded and exploited by some of their peers, while others look onwith little or no moral outrage or inclination to intervene. More potentlystill, an NIR account of the Innocent Profiteer effect could offer us tractionon the widespread historical and contemporary (Adinkrah, 2004) incidenceof witchcraft accusations; where the coincidence of some people?s loss withother, low-status individuals? gain (or lack of similar loss) prompts violent,sometimes deadly, retaliation by their societies.As ever more comparative (e.g., Kitayama & Uskul, 2011; Nisbett, 2004)and phylogenetic (e.g., Currie et al., 2010; Mace & Jordan, 2011) researchdemonstrates that deep difference in human reasoning and behaviour aretransmitted culturally, the onus increasingly falls on psychologists to movebeyond extrapolating from convenient but unrepresentative western under-graduate samples (Henrich et al., 2010), and begin detailing the specificdevelopmental, cultural and evolutionary trajectories that shape psycholog-ical effects. This especially true for moral psychology, where the universalityand origins of moral intuitions are a focus of inquiry (e.g., Mikhail, 2007).Extrapolating and testing the predictions of formal evolutionary models isan important and underutilized tool for probing the ultimate roots of mod-ern moral intuitions. We hope that this simple study can provide a modelfor further integration of formal evolutionary theory and moral psychology.160Chapter 6ConclusionsTypical graduate dissertations in social psychology focus on one specific em-pirical phenomenon, such as prejudice, helping or religious belief. Socialpsychologists probe the intricacies of such phenomena with clever, some-times wonderfully creative empirical techniques. Theoretical descriptionsof social psychological phenomena usually evolve alongside the discovery ofnew evidence about them. The theoretical terms in which these descriptionsare couched mix together concepts from other social-psychological investi-gations, folk-psychology and culture-specific ideas about the functional en-tities that inhabit minds. Often the assumptions made by such theories areimplicitly hidden in the chaotic intellectual history of these terms and con-cepts. Gradually, collectively, these assumptions and concepts are refinedas researchers (who are motivated to publish papers) preferentially borrowand recombine the concepts that have previously been most successful ingenerating exciting empirical effects.There is a lot to recommend this approach. For one thing it is highlygenerative. Social psychologists have unveiled an impressive terrain of pre-viously unsuspected human peculiarity (for the distribution of effect sizesthey have discovered, see Richard et al., 2003). When the microscope wasfirst discovered, it was a terrific idea for countless scientists to peer throughit at whatever struck their fancy and describe the most exciting things theysaw. This early excited phase of exploration provides the raw fodder from161which more general theories are built. Our recently discovered statisticaland experimental methods are like a macroscope, lending us vision of an-other world not visible with the naked eye. It is a world inhabited by subtlepatterns of human behaviour and judgement only apparent on aggregate, inthe mean differences between individuals; individuals experimentally influ-enced just so.I took a different approach.Rather than trying to explain a particular contemporary empirical phe-nomenon, this dissertation is a contribution to the top-down challenge ofretracing the biological origins of our social psychology. Rather than recom-bining the stew of often only partially mutually-consistent contemporarypsychological concepts into a good description of one aspect of our socialpsychology and working up towards a more general understanding, this dis-sertation embarks from a purely theoretical challenge: explaining how ourancestors? biological realities could have given rise to a social psychology inthe first place.To rigorously confront this challenge, I have restricted myself to onlyinvoking concepts, terms and ideas that can be clearly, formally definedfrom the stand-point of evolutionary biology. I only allowed myself to useideas that I would make sense to an abstract, non-human, external observerwho had so far only managed make sense of chimpanzees.I looked for combinations of these concepts which could generate thekind of explosion of cooperative, complex, cultural peculiarity that charac-terises humans today. I considered only combinations that could be statedwith formal, mathematical rigour and were consistent with the principles ofevolution by natural selection. My efforts interface with those of scholarsacross the sciences, who are already embarked on this enterprise.In chapter 2, I reviewed an emerging picture of the evolutionary changesthat launched our species on its unique trajectory. At the center of thispicture is the idea that ?culture??our ability to transmit complex, encodedinformation across generations?made possible exciting new forms of social-ity. It let us cooperate in ways that no other species has previously accom-plished. These accounts are plagued by a puzzle: the cooperative dilemma162of culture, the evil teacher problem. An evolving species cannot accumulatea shared corpus of a sophisticated cultural knowledge until some very gen-eral, very robust mechanism ensures they cooperate. Why? Because cultureitself is a valuable public good that you can only acquire from other minds,and so is eminently exploitable.In chapter 3, I proposed a novel solution to the cooperative dilemmaof culture: Negative Indirect Reciprocity. I took one of the best existingcandidates for a uniquely-human solution to the dilemma (reputation) andhoned it until it assumed as little pre-existing culture as possible. This newmodel tells a precise story about one part of humans? transition to modernitywhich, if accurate, can help explain why more sophisticated behaviours,cognitive adaptations and institutions exist. The model hinges on severalkey assumptions about our ancestors? now unobservable psychology.In chapter 3, I went on to argue that robust biases in contemporary hu-man cognition?negativity- and omission-biases?provide support for theseassumptions. However NIR?s assumptions are also more specific. NIR as-sumes that ancestral patterns of indirect reciprocity (in particular, 2nd-orderreputational judgements) were biased in specific ways. To the best of myknowledge, prior to this dissertation no direct evidence existed of negativity-and omission-biases in 2nd-order reputational judgements.NIR?s assumptions can be tested using the techniques of social psychol-ogy. If our ancestor?s socio-ecology was biased in the ways assumed byNIR, then our contemporary reputational judgements ought to be similarlybiased. The subsequent chapters carried out these tests.From the vantage of NIR, there are a real and important differencesbetween positive cooperation?paying an absolute cost to beget an abso-lute benefit for others?and negative cooperation?paying a relative cost(by forfeiting the benefits of exploitation) to bring about a relative benefit(not having to suffer a loss). This stands in stark contrast the perspec-tive implied by the mainstay of current empirical and theoretical inquiryinto cooperation, which typically only considers positive cooperation andpunishment (paying a cost to cause a cost, a very different phenomenon).Our current scholarly paradigm implicitly (and even explicitly in conver-163sations I?ve had with influential scholars) assumes that positive and nega-tive cooperation are equivalent?that insights about one are mere corollariesof investigations of the other. The evidence presented in this dissertationsspeaks against this assumption. Chapters 4 and 5 documented studies wherea small manipulation of dilemma valence?the difference between addingand taking pieces of paper from an unattended pile in chapter 4, betweeninvesting indirectly in a disaster or a miracle in chapter 5?generated vastlydifferent moral and reputational reactions from participants. NIR (chapter3) suggests that the valence of cooperation matters in theory, chapters 4and 6 show that it matters in practice too.An NIR perspective on our evolutionary history assumes that negativedilemmas influence reputations more than positive ones. This was evidencedin contemporary cognition by both chapters 4 and 5. In both cases manip-ulations designed to worsen an individual?s reputation by making it clearto participants that they exploit others were more effective than attemptsto improve their reputation by making it clear they help others. That is,negative dilemmas seem to have a stronger effect on reputations than pos-itive ones. This effect is consistent with the ?negativity-bias? that pervadesmany domains of human cognition. The generality of this bias suggests itmay be an ancient genetic adaptation to cognition, rather than a recentcultural adaptation in reputation assessment. This generality strengthensour inference from our contemporary observations to ancestral populations,building our confidence that our ancestors reputational evaluations wherenegativity-biased too.In both chapters 4 and 5, second-order indirect reciprocity?that is, thedegree to which the reputational consequences of an action depend on thereputation of the action?s target?was more evident (indeed, exclusively ev-ident in Chapter 5) in the context of negative cooperative dilemmas.NIR makes more precise predictions still about second order indirectreciprocity in the context negative cooperative dilemmas. It predicts that?exploiting good guys is bad?. We saw this in both studies. Individuals whoexploited a well-reputed individual were themselves judged more negativelyboth within (chapter 4) and between subjects (chapter 5). Additionally,164and critically for its dynamics to be plausible, NIR predicts that ?exploitingbad guys isn?t as bad as exploiting good guys?. Again, this is just what weobserved in chapters 4 and 5. This 2nd-order reputational effect is entailedby neither negativity- nor omission-biases, it is a prediction uniquely entailedby NIR.All up, the studies in chapters 4 and 5 support the claim that our con-temporary reputation-relevant intuitions were shaped by the ancestral so-cioecology described in chapter 3.That said, these findings should not be overstated. They are a necessaryfirst step in testing NIR. They provide initial support and motivation forfurther investigation, rather than constituting conclusive evidence on theirown. This early evidence provides justification for undertaking costlier butpotentially more influentially potent work, such as attempts to directly ob-serve NIR dynamics in contemporary small scale societies.In chapter 4, in addition to testing these predictions, I also attempted toexpand our understanding of indirect reciprocity by full surveying the spaceof possible second-order continuous reputation assessment rules. Previouswork on this topic has typically employed the method of experimental eco-nomic games. This powerful tool allows precise control of participants? mo-tivations, but pays for it with ecological validity; participants interact withanonymous representations on a computer screen rather than other real,concrete people. This previous work has drawn inconsistent conclusionsabout second-order indirect reciprocity (caring about the target of some-one?s actions, when you judge them). Some studies observe it, others don?t.Some researchers have suggested that second-order indirect reciprocity istoo difficult for humans.I suspected that people might just struggle with abstract, anonymousrepresentations of reputation. Our reciprocity-intuitions might require aspecific cognitive representation of the people involved in a situation, richwith emotions and associations. So, I made the opposite trade-off. I sacri-ficed anonymity and had participants judge very specific, though fictitious,characters involved in interactions that they might well encounter in reallife.165I found clear, cross-cultural evidence of first order indirect reciprocity,especially in negative dilemmas. I also found reasonable evidence of secondorder indirect reciprocity, especially among Americans. However, when Idiscretely categorised each participant?s judgements into one of the 81 possi-ble assessment rules, the most common was first-order scoring assessment?where a target?s reputation doesn?t matter. This suggests a puzzle for indi-rect reciprocity theorists, since our data also suggests that some people usesecond-order standing assessment, which models suggest is more evolution-arily viable.Chapter 5 is particularly exciting because it presents a novel test ofNIR. Driven by an interest in how the law accords with our moral intuitions(specifically the question of whether causation is necessary for guilt) NoelInbar and his colleagues documented a peculiar psychological phenomenon.People condemn those who profit while others lose, even when there is nocausal connection between them. I call this the ?lucky profiteer effect?. In-bar?s investigations of this phenomenon made no explicit connection to NIR,nor indirect reciprocity nor reputation more broadly. Nor was NIR devel-oped to explain this phenomenon, nor with it in mind. NIR?s ability topredict the lucky profiteer effect is therefore a genuinely novel test of thehypothesis.Though NIR superficially fits the lucky profiteer effect?both involvenegative judgements of someone who profits while others suffer?it entailsan additional prediction. The effect should be ameliorated if the target ofloss has a bad reputation. This additional prediction was not foreseen bythe discovers of the lucky profiteer effect nor, as far as I know, is it impliedor required by the moral-psychological theories on which their explanationsare based. I tested the prediction and found that it held.The patterns of variability I observed in people?s judgements of the luckyprofiteer are consistent with humans? very robust negativity- and omission-biases, supporting NIR?s inference from contemporary to ancestral cogni-tion. However the 2nd-order reputational effect?the amelioration of thesepatterns when victims are badly reputed?is uniquely predicted by NIR.The most impactful social psychological discoveries usually excite and166surprise us because they are counter-intuitive. This refinement of the luckyprofiteer effect is just the opposite. It seems entirely natural to us that peoplewould condemn those who profit on the suffering of others, but less so if theirprofiteering on the misfortune of ?bad guys?. This lack of novelty is just whatwe would expect if an NIR socioecology had shaped our intuitions. Whatmakes this observation exciting is not its novelty, but the fact that the theorythat predicted it was derived a priori from theoretical considerations ofearly human evolution, rather than being an empirically-refined descriptionof people?s moral behaviour.6.1 Near future directionsThis dissertation documents early theoretical and empirical steps towardsunderstanding the ancestral socio-ecological dynamics that shaped our socialpsychology. Much remains to be done.The theoretical model in chapter 3 demonstrates the plausibility of a so-cioecology where selfish, individually adaptive, promiscuous social learninggives rise to reputations. These in turn foster a socioecology where selfish,individually adaptive exploitation actually promotes prosociality, formingthe substrate of more complex forms of information sharing and collectiveaction. However chapter 3 is just one utterance in an ongoing conversationabout the circumstances that set our species in motion. It answers somequestions (?could indirect reciprocity arise in a species without culturallywell-defined social roles and responsibilities??), but raises others. Theseinclude? Could selection hone strategies to prey on the socioecology of indirectreciprocity, as it could for direct reciprocity (e.g., Boyd & Lorberbaum,1987; van Veelen et al., 2012)?? How does migration between groups change NIR dynamics? This is aparticularly pressing question since reasonable amounts of migrationcould have been a prerequisite for the accumulation of complex culture(Powell et al., 2009).167? What other mechanisms could solve the cooperative dilemma of cul-ture? Plausible candidates include cognitive mechanisms for reducingcredulity (e.g., Henrich, 2009), and learning biases that create strongasymmetries in social influence (e.g., Henrich & Gil-White, 2001) suchthat the most influential individuals benefit by having their cooper-ative acts widely imitated. Do these mechanisms synergise with ordisrupt NIR?On the empirical front, chapter 4 suggests that NIR fits our reputationalintuitions and chapter 5 suggests it can explain surprising novel phenomena.However both studies have limitations.Though I strove to carefully balanced the positively and negatively va-lenced scenarios in chapter 4?they differed only in whether an individualadded or took pieces of paper from a pile?they may still have tapped verydifferent, culturally-established norms. It is possible that these particularnorms dictated our participants? very different responses rather than thevalence of the cooperative dilemma itself.A clear answer will not emerge until these methods are replicated withmany different scenarios, each carefully designed to compare positively- andnegatively-valenced variants of its cooperative dilemma.It is also possible that the patterns observed here are the consequence ofconvergent cultural evolution. For instance, patterns in participants? judge-ments could have been artefacts of culturally evolved nuances in how peopleapply moral labels such as ?good person? (e.g., Gidron et al., 1993; Skowron-ski & Carlston, 1987). The convergent evidence from my more behaviouralquestions (about doing people favours and returning their money) help easethis concern, but the strongest evidence would require cross-cultural be-havioural studies.I merely took the first steps towards to providing a cross-cultural ground-ing for these effects. I observed that Americans and Indians both showedsimilar trends, though the Americans? were substantially greater in magni-tude. A strong genetic interpretation of NIR predicts that these reputationalintuitions should be common to all humans. We cannot confidently test this168claim by examining any two cultures, however distinct; we would need toask similar questions of many different, culturally remote individuals.The results I documents in chapter 4 suggest that people do make 2nd-order reputational judgments, but that these might not be easily tappedby the abstract, anonymous, impersonal methods of standard experimen-tal economic games. This opens the door for creative researchers to designnew non-verbal and behavioural studies of peoples? reputational judgmentsthat do not require anonymity. That is, the challenge is now to test indi-rect reciprocity, including negative indirect reciprocity, by experimentallymanipulating what people?s real reputational reactions are to a situation,rather than merely asking them to imagine and report what they would do.The profiteer effect examined in chapter 5 is also a first step. Whileit is exciting that NIR accurately predicts novel phenomena, the real testof a distal theory is whether it fits many different phenomena across manydifferent contexts and domains. While future work could certainly furtherdocument the details of the lucky profiteer effect, the best test of NIR willbe whether it meets the challenge of new, very different novel tests.6.2 Distant future directionsI open this work by raising the possibility that there might exist some centralexplanatory principles that provide a deep, satisfying answer the question?what am I?? I hoped my work would contribute to the long road to theirdiscovery. But how could insights into ancestral social dynamics possibly dothat?Imagine that, by formally exploring the theoretical possibilities and test-ing their implications for contemporary cognition, we did eventually arriveat a correct and mostly complete description of the socio-ecological dynam-ics that drove our ancestors along the evolutionary road to us. Why wouldknowing about ancient history help us understand ourselves today?Minds are information processing machines. They use sensory infor-mation to reconstruct an external reality. The functional and mechanisticdetails of how they do this are being carefully and skilfully reconstructed by169cognitive and developmental psychologists. However to achieving this endwe may also require an objectively correct description of the external worldthat our minds are grappling with. In particular, of the socio-ecologicaldynamics that result from minds sharing information.The core insight of culture-gene coevolutionary theories is that our cog-nitive reconstructions of the outside world fundamentally changed when ourancestors began honestly transmitting complex cultural information. Thereconstructive process became a collaborative endeavour, one that is muchmore powerful and generative, and builds a far more accurate picture thanany of us could accomplish alone. Understanding the details of this pro-cess is central to understanding humans. It involves understanding the coredilemmas and social dynamics that cultural learning generates, and the waysour minds have responded and adapted to them.Today the realities our minds reconstruct are fantastically complicated.They are peopled by entities such as ?software?, ?silent partners?, ?symbiosis?,?sacrilege?, ?self-esteem? and the ?sanguine humor?. These representationsreally do change how we act, interact, think, form relationships, build andsustain institutions, and so on. Much of who we are depends on how weacquire these ideas and employ them. I believe that if our models of humancognition are to ever come close to grappling with this complexity, theymust intersect with accurate models of the cultural information landscapeon which it occurs.The interface between ?models of the origin and evolution of culture?and ?models of psychology and cognition? is exciting. It suggests that soci-ology could interweave tightly with social psychology, allowing us to buildecological models that simultaneously understand the complex interplay ofindividual minds and emergent social institutions and phenomena, ratherthan assuming that one is constant while we examine the other. It holdsthe prospect of new approaches to history which explicitly incorporate ourgrowing understanding of both the universal and culturally mailable waysthat minds interact with evolving cultural institutions and concepts. Itsuggests the possibility of a personality psychology that could describe andpredict an ?ecology of personalities? for a given contemporary or even histor-170ical group, including their long-run frequencies, developmental trajectoriesand interactions between them, rather than just post-hoc cataloguing per-sonality differences and their correlates. It is, in short, a promising spaceof ideas in which we might find an accurate model of human minds and thecomplex ecological interactions between them. It is a space in which wemight answer ?who am I??The bottom-up work of accurately modeling of our cognitive processesreceives a great deal of funding, attention and researcher effort. Top-downinquiry into the origins and nature of the cultural socioecology our mindsinhabit is comparatively scarce. I hope this dissertation contributes tothis top-down effort in three ways. First, by highlighting the cooperativedilemma of culture. This will hopefully restrict the space of unproductive,implausible speculation and help focus inquiry into theoretically potent puz-zles. Second, by drawing attention the importance of dilemma valence?aninfluential dimension of cooperation and one that might be pivotal to solvingthe evil teacher problem, but that has been largely ignored thus far. Third,by drawing attention to one particular socio-ecological dynamic, NegativeIndirect Reciprocity, which may have been and may continue to be a driverof the cooperative foundation of our social psychology.171BibliographyAbbot, P., Abe, J., Alcock, J., Alizon, S., Alpedrinha, J. A. C., Andersson,M., Andre, J.-B., van Baalen, M., Balloux, F., Balshine, S., & Others(2011). Inclusive fitness theory and eusociality. Nature, 471(7339),E1?-E4. ? pages: 14Adinkrah, M. (2004). Witchcraft accusations and female homicidevictimization in contemporary Ghana. Violence against women, 10(4),325?356. ? pages: 141, 160Aiello, L. C., & Wheeler, P. (1995). The expensive-tissue hypothesis: thebrain and the digestive system in human and primate evolution. Currentanthropology, 36(2), 199?221. ? pages: 24, 29, 89Axelrod, R., & Hamilton, W. (1981). The evolution of cooperation.Science, 211(4489), 1390?1396. ? pages: 21, 89, 138Barkow, J. H., Cosmides, L., & Tooby, J. (1992). The Adapted Mind:Evolutionary Psychology and the Generation of Culture. OxfordUniversity Press. ? pages: 9Baron, J., & Ritov, I. (2004). Omission bias, individual differences, andnormality. Organizational Behavior and Human Decision Processes,94(2), 74?85. ? pages: 33, 46, 134, 149Barrett, H. C., & Broesch, J. (2012). Prepared social learning aboutdangerous animals in children. Evolution and Human Behavior. ?pages: 12Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001).Bad is stronger than good. Review of general psychology, 5(4), 323. ?pages: 4, 32, 46, 134, 156172Bell, A. V., Richerson, P. J., & McElreath, R. (2009). Culture rather thangenes provides greater scope for the evolution of large-scale humanprosociality. Proceedings of the National Academy of Sciences of theUnited States of America, 106(42), 17671?17674. ? pages: 10Berger, U. (2011). Learning to cooperate via indirect reciprocity. Gamesand Economic Behavior, 72(1), 30?37. ? pages: 93Birch, S. A. J., Akmal, N., & Frampton, K. L. (2009). Two-year-olds arevigilant of others? non-verbal cues to credibility. Developmental science,13(2), 363?369. ? pages: 25Birch, S. A. J., Vauthier, S. A., & Bloom, P. (2008). Three-andfour-year-olds spontaneously use others? past performance to guide theirlearning. Cognition, 107(3), 1018?1034. ? pages: 25Bolton, G. E., Katok, E., & Ockenfels, A. (2005). Cooperation amongstrangers with limited information about reputation. Journal of PublicEconomics, 89(8), 1457?1468. ? pages: 97, 140Boyd, R., Gintis, H., & Bowles, S. (2010). Coordinated punishment ofdefectors sustains cooperation and can proliferate when rare. Science,328(5978), 617?620. ? pages: 20, 28, 137, 139Boyd, R., & Lorberbaum, J. P. (1987). No pure strategy is evolutionarilystable in the repeated Prisoner?s Dilemma game. Nature, 327 , 58?59. ?pages: 21, 89, 138, 167Boyd, R., & Richerson, P. J. (1988). Culture and the evolutionary process.University of Chicago Press. ? pages: 13, 24, 26, 53Boyd, R., & Richerson, P. J. (1989). The evolution of indirect reciprocity.Social Networks, 11(3), 213?236. ? pages: 93Boyd, R., & Richerson, P. J. (1996). Why culture is common, but culturalevolution is rare. In Proceedings of the British Acadamy, vol. 88, (pp.77?94). ? pages: 13Boyd, R., & Richerson, P. J. (2005). The Origin and Evolution of Cultures(Evolution and Cognition). Oxford University Press, USA. ? pages: 10,11, 13, 30, 139Boyd, R., Richerson, P. J., & Henrich, J. (2011a). Rapid culturaladaptation can facilitate the evolution of large-scale cooperation.Behavioral ecology and sociobiology, 65(3), 431?444. ? pages: 14173Boyd, R., Richerson, P. J., & Henrich, J. (2011b). The cultural niche:Why social learning is essential for human adaptation. Proceedings ofthe National Academy of Sciences, 108(Supplement 2), 10918?10925. ?pages: 24Brandt, H., & Sigmund, K. (2005). Indirect reciprocity, image scoring, andmoral hazard. Proceedings of the National Academy of Sciences of theUnited States of America, 102(7), 2666?2670. ? pages: 140Brandt, H., & Sigmund, K. (2006). The good, the bad and thediscriminator?errors in direct and indirect reciprocity. Journal oftheoretical biology, 239(2), 183?94. ? pages: 94Brandt, H., Sigmund, K., & Others (2004). The logic of reprobation:assessment and action rules for indirect reciprocation. Journal ofTheoretical Biology, 231(4), 475. ? pages: 140Brown, C., & Laland, K. N. (2003). Social learning in fishes: a review.Fish and Fisheries, 4(3), 280?288. ? pages: 11Buttelmann, D., Zmyj, N., Daum, M., & Carpenter, M. (2012). SelectiveImitation of In-Group Over Out-Group Members in 14-Month-OldInfants. Child Development. ? pages: 25Cacioppo, J. T., & Berntson, G. G. (1994). Relationship between attitudesand evaluative space: A critical review, with emphasis on theseparability of positive and negative substrates. Psychological bulletin,115(3), 401. ? pages: 32, 46, 134, 156Card, N. A., Stucky, B. D., Sawalani, G. M., & Little, T. D. (2008). Directand indirect aggression during childhood and adolescence: ameta-analytic review of gender differences, intercorrelations, andrelations to maladjustment. Child development, 79(5), 1185?229. ?pages: 44Carpenter, M., Uebel, J., & Tomasello, M. (2013). Being MimickedIncreases Prosocial Behavior in 18-Month-Old Infants. Childdevelopment. ? pages: 26Cavalli-Sforza, L. L., & Cavalli-Sforza, L. L. (2000). Genes, People, andLanguages. North Point Press. ? pages: 10174Cavalli-Sforza, L. L., & Feldman, M. W. (1981). Cultural Transmissionand Evolution: A Quantitative Approach.(MPB-16), vol. 16. PrincetonUniversity Press. ? pages: 13Charness, G., Du, N., & Yang, C.-L. (2011). Trust and trustworthinessreputations in an investment game. Games and Economic Behavior,72(2), 361?375. ? pages: 96Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: Theperception?behavior link and social interaction. Journal of personalityand social psychology, 76(6), 893. ? pages: 26Chartrand, T. L., & van Baaren, R. (2009). Human mimicry. Advances inexperimental social psychology, 41, 219?274. ? pages: 26, 27Chudek, M., Brosseau-Liard, P., Birch, S., & Henrich, J. (2013a).Culture-gene coevolutionary theory and children?s selective sociallearning. In B. Banaji, & S. Gelman (Eds.) Navigating the Social World:What infants, children, and other species can teach us. OxfordUniversity Press. ? pages: 25Chudek, M., Heller, S., Birch, S., & Henrich, J. (2012). Prestige-biasedcultural learning: bystander?s differential attention to potential modelsinfluences children?s learning. Evolution and Human Behavior, 33(1),46?56. ? pages: 25Chudek, M., & Henrich, J. (2011). Culture-gene coevolution,norm-psychology and the emergence of human prosociality. Trends incognitive sciences, 15(5), 218?26. ? pages: 14, 29, 41, 44, 138, 139Chudek, M., McNamara, R., Birch, S., Bloom, P., & Henrich, J.(forthcoming). Developmental and Cross-Cultural Evidence for IntuitiveDualism. ,. ? pages: 12, 23Chudek, M., Zhao, W., & Henrich, J. (2013b). Culture-Gene Coevolution,Large-Scale Cooperation and the Shaping of Human Social Psychology.In K. Sterelny, R. Joyce, B. Calcott, & B. Fraser (Eds.) Cooperation andits Evolution, Vol 2: Agents and Mechanisms. MIT Press. ? pages: 15,17, 90, 138Clark, J. D., & Harris, J. W. K. (1985). Fire and its roles in early hominidlifeways. African Archaeological Review, 3(1), 3?27. ? pages: 159175Cribari-Neto, F., & Zeileis, A. (2010). Beta regression in R. Journal ofStatistical Software, 34(2). ? pages: 102, 110, 142Csibra, G., & Gergely, G. (2009). Natural pedagogy. Trends in cognitivesciences, 13(4), 148?53. ? pages: 10, 17, 29Currie, T. E., Greenhill, S. J., Gray, R. D., Hasegawa, T., & Mace, R.(2010). Rise and fall of political complexity in island South-East Asiaand the Pacific. Nature, 467(7317), 801?804. ? pages: 158, 160Cushman, F., Young, L., & Hauser, M. (2006). The role of consciousreasoning and intuition in moral judgment testing three principles ofharm. Psychological science, 17(12), 1082?1089. ? pages: 33, 46, 134,149Daly, M., & Wilson, M. (1988). Homicide. Aldine. ? pages: 28Day, M. H. (1969). Omo human skeletal remains. Nature, 222(5199),1135?8. ? pages: 9Deacon, T. W. (1998). The Symbolic Species: The Co-evolution ofLanguage and the Brain. W. W. Norton & Company. ? pages: 10Dean, L. G., Kendal, R. L., Schapiro, S. J., Thierry, B., & Laland, K. N.(2012). Identification of the social and cognitive processes underlyinghuman cumulative culture. Science, 335(6072), 1114?1118. ? pages: 10Dennett, D. C. (1991). Real patterns. The Journal of Philosophy, 88(1),27?51. ? pages: 4DeScioli, P., Christner, J., & Kurzban, R. (2011). The omission strategy.Psychological Science, 22(4), 442?446. ? pages: 33, 46, 134Doebeli, M., & Hauert, C. (2005). Models of cooperation based on thePrisoner?s Dilemma and the Snowdrift game. Ecology Letters, 8(7),748?766. ? pages: 21, 89Elworty, F. T. (1895). The Evil Eye: The classic account of an ancientsuperstition. Courier Dover Publications. ? pages: 141Endicott, P., Ho, S. Y. W., Metspalu, M., & Stringer, C. (2009).Evaluating the mitochondrial timescale of human evolution. Trends inecology & evolution, 24(9), 515?521. ? pages: 159176Engelmann, D., & Fischbacher, U. (2009). Indirect reciprocity andstrategic reputation building in an experimental helping game. Gamesand Economic Behavior, 67(2), 399?407. ? pages: 96, 140Enquist, M., Eriksson, K., & Ghirlanda, S. (2008). Critical social learning:A solution to Rogers?s paradox of nonadaptive culture. AmericanAnthropologist, 109(4), 727?734. ? pages: 11Evans-Pritchard, E. E. E. E. (1976). Witchcraft, oracles, and magic amongthe Azande.. Clarendon Press. ? pages: 141Fehr, E., & Henrich, J. (2003). Is Strong Reciprocity a Maladaptation. InP. Hammerstein (Ed.) Genetic and Culture Evolution of Cooperation.MIT Press. ? pages: 90Ferrari, S., & Cribari-Neto, F. (2004). Beta regression for modelling ratesand proportions. Journal of Applied Statistics, 31(7), 799?815. ? pages:102, 110, 142Fiske, S. T. (1980). Attention and weight in person perception: Theimpact of negative and extreme behavior. Journal of Personality andSocial Psychology, 38(6), 889. ? pages: 33Fogarty, L., Strimling, P., & Laland, K. N. (2011). The evolution ofteaching. Evolution, 65(10), 2760?2770. ? pages: 26Frankenhuis, W. E. (2010). Did insecure attachment styles evolve for thebenefit of the group? Frontiers in psychology, 1, 172. ? pages: 11G?chter, S., & Herrmann, B. (2009). Reciprocity, culture and humancooperation: previous insights and a new cross-cultural experiment.Philosophical transactions of the Royal Society of London. Series B,Biological sciences, 364(1518), 791?806. ? pages: 97Galef, B. G., & Whiskin, E. E. (2008). ?Conformity?in Norway rats?Animal Behaviour, 75(6), 2035?2039. ? pages: 12Garrigan, D., & Hammer, M. F. (2006). Reconstructing human origins inthe genomic era. Nature reviews. Genetics, 7(9), 669?80. ? pages: 10Geritz, S. A. H., Kisdi, E., Mesze?NA, G., & Metz, J. A. J. (1997).Evolutionarily singular strategies and the adaptive growth andbranching of the evolutionary tree. Evolutionary Ecology, 12(1), 35?57.? pages: 36177Gervais, W. M., Willard, A. K., Norenzayan, A., & Henrich, J. (2011).The cultural transmission of faith: Why innate intuitions are necessary,but insufficient, to explain religious belief. Religion, 41(3), 389?410. ?pages: 14Gidron, D., Koehler, D. J., & Tversky, A. (1993). Implicit quantification ofpersonality traits. Personality and Social Psychology Bulletin, 19, 594.? pages: 168Gignoux, C. R., Henn, B. M., & Mountain, J. L. (2011). Rapid, globaldemographic expansions after the origins of agriculture. Proceedings ofthe National Academy of Sciences, 108(15), 6044?6049. ? pages: 11Gowlett, J. A. J., Harris, J. W. K., Walton, D., & Wood, B. A. (1981).Early archaeological sites, hominid remains and traces of fire fromChesowanja, Kenya. Nature, 294(5837), 125?129. ? pages: 159Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservativesrely on different sets of moral foundations. Journal of personality andsocial psychology, 96(5), 1029. ? pages: 137Greene, J. D., Cushman, F. A., Stewart, L. E., Lowenberg, K., Nystrom,L. E., & Cohen, J. D. (2009). Pushing moral buttons: The interactionbetween personal force and intention in moral judgment. Cognition,111(3), 364?371. ? pages: 137Griffiths, T. L., Sobel, D. M., Tenenbaum, J. B., & Gopnik, A. (2011).Bayes and blickets: Effects of knowledge on causal induction in childrenand adults. Cognitive science, 35(8), 1407?1455. ? pages: 12Gurven, M., Kaplan, H., & Gutierrez, M. (2006). How long does it take tobecome a proficient hunter? Implications for the evolution of extendeddevelopment and long life span. Journal of Human Evolution, 51(5),454?470. ? pages: 20Hamilton, W. (1964). The genetical evolution of social behaviour. I.Journal of Theoretical Biology, 7(1), 1?16. ? pages: 28, 138Hamlin, K. J., Wynn, K., & Bloom, P. (2010). Three-month-olds show anegativity bias in their social evaluations. Developmental science, 13(6),923?929. ? pages: 33Harris, P. (2012). Trusting What You?re Told. Harvard University Press.? pages: 18178Henrich, J. (2004). Demography and cultural evolution: how adaptivecultural processes can produce maladaptive losses: the Tasmanian case.American Antiquity, (pp. 197?214). ? pages: 14Henrich, J. (2009). The evolution of costly displays, cooperation andreligion:: credibility enhancing displays and their implications forcultural evolution. Evolution and Human Behavior, 30(4), 244?260. ?pages: 13, 24, 26, 29, 30, 91, 168Henrich, J., & Boyd, R. (2001). Why people punish defectors. Journal oftheoretical biology, 208(1), 79?89. ? pages: 26, 91Henrich, J., & Boyd, R. (2008). Division of labor, economic specialization,and the evolution of social stratification. Current Anthropology, 49(4),715?724. ? pages: 14Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H.,McElreath, R., Alvard, M., Barr, A., Ensminger, J., & Others (2005). ?Economic man? in cross-cultural perspective: Behavioral experiments in15 small-scale societies. Behavioral and brain sciences, 28(6), 795?814.? pages: 158Henrich, J., Boyd, R., & Richerson, P. J. (2012). The puzzle ofmonogamous marriage. Philosophical Transactions of the Royal SocietyB: Biological Sciences, 367(1589), 657?669. ? pages: 14Henrich, J., & Broesch, J. (2011). On the nature of cultural transmissionnetworks: evidence from Fijian villages for adaptive learning biases.Philosophical Transactions of the Royal Society B: Biological Sciences,366(1567), 1139?1148. ? pages: 17Henrich, J., & Gil-White, F. J. (2001). The evolution of prestige: freelyconferred deference as a mechanism for enhancing the benefits ofcultural transmission. Evolution and Human Behavior, 22(3), 165?196.? pages: 13, 24, 26, 30, 53, 91, 168Henrich, J., Heine, S. J., Norenzayan, A., & Others (2010). The weirdestpeople in the world. Behavioral and Brain Sciences, 33(2-3), 61?83. ?pages: 10, 97, 144, 160Henrich, J., & Henrich, N. (2007). Why humans cooperate: A cultural andevolutionary explanation. Oxford University Press. ? pages: 45179Henrich, J., & Henrich, N. (forthcoming). Fairness without punishment:behavioral experiments in the Yasawa Island, Fiji. In J. Henrich, &J. Ensminger (Eds.) Experimenting with Social Norms: Fairness andPunishment in Cross-Cultural Perspective, (pp. 171?218). ,. ? pages: 43Henrich, J., McElreath, R., Barr, A., Ensminger, J., Barrett, C.,Bolyanatz, A., Cardenas, J. C., Gurven, M., Gwako, E., Henrich, N., &Others (2006). Costly punishment across human societies. Science,312(5781), 1767?1770. ? pages: 10Herrmann, E., Call, J., Hern?ndez-Lloreda, M. V., Hare, B., & Tomasello,M. (2007). Humans have evolved specialized skills of social cognition:the cultural intelligence hypothesis. Science (New York, N.Y.),317(5843), 1360?6. ? pages: 10Herrmann, E., Hern?ndez-Lloreda, M. V., Call, J., Hare, B., & Tomasello,M. (2010). The structure of individual differences in the cognitiveabilities of children and chimpanzees. Psychological science, 21(1),102?110. ? pages: 10Higham, J. P., & Maestripieri, D. (2010). Revolutionary coalitions in malerhesus macaques. Behaviour, 147 , 13(14), 1889?1908. ? pages: 29, 138Hudjashov, G., Kivisild, T., Underhill, P. A., Endicott, P., Sanchez, J. J.,Lin, A. A., Shen, P., Oefner, P., Renfrew, C., Villems, R., & Others(2007). Revealing the prehistoric settlement of Australia by Ychromosome and mtDNA analysis. Proceedings of the National Academyof Sciences, 104(21), 8726?8730. ? pages: 159Inbar, Y., Pizarro, D. A., & Cushman, F. (2012). Benefiting FromMisfortune. Personality and Social Psychology Bulletin, 38(1), 52?62. ?pages: 140Ingram, C. J. E., Mulcare, C. A., Itan, Y., Thomas, M. G., & Swallow,D. M. (2009). Lactose digestion and the evolutionary genetics of lactasepersistence. Human genetics, 124(6), 579?591. ? pages: 158Kalish, M. L., Griffiths, T. L., & Lewandowsky, S. (2007). Iteratedlearning: Intergenerational knowledge transmission reveals inductivebiases. Psychonomic Bulletin & Review, 14(2), 288?294. ? pages: 12Kaplan, H. S., & Robson, A. J. (2002). The emergence of humans: Thecoevolution of intelligence and longevity with intergenerational transfers.180Proceedings of the National Academy of Sciences, 99(15), 10221?10226.? pages: 20Karkanas, P., Shahack-Gross, R., Ayalon, A., Bar-Matthews, M., Barkai,R., Frumkin, A., Gopher, A., & Stiner, M. C. (2007). Evidence forhabitual use of fire at the end of the Lower Paleolithic: Site-formationprocesses at Qesem Cave, Israel. Journal of human evolution, 53(2),197?212. ? pages: 159Kitayama, S., & Uskul, A. K. (2011). Culture, mind, and the brain:Current evidence and future directions. Annual review of psychology, 62,419?449. ? pages: 160Knobe, J. (2003). Intentional action and side effects in ordinary language.Analysis, 63(279), 190?194. ? pages: 33Knobe, J. (2010). Person as scientist, person as moralist. Behavioral andBrain Sciences, 33(4), 315. ? pages: 33, 137Kotrschal, A., Rogell, B., Bundsen, A., Svensson, B., Zajitschek, S.,Br?nnstr?m, I., Immler, S., Maklakov, A. A., & Kolm, N. (2013).Artificial selection on relative brain size in the guppy reveals costs andbenefits of evolving a larger brain. Current Biology. ? pages: 24, 89Kuhn, T. S. (1996). The structure of scientific revolutions, vol. 2.University of Chicago press. ? pages: 1Lachmann, M., & Bergstrom, C. T. (2004). The disadvantage ofcombinatorial communication. Proceedings of the Royal Society ofLondon. Series B: Biological Sciences, 271(1555), 2337?2343. ? pages:20, 29Lakin, J. L., Chartrand, T. L., & Arkin, R. M. (2008). I Am Too Just LikeYou Nonconscious Mimicry as an Automatic Behavioral Response toSocial Exclusion. Psychological Science, 19(8), 816?822. ? pages: 27Lakin, J. L., Jefferis, V. E., Cheng, C. M., & Chartrand, T. L. (2003). Thechameleon effect as social glue: Evidence for the evolutionarysignificance of nonconscious mimicry. Journal of nonverbal behavior,27(3), 145?162. ? pages: 26, 27Laland, K. N. (2004). Social learning strategies. Learning & behavior,32(1), 4?14. ? pages: 10, 11, 13, 30181Laland, K. N., Odling-Smee, J., & Feldman, M. W. (2001). Cultural nicheconstruction and human evolution. Journal of Evolutionary Biology,14(1), 22?33. ? pages: 10, 13Laland, K. N., Odling-Smee, J., & Myles, S. (2010). How culture shapedthe human genome: bringing genetics and the human sciences together.Nature reviews. Genetics, 11(2), 137?48. ? pages: 10, 14, 158Lancaster, J. B., Kaplan, H., Hill, K., & Hurtado, A. M. (2000). Theevolution of life history, intelligence and diet among chimpanzees andhuman foragers. Perspectives in ethology: Evolution, culture, andbehavior, 13, 47?72. ? pages: 20Langergraber, K. E., Mitani, J. C., & Vigilant, L. (2007). The limitedimpact of kinship on cooperation in wild chimpanzees. Proceedings ofthe National Academy of Sciences, 104(19), 7786. ? pages: 29, 138Leighton, J., Bird, G., Orsini, C., & Heyes, C. (2010). Social attitudesmodulate automatic imitation. Journal of Experimental SocialPsychology, 46(6), 905?910. ? pages: 27Leimar, O., & Hammerstein, P. (2001). Evolution of cooperation throughindirect reciprocity. Proceedings. Biological sciences / The RoyalSociety, 268(1468), 745?53. ? pages: 22, 28, 30, 93, 106Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobiledestruction: An example of the interaction between language andmemory. Journal of verbal learning and verbal behavior, 13(5), 585?589.? pages: 25Lyons, D. E., Young, A. G., & Keil, F. C. (2007). The hidden structure ofoverimitation. Proceedings of the National Academy of Sciences,104(50), 19751. ? pages: 25Mace, R., & Jordan, F. M. (2011). Macro-evolutionary studies of culturaldiversity: a review of empirical studies of cultural transmission andcultural adaptation. Philosophical Transactions of the Royal Society B:Biological Sciences, 366(1563), 402?411. ? pages: 158, 160Mathew, S., & Boyd, R. (2011). Punishment sustains large-scalecooperation in prestate warfare. Proceedings of the National Academy ofSciences, 108(28), 11375?11380. ? pages: 95182McDougall, I., Brown, F. H., & Fleagle, J. G. (2005). Stratigraphicplacement and age of modern humans from Kibish, Ethiopia. Nature,433(7027), 733?6. ? pages: 9McElreath, R., Boyd, R., & Richerson, P. J. (2003). Shared norms and theevolution of ethnic markers. Current Anthropology, 44(1), 122?129. ?pages: 14McGuigan, N., Makinson, J., & Whiten, A. (2011). From over-imitation tosuper-copying: Adults imitate causally irrelevant aspects of tool use withhigher fidelity than young children. British Journal of Psychology,102(1), 1?18. ? pages: 25Merrell, K. W., Gueldner, B. a., Ross, S. W., & Isava, D. M. (2008). Howeffective are school bullying intervention programs? A meta-analysis ofintervention research. School Psychology Quarterly, 23(1), 26?42. ?pages: 44Mesoudi, A. (2007). Using the methods of experimental social psychologyto study cultural evolution. Journal of Social, Evolutionary, andCultural Psychology, 1(2), 35?58. ? pages: 138Mesoudi, A., Whiten, A., Laland, K. N., & Others (2006). Towards aunified science of cultural evolution. Behavioral and Brain Sciences,29(4), 329?346. ? pages: 138, 139Mikhail, J. (2007). Universal moral grammar: Theory, evidence and thefuture. Trends in cognitive sciences, 11(4), 143?152. ? pages: 160Milinski, M., & Rockenbach, B. (2012). On the interaction of the stick andthe carrot in social dilemmas. Journal of theoretical biology, 299,139?43. ? pages: 97Milinski, M., Semmann, D., Bakker, T. C. M., & Krambeck, H. J. (2001).Cooperation through indirect reciprocity: image scoring or standingstrategy? Proceedings of the Royal Society of London. Series B:Biological Sciences, 268(1484), 2495?2501. ? pages: 140Mischel, W. (2009). Becoming a cumulative science. APS Observer, 22(1).? pages: 3Mithen, S. (1996). The Prehistory of the Mind: The Cognitive Origins ofArt, Religion and Science. Thames & Hudson. ? pages: 10183Nakahashi, W., Wakano, J. Y., & Henrich, J. (2012). Adaptive SocialLearning Strategies in Temporally and Spatially Varying Environments.Human Nature, (pp. 1?33). ? pages: 13, 30Navarrete, C. D. C., & Fessler, D. M. T. D. (2005). Normative Bias andAdaptive Challenges: A Relational Approach to Coalitional Psychologyand a Critique of Terror Management Theory. Evolutionary Psychology,3, 297?-325. ? pages: 11Nielsen, M., & Tomaselli, K. (2010). Overimitation in Kalahari Bushmanchildren and the origins of human cultural cognition. PsychologicalScience, 21(5), 729?736. ? pages: 25Nielsen, R., Hellmann, I., Hubisz, M., Bustamante, C., & Clark, A. G.(2007). Recent and ongoing selection in the human genome. NatureReviews Genetics, 8(11), 857?868. ? pages: 158Nisbet, R. E. (2003). Geography of Thought. Nicholas Brealey Publishing.? pages: 9Nisbett, R. (2004). The geography of thought: How Asians and Westernersthink differently... and why. Free Press. ? pages: 160Norenzayan, A., & Shariff, A. F. (2008). The origin and evolution ofreligious prosociality. Science (New York, N.Y.), 322(5898), 58?62. ?pages: 14, 28Nowak, M. A., & Sigmund, K. (1998). Evolution of indirect reciprocity byimage scoring. Nature. ? pages: 93, 106, 140Nowak, M. A., & Sigmund, K. (2005). Evolution of indirect reciprocity.Nature, 437(7063), 1291?8. ? pages: 10Nowak, M. A., Tarnita, C. E., & Wilson, E. O. (2010). The evolution ofeusociality. Nature, 466(7310), 1057?1062. ? pages: 14Ohtsuki, H., & Iwasa, Y. (2004). How should we definegoodness??reputation dynamics in indirect reciprocity. Journal ofTheoretical Biology, 231(1), 107?120. ? pages: 94Ohtsuki, H., Iwasa, Y., & Others (2006). The leading eight: social normsthat can maintain cooperation by indirect reciprocity. Journal ofTheoretical Biology, 239(4), 435?444. ? pages: 22, 51184Otto, S. P., & Day, T. (2011). A biologist?s guide to mathematical modelingin ecology and evolution, vol. 13. Princeton University Press. ? pages:81Over, H., & Carpenter, M. (2009). Priming third-party ostracism increasesaffiliative imitation in children. Developmental science, 12(3), F1?-F8.? pages: 27Over, H., Carpenter, M., Spears, R., & Gattis, M. (2013). Childrenselectively trust individuals who have imitated them. SocialDevelopment. ? pages: 27O?Brien, M. J., & Laland, K. N. (2012). Genes, Culture, and Agriculture:An Example of Human Niche Construction. Current Anthropology,53(4), pp. 434?470. ? pages: 14Panchanathan, K., & Boyd, R. (2004). Indirect reciprocity can stabilizecooperation without the second-order free rider problem. Nature,432(7016), 499?502. ? pages: 22, 28, 29, 57, 94, 95, 137, 139Panchanathan, K., Boyd, R., & Others (2003). A tale of two defectors: theimportance of standing for evolution of indirect reciprocity. Journal ofTheoretical Biology, 224(1), 115?126. ? pages: 22, 93, 140Park, J. H., Schaller, M., & Van Vugt, M. (2008). Psychology of humankin recognition: Heuristic cues, erroneous inferences, and theirimplications. Review of General Psychology, 12(3), 215. ? pages: 28Patterson, N., Richter, D. J., Gnerre, S., Lander, E. S., & Reich, D.(2006). Genetic evidence for complex speciation of humans andchimpanzees. Nature, 441(7097), 1103?8. ? pages: 9Peng, Y., Shi, H., Qi, X., Xiao, C., Zhong, H., Ma, R. Z., & Su, B. (2010).The ADH1B Arg47His polymorphism in East Asian populations andexpansion of rice domestication in history. BMC evolutionary biology,10(1), 15. ? pages: 158Perry, S., & Manson, J. H. (2008). Manipulative monkeys: the capuchins ofLomas Barbudal. Harvard Univ Pr. ? pages: 29, 138Pinker, S. (2010). The cognitive niche: Coevolution of intelligence,sociality, and language. Proceedings of the National Academy ofSciences, 107(Supplement 2), 8993?8999. ? pages: 10, 23, 24, 89185Powell, A., Shennan, S., & Thomas, M. G. (2009). Late Pleistocenedemography and the appearance of modern human behavior. Science,324(5932), 1298?1301. ? pages: 167Rakoczy, H., Warneken, F., & Tomasello, M. (2009). Young children?sselective learning of rule games from reliable and unreliable models.Cognitive Development, 24(1), 61?69. ? pages: 25Rand, D. G. (2012). The promise of Mechanical Turk: how online labormarkets can help theorists run behavioral experiments. Journal oftheoretical biology, 299, 172?9. ? pages: 109, 131, 146Rankin, D. J., & Eggimann, F. (2009). The evolution of judgement bias inindirect reciprocity. Proceedings. Biological sciences / The RoyalSociety, 276(1660), 1339?45. ? pages: 93Rendell, L., Fogarty, L., & Laland, K. N. (2010). Rogers? paradox recastand resolved: population structure and the evolution of social learningstrategies. Evolution, 64(2), 534?548. ? pages: 11Rendell, L., Fogarty, L., & Laland, K. N. (2011). Runaway cultural nicheconstruction. Philosophical Transactions of the Royal Society B:Biological Sciences, 366(1566), 823?835. ? pages: 14Richard, F. D., Bond Jr, C. F., & Stokes-Zoota, J. J. (2003). One HundredYears of Social Psychology Quantitatively Described. Review of GeneralPsychology, 7(4), 331. ? pages: 161Richerson, P. J., & Boyd, R. (2004). Not By Genes Alone: How CultureTransformed Human Evolution. University Of Chicago Press. ? pages:10, 13, 28, 139Ritov, I., & Baron, J. (1999). Protected values and omission bias.Organizational Behavior and Human Decision Processes, 79(2), 79?94.? pages: 33, 46, 134Rodkin, P. C., & Berger, C. (2008). Who bullies whom? Social statusasymmetries by victim gender. International Journal of BehavioralDevelopment, 32(6), 473?485. ? pages: 44Roebroeks, W., & Villa, P. (2011). On the earliest evidence for habitualuse of fire in Europe. Proceedings of the National Academy of Sciences,108(13), 5209?5214. ? pages: 159186Rogers, A. R. (1988). Does biology constrain culture? AmericanAnthropologist, 90(4), 819?831. ? pages: 11Rozin, P., & Royzman, E. B. (2001). Negativity bias, negativitydominance, and contagion. Personality and social psychology review,5(4), 296?320. ? pages: 32, 33, 46, 134, 157Seinen, I., & Schram, A. (2006). Social status and group norms: Indirectreciprocity in a repeated helping experiment. European EconomicReview, 50(3), 581?602. ? pages: 96Senghas, A., Kita, S., & ?zy?rek, A. (2004). Children creating coreproperties of language: Evidence from an emerging sign language inNicaragua. Science, 305(5691), 1779?1782. ? pages: 24Seyfarth, R. M., Cheney, D. L., & Marler, P. (1980). Monkey responses tothree different alarm calls: evidence of predator classification andsemantic communication. Science, 210(4471), 801. ? pages: 20Seyfarth, R. M., Silk, J. B., & Cheney, D. L. (2012). Variation inpersonality and fitness in wild female baboons . Proceedings of theNational Academy of Sciences, 109(42), 16980?16985. ? pages: 138Sigmund, K. (2012). Moral assessment in indirect reciprocity. Journal oftheoretical biology, 299, 25?30. ? pages: 22Sigmund, K., De Silva, H., Traulsen, A., & Hauert, C. (2010). Sociallearning promotes institutions for governing the commons. Nature,466(7308), 861?863. ? pages: 21, 28, 95, 137, 139Silk, J. B. (2002). Using the?F?-word in primatology. Behaviour, (pp.421?446). ? pages: 29, 138Simpson, B., & Willer, R. (2008). Altruism and Indirect Reciprocity: TheInteraction of Person and Situation in Prosocial Behavior. SocialPsychology Quarterly, 71(1), 37?52. ? pages: 96Skowronski, J. J., & Carlston, D. E. (1987). Social judgment and socialmemory: The role of cue diagnosticity in negativity, positivity, andextremity biases. Journal of Personality and Social Psychology, 52(4),689?699. ? pages: 33, 168Smith, J. M. (1964). Group selection and kin selection. Nature, 201,1145?1147. ? pages: 28, 138187Smith, J. M., & Szathm?ry, E. (1997). The Major Transitions inEvolution. Oxford University Press. ? pages: 20Smithson, M., & Verkuilen, J. (2006). A better lemon squeezer?Maximum-likelihood regression with beta-distributed dependentvariables. Psychological Methods, 11(1), 54. ? pages: 102, 110, 142Sommerfeld, R. D., Krambeck, H.-J., & Milinski, M. (2008). Multiplegossip statements and their effect on reputation and trustworthiness.Proceedings of the Royal Society B: Biological Sciences6, 275(1650),2529?36. ? pages: 96Sommerfeld, R. D., Krambeck, H.-J., Semmann, D., & Milinski, M. (2007).Gossip as an alternative for direct observation in games of indirectreciprocity. Proceedings of the National Academy of Sciences of theUnited States of America, 104(44), 17435?40. ? pages: 96Spelke, E. S., & Kinzler, K. D. (2006). Core knowledge. Developmentalscience, 10(1), 89?96. ? pages: 12Spranca, M., Minsk, E., & Baron, J. (1991). Omission and commission injudgment and choice. Journal of Experimental Social Psychology, 27(1),76?105. ? pages: 33, 46, 134, 149Stenberg, G. (2009). Selectivity in infant social referencing. Infancy, 14(4),457?473. ? pages: 25Sterelny, K. (2012). The evolved apprentice: how evolution made humansunique. MIT Press. ? pages: 10, 26Stewart-Williams, S. (2007). Altruism among kin vs. nonkin: Effects ofcost of help and reciprocal exchange. Evolution and Human Behavior,28(3), 193?198. ? pages: 28Suzuki, S., & Akiyama, E. (2007a). Evolution of indirect reciprocity ingroups of various sizes and comparison with direct reciprocity. Journalof theoretical biology, 245(3), 539?52. ? pages: 94Suzuki, S., & Akiyama, E. (2007b). Three-person game facilitates indirectreciprocity under image scoring. Journal of theoretical biology, 249(1),93?100. ? pages: 94Tomasello, M. (1999). The Cultural Origins of Human Cognition. HarvardUniversity Press. ? pages: 10188Top?l, J., Gergely, G., Miklosi, A., Erdohegyi, A., & Csibra, G. (2008).Infants? Perseverative Search Errors Are Induced by PragmaticMisinterpretation. Science, 321(5897), 1831?1834. ? pages: 25Trivers, R. L. (1971). The evolution of reciprocal altruism. Quarterlyreview of biology, (pp. 35?57). ? pages: 21, 89, 138Uchida, S., & Sigmund, K. (2010). The competition of assessment rules forindirect reciprocity. Journal of theoretical biology, 263(1), 13?19. ?pages: 94Ule, A., Schram, A., Riedl, A., & Cason, T. N. (2009). Indirectpunishment and generosity toward strangers. Science (New York, N.Y.),326(5960), 1701?4. ? pages: 96, 97van Veelen, M., Garc?a, J., Rand, D. G., & Nowak, M. A. (2012). Directreciprocity in structured populations. Proceedings of the NationalAcademy of Sciences, 109(25), 9929?9934. ? pages: 21, 89, 167Walker, R., Gurven, M., Hill, K., Migliano, A., Chagnon, N., De Souza, R.,Djurovic, G., Hames, R., Hurtado, A. M., Kaplan, H., & Others (2006).Growth rates and life histories in twenty-two small-scale societies.American Journal of Human Biology, 18(3), 295?311. ? pages: 20Walker, R., Hill, K., Kaplan, H., & McMillan, G. (2002). Age-dependencyin hunting ability among the Ache of Eastern Paraguay. Journal ofHuman Evolution, 42(6), 639?657. ? pages: 20Watts, D. P. (2002). Reciprocity and interchange in the social relationshipsof wild male chimpanzees. Behaviour, (pp. 343?370). ? pages: 30, 138Waxman, D., & Gavrilets, S. (2005). Twenty Questions on AdaptiveDynamics. Journal of Evolutionary Biology, 18(5), 1139?1154. ? pages:36, 77Wedekind, C. (2000). Cooperation Through Image Scoring in Humans.Science, 288(5467), 850?852. ? pages: 96Wedekind, C., & Braithwaite, V. a. (2002). The long-term benefits ofhuman generosity in indirect reciprocity. Current biology : CB, 12(12),1012?5. ? pages: 96West, S. A., El Mouden, C., & Gardner, A. (2011). Sixteen commonmisconceptions about the evolution of cooperation in humans. Evolutionand Human Behavior, 32(4), 231?262. ? pages: 14189Whitehead, H., Rendell, L., Osborne, R. W., & W?rsig, B. (2004). Cultureand conservation of non-humans with reference to whales and dolphins:review and new directions. Biological Conservation, 120(3), 427?437. ?pages: 11Whiten, A., Goodall, J., McGrew, W. C., Nishida, T., Reynolds, V.,Sugiyama, Y., Tutin, C. E. G., Wrangham, R. W., & Boesch, C. (1999).Cultures in chimpanzees. Nature, 399(6737), 682?685. ? pages: 11Whiten, A., McGuigan, N., Marshall-Pescini, S., & Hopper, L. M. (2009).Emulation, imitation, over-imitation and the scope of culture for childand chimpanzee. Philosophical Transactions of the Royal Society B:Biological Sciences, 364(1528), 2417?2428. ? pages: 25Wilson, E. O. (2012). The Social Conquest of Earth. Liveright. ? pages:10Wrangham, R. (2009). Catching fire: how cooking made us human. Basicbooks. ? pages: 159Yamamichi, M., Gojobori, J., & Innan, H. (2012). An autosomal analysisgives no genetic evidence for complex speciation of humans andchimpanzees. Molecular biology and evolution, 29(1), 145?156. ? pages:9Yeung, S., & Griffiths, T. L. (2011). Estimating human priors on causalstrength. Proceedings of the 33rd Annual Conference of the CognitiveScience Society. ? pages: 12190


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items