the causal effect of cultural identity on...
TRANSCRIPT
The Causal Effect of Cultural Identity onCooperation
Jeffrey V. Butler and Dietmar Fehr∗
July 31, 2018
Abstract
In economics, a recent consensus around a definition of cultureas shared beliefs and values among people groups has led to rapidgrowth in the cultural economics literature. Culture’s impact on non-kin cooperation has been singled out as being particularly importanttheoretically. Empirical causal evidence running from culture (beliefsand values) to behavior and outcomes remains relatively scant. In thispaper we adopt a viewpoint implied by social identity, a contempo-raneous and also rapidly growing economics literature, that culture isone aspect of an individual’s multi-faceted self-concept. We then use“priming,” a technique successfully employed in the identity litera-ture, to reveal the causal impact of culture on behavior. In two exper-iments involving first- and second-generation Chinese immigrants ata large US public university, we prime participants’ American or Chi-nese cultural backgrounds and observe the impact this has on coop-eration with strangers. In both experiments, culture has a substantialimpact on cooperation. Moreover, comparing behavior across experi-ments, our results suggest that culture’s effect on beliefs formation isas important as its effect on cooperative preferences.
∗Butler: University of California, Merced, [email protected]. Fehr: University ofHeidelberg, [email protected]. We thank participants at the ESA Meet-ings in Heidelberg 2015 and in Tucson 2016, the Arne Ryde Workshop 2016 in Lund, theSPI Conference 2017 in Chicago, BABEEW 2018 at the University of San Francisco, and theERINN Workshop at Oxford University 2018 as well as seminar participants at Florida StateUniversity, Georgia State University and the University of California, Merced. We thankYannick Reichlin for research assistance. Dietmar Fehr gratefully acknowledges financialsupport from the Berlin Social Science Center (WZB). This study was approved by the In-stitutional Review Board (IRB) at UC Berkeley (2014-03-6094). The authors declare that theyhave no relevant or material financial interests that relate to the research described in thispaper.
1 Introduction
An emerging consensus on a concrete definition of culture as beliefs andvalues shared among people and groups (Guiso, Sapienza and Zingales,2006; Bisin and Verdier, 2008; Tabellini, 2010a; Fernandez, 2011)1 has ledto rapid growth in the economics literature on culture. While the ques-tion of whether culture matters for economic outcomes is seemingly settled(Fernandez and Fogli, 2006; Tabellini, 2010b; Alesina, Giuliano and Nunn,2013; Algan and Cahuc, 2013; Alesina and Giuliano, 2015; Lowes et al.,2017), the channels through which culture affects behavior and outcomes,and their relative importance, are important open questions.
In this paper, we study the impact that culture has on cooperation. Put-nam identifies the propensity to cooperate as one of a handful of key traitsdefining social capital (Alesina and Giuliano, 2015, pp. 898-899), whicheconomists have, in turn, argued facilitates the well functioning of economiesand societies (inter alia, Knack and Keefer, 1997; Algan and Cahuc, 2013).We focus on two distinct channels through which culture may affect coop-eration, as suggested by the definition above. Culture may affect behaviordirectly through preferences, by instilling cooperation as a virtue (Tabellini,2008) or by coloring beliefs about how others will behave and what otherswill expect.By affecting social perception, culture may indirectly induce co-operation in various ways, including: through a direct preference for con-formity; through an aversion to disappointing others; or by serving as a co-ordinating device in a repeated-game setting (Asch, 1955; Bernheim, 1994;Bicchieri, 2006; Charness and Dufwenberg, 2006; Battigalli and Dufwen-berg, 2007; Butler et al., 2016).
Our goal of investigating how culture causes cooperation obviously re-quires exogenous variation in culture. To accomplish this, we combine in-sights from various literatures. First, we depart from much of the litera-ture on culture in economics and adopt the viewpoint that culture is oneaspect of an individual’s multi-faceted self-concept, or identity. Buildingon decades of research in sociology and psychology beginning with Tajfelet al. (1971), Akerlof and Kranton introduced the concept of identity into
1For example, Guiso, Sapienza and Zingales (2006, p. 23) define culture as “[...] thosecustomary beliefs and values that ethnic, religious, and social groups transmit fairly un-changed from generation to generation.” Similarly, Bisin and Verdier (2008) refer to cultureas “preferences, beliefs, and norms that govern human behavior [...]” while Fernandez(2011, p.482) provides a working definition of culture as a “[...] distribution of social prefer-ences and beliefs.” Tabellini (2010a) measures culture by aggregating “[...] specific indica-tors of values and beliefs [...].”
1
economics (Akerlof and Kranton, 2000, 2002, 2005, 2010). They conceptu-alize identity as a set of social categories to which prescriptions about theideal traits and behaviors of category members are attached. For behav-ioral content, Akerlof and Kranton proffer a model of preferences in whichstandard economic utility is augmented by identity utility: a preference forbelonging to a particular social category and for living up to the prescrip-tions of that social category. Because relevant social categories may varywith context, incorporating identity utility provides “...a theory of decisionmaking where social context matters” (Akerlof and Kranton, 2010, p. 6).
Although there is an intuitive connection between identity’s category-specific prescriptions and the beliefs and values that define culture, to datethe bodies of economic research on culture and on identity remain largelyseparate, with only a few prominent exceptions including Besley and Pers-son (2018). Connecting these two literatures may benefit both. In particular,the role of cultural identity in coloring beliefs about others is central to theculture literature but absent from the identity economics literature.2
Second, we build on a host of research documenting differences in West-ern and Eastern Cultures (see e.g., Nisbett et al., 2001; Nisbett, 2010). Mostof these differences can be tied to the individualism-collectivism distinc-tion (e.g., Triandis, 2001; Oyserman, Coon and Kemmelmeier, 2002; Heine,2015). Individualism is a cultural disposition that puts emphasis on inde-pendence, freedom of choice, and prioritizes self over social groups. Col-lectivism views individuals primarily as parts of a greater whole and thusemphasizes interdependence, conformity and the common good. Variationalong this cultural dimension, for example, has implications on long-rungrowth (Gorodnichenko and Roland, 2011, 2017) and should intuitively bean important factor enhancing cooperation. Accordingly, we focus on twowidely studied nationalities – Chinese and American – that are at oppositeends of the collectivism vs. individualism spectrum.3
2The literature on the cultural transmission of trust can be seen in this vein, as trust ispartially about beliefs about others’ trustworthiness (Sapienza, Toldra-Simats and Zingales,2013; Butler et al., 2016; Butler, Giuliano and Guiso, 2015, 2016). More generally, Besley andPersson (2018, p. 4) assert “...the most common approach among economists who study[corporate] cultures focuses on beliefs.” At the same time, Akerlof and Kranton’s model ofidentity (Akerlof and Kranton, 2000, 2002, 2005, 2010) takes place in a complete and per-fect information setting with no role for beliefs formation, while Benabou and Tirole (2003,2006a,b) model identity as beliefs about oneself and leave outside the model how these be-liefs might color beliefs about others. This oversight is not characteristic of the social psy-chological literature on identity where there is a long history of asking how categorizationaffects social perception (see the discussion in Butler, 2018).
3On the 0 to 100 Hofstede scale of individualism, with 100 indicating a fully individualist
2
Adopting the view that culture is an identity allows us to reveal thecausal impact of culture on preferences, beliefs and behavior through theexperimental technique of “priming”: the use of subtle situational cuesto make a specific social category temporarily salient amd thereby inducebehavior prescribed by the salient social category (Benjamin et al., 2010,p. 1914). Priming has been successfully used in several studies in the iden-tity economics literature (e.g. Benjamin et al., 2010; Cohn, Fehr and Marechal,2014; Cohn, Marechal and Noll, 2015; Chen et al., 2014). In two separateexperiments we recruit individuals simultaneously belonging to both Chi-nese and American cultures, “prime” one of these cultural identities, andexamine how this affects cooperation.
In our first experiment, we prime cultural identity and have partici-pants play an anonymous one-shot Prisoners’ Dilemma game. Behaviorhere should reflect the effect of culture on cooperation through both the be-liefs channel and through the preferences channel – the total effect of cul-ture on cooperation. In our second experiment, we attempt to shut downthe beliefs channel by transmitting credible concrete information about thelikelihood that others cooperated in an identical situation. We conjecturethat this information will weaken or eliminate culture’s impact on the be-liefs formation process while leaving its effect on preferences intact. Com-paring behavior across these two experiments should therefore reveal therelative strength of culture’s capacity to color beliefs.
Much of the recent empirical research on the role of culture in economicoutcomes is based on the “epidemiological approach” that exploits culturalvariation among immigrants holding the economic and institutional settingfixed. More precisely, its logic is that inherited beliefs and values vary ex-ogenously with respect to second-generation immigrants’ experiences intheir countries of residence, so that correlation between country-of-originculture and country-of-residence outcomes admits a causal interpretation(e.g., Fernandez and Fogli, 2006; Giuliano, 2007; Fernandez and Fogli, 2009;Algan and Cahuc, 2010; Luttmer and Singhal, 2011). While this approachrules out important confounds such as reverse causation, concerns aboutother sources of endogeneity, such as omitted variable bias remain (see
culture and 0 indicating a fully collectivist culture, China scores 20 while the US scores91 (Hofstede, Hofstede and Minkov, 2010). At the same time, recent research has arguedthat there is substantial variation in collectivism within China due to historical agriculturalpractices – rice cultivation vs. wheat cultivation (Talhelm et al., 2014). We address thisconcern using information about whether our participants parents were born north of theYangtze river in China, where historical reliance on wheat cultivation may have given riseto a more individualistic culture.
3
Fernandez (2008, 2011) and Alesina and Giuliano (2015) for a more thor-ough discussion). By sidestepping many of the endogeneity issues plagu-ing the existing cultural economics research, our experimental methodol-ogy allows us to examine the causal impact of cultural identity on coopera-tion precisely enough to plausibly identify its effects on separate channels –beliefs and preferences. Thus, our approach complements the existing em-pirical approaches, which feature higher external validity at the expense ofmeasurement precision.
As a preview of our results, in our first experiment we find that prim-ing American identity substantially increases cooperation with anonymousstrangers in the Prisoners’ Dilemma game. US-primed participants areabout 25 percentage points more likely to cooperate than their counterpartsin Chinese-prime treatments, who cooperate only about 40% of the time.
This result is at first glance surprising in light of recent research. Ina study closely related to our first experiment, LeBoeuf, Shafir and Bayuk(2010) find the opposite pattern in a hypothetical Prisoners’ Dilemma game.Other research indicates that traits related to cooperation such as positivereciprocity, altruism, and trust are consistently more pronounced in Chinathan in the US (Falk et al., 2017) and suggests that collectivist culturescan be traced back to “collaborative” irrigation practices in (pre-industrial)agriculture (e.g., Buggle, 2017; Talhelm et al., 2014). However, this evi-dence typically considers either hypothetical decisions or situations withcomplete information where people know what to expect from others. Ifwe move instead to situations where uncertainty about others’ behaviorlooms larger, one may observe, as we do, less cooperation in collectivistcultures (see also Alesina and Giuliano, 2015; Enke, 2017; Gaechter, Schulzand Thoeni, 2017).4
In our second experiment – which ideally reflects culture’s impact onpreferences only – we find the exact opposite: US-primed participant coop-erate only 41% of the time, while Chinese-primed participants are substan-tially and significantly more likely to cooperate (59% of them do).
Taken together, our study complements the findings from previous cross-cultural research suggesting that small-scale societies with greater marketintegration are more cooperative and fair in anonymous transactions (Hen-rich et al., 2001, 2005, 2010). This finding echoes a common observationthat collectivist cultures tend to have a small radius of trust and coopera-
4Other recent research has shown that China’s One-Child Policy (OCP) may have ledto less cooperation within society. Cameron et al. (2013) show that subjects born after theintroduction of the OCP are less trusting, less trustworthy, more risk-averse, and more pes-simistic – traits that are related to cooperation – than subjects born before the OCP.
4
tion, i.e., that individuals mainly interact in closed circles where they canexpect preferential treatment from others, while the opposite is true for in-dividualistic cultures. These are characterized as more open to interactionswith strangers and at the same time by more uncertain expectations aboutothers’ behavior (e.g., Yamagishi, Cook and Watabe, 1998; Alesina and Giu-liano, 2015; Enke, 2017).5 In Experiment 1 this uncertainty features moreprominently than in Experiment 2 where individuals received informationabout previous cooperation rates in otherwise similar situations. As such,priming a “collectivist” cultural identity in Experiment 2 may be reminis-cent of a “collectivist” frame that promotes cooperation. On the other hand,in the context of Experiment 1 interacting with a stranger is something onewants to avoid under a “collectivist” prime and thus results in low coopera-tion, while such interactions are common under an “individualistic” primeleading to high cooperation.
More generally, the implication of our findings is that culture’s impacton cooperation operates as, if not more, strongly through its effects on so-cial perception as through its direct effects on preferences. Indeed the ef-fects on social perception are so strong that weakening or shutting themdown can reverse culture’s overall impact on cooperation.
The remainder of the paper proceeds as follows. First, we outline aframework through which culture’s impact on beliefs may impact behav-ior. Then we describe in detail our first experiment, followed by results.Next, we detail our second experiment and results. In the final section wesummarize our findings across both experiments and provide concludingremarks.
2 Framework
To explain cooperation in a (one-shot) Prisoners’ Dilemma (PD) game re-quires assuming that material payoffs are not the only component of utility– otherwise, cooperation is a dominated action. In an Akerlof and Kranton-style identity utility model, overall utility is essentially a weighted aver-age of economic utility stemming from material incentives and identityutility. Individuals lose identity utility from failing to live up to identity-contingent prescriptions concerning traits and behaviors. In the PD, if we
5Relatedly, Hajikhameneh and Kimbrough (2018) provide evidence from a lab experi-ment that more “collectivistically-thinking” subjects are less likely quit existing relation-ships to initiate more profitable trades with strangers than “individualistic-thinking” sub-jects.
5
think of cooperation as prescribed by a collectivist identity but not an in-dividualist identity, then we have straightforward predictions about howcultural identity should matter: a collectivist (individualist) culture should,ceterus paribus, increase (decrease) cooperation.
An important caveat is that Akerlof and Kranton develop their model inthe context of complete and perfect information. When outcomes involvestrategic uncertainty, as in the PD, or incomplete information about others’types, overall expected utility may additionally depend on beliefs. Mostobviously, since Social Identity Theory, beginning with Tajfel et al. (1971),is a theory of group-contingent preferences, uncertainty about what groupsothers belong to should translate into uncertainty about the utility conse-quences of particular actions. While this observation gives scope for beliefsto matter, since Akerlof and Kranton are silent on how beliefs will matter,we must look outside of their model for predictions.
Perhaps the simplest channel for beliefs about others to matter in thePD is through a direct preference for conformity. If individuals prefer toconform to, e.g., the behavior of the majority, then a belief that the majoritywill choose to cooperate – for whatever reason, including formal identityutility considerations – may induce an individual to cooperate as well.
Another possible mechanism is given by Bicchieri (2006), who providesa useful framework illustrating how the prescriptions attached to particularidentities may affect PD cooperation through beliefs. In her view, (psychic)utility costs from violating others’ “expectations” may turn the PD into acoordination game.6 If there is uncertainty about what others expect, andexpectations are summarized as a player’s type, the overall situation can berepresented by a Bayesian game such as that depicted in Figure 1 (adaptedfrom Bicchieri (2006, p.27)).7
In Figure 1 the letters B, S, T, and W refer to the Best, Second Best, ThirdBest and Worst outcomes in terms of utility. With probability 1− p an indi-vidual is matched with a co-player who expects cooperation and the utilityconsequences from disappointing the co-player are large enough to makecooperation undominated. This could be, for example, the case if the co-
6In her formulation, expectations are not literally Player B’s mathematical beliefs aboutplayer A’s behavior, which would give rise to a psychological game (Geanakoplos, Pearceand Stacchetti, 1989; Battigalli and Dufwenberg, 2007; Bernheim, 1994) since then A’s beliefsabout B’s beliefs would affect utility. Instead, “expectations” are shorthand for the norms,social or moral (Schram and Charness, 2015), a player subscribes to.
7In the more complex psychological game case where “expectations” are taken to mean,e.g., the row player’s (second-order) beliefs about the column player’s (first-order) beliefsabout the row player’s behavior, a Bayesian game representation may still be empiricallyvalid (see the discussion in Butler et al. (2016)).
6
Figure 1: Bayesian Game, adapted from Bicchieri (2006)
Nature
p 1− p
C DC B,B W,TD T,W S,S
Coordination Game
C DC S,S W,BD B,W T,T
Prisoners’ Dilemma
1
player believes that both share an identity that prescribes cooperation. Ifthis describes both players’ beliefs and preferences, then the players areengaged in a coordination game. With the remaining probability, p, the co-player does not expect cooperation.8 If cooperation is not expected, andidentity utility is itself not strong enough to offset the material gains fromdefecting, then the players are engaged in a PD. Clearly, in this Bayesiangame, whether cooperation obtains depends on beliefs about the fractionof players in the population who expect cooperation, 1− p.
In the context of our experiments, information about the likelihood ofcooperation in previous identical situations may serve as a proxy for thispopulation parameter. We remain agnostic on the precise model relatingbeliefs to behavior, noting only that many models imply a positive rela-tionship between beliefs about the likelihood others will cooperate (or willexpect cooperation) and an individual’s propensity to cooperate.
8This could be the case if the co-player does not believe both players belong to the sameidentity group and intergroup cooperation is not prescribed; it could also be the case thatthe individual believes the co-player belongs to a group that does not prescribe cooperation.
7
3 Experiment 1
3.1 Experimental design and procedures
Our first experiment consisted primarily of two phases: an identity prim-ing phase and a game-playing phase. In the identity priming phase par-ticipants answered a short questionnaire, which began with filler questionsabout demographics and student life. The last three questions of each ques-tionnaire constituted our prime, which varied across treatments. About40 percent of participants were randomly assigned questions making theirChinese identity temporarily salient (e.g. “what is your favorite Chineseholiday?”), while another 40 percent received questions making their USidentity salient (e.g. “what is your favorite US holiday?”). The remaining20 percent served as a control group and answered questions unrelated totheir Chinese or US identity (e.g. “What is your favorite place on cam-pus?”).9 See the supplementary material for the questionnaire and the fullset of instructions.
The game-playing phase was identical across all three treatments andconsisted of several experimental games and individual decisions. Imme-diately after the identity prime, participants played a one-shot Prisoner’sDilemma (PD) in randomly assigned anonymous pairs with payoff param-eters (indicating dollar amounts) B = 12, S = 11, T = 5, and W = 0(see Figure S1 in the supplementary material). Note that the symmetryof the PD permits use of the direct response method, making it particu-larly appropriate when investigating decisions potentially involving a sub-stantial non-deliberative component, such as cooperation. We familiarizedstudents with, and quizzed students on, the bi-matrix notation used to de-scribe the PD at the very beginning of each session – well before the prim-ing instrument. This allowed us to implement the PD immediately follow-ing the identity prime, when the effects were most likely to be detectable.
9Our priming instrument follows closely LeBoeuf, Shafir and Bayuk (2010), but isweaker in some respects and stronger in others. In particular, we take the two US-primeand Chinese-prime questions mentioned verbatim from their study, but ask them both inEnglish whereas in their study the Chinese-prime materials were presented in Chinese. Inthis sense, our priming intrument may be weaker. However, we add free-form responsequestions to our priming instrument which they do not, which may make our instrumentstronger relative to theirs. The primary benefit of writing all materials in English is that itdoes not preclude Chinese participants who are not fluent in Chinese. It is conceivable thatnot all second- and later-generation Chinese immigrants can speak or read Chinese. Beingstudents at an American university, however, all potential participants should have beenfamiliar with English.
8
Following the PD, we randomly and anonymously re-paired partici-pants to play a sequential version of the PD (SPD). Because we were partic-ularly interested in participants’ decisions as a second mover, we presentedthe game in its normal form (see Figure S2 in the supplementary material)and participants played both roles – first as a row player (second mover)and then as a column player (first mover).10
Next, participants were given three dishonesty opportunities. For eachof these opportunities, participants rolled a 10-sided die in private and re-ported the result, which determined their own earnings. Across each ofthese three opportunities we varied who else the die roll affected, pittingown earnings against: i) nobody; ii) another randomly chosen participantin the same session; iii) a well-known charity. The mapping between earn-ings consequences and the number on which the die landed was random-ized across individuals in order to minimize the influence of unintentionalcheating (e.g., confusing a “6” for a “9”).
After the dishonesty opportunities, we elicited participants’ risk atti-tudes using a multiple-price list procedure. The procedure included 18 de-cisions between a certain payment of $15 and a binary lottery paying $30with probability x and $0 with probability (1− x), where x increased from0.25 in increments of 0.03.11
Finally, participants answered a short questionnaire including socio-economic demographics, hypothetical measures of risk and time prefer-ences and measures for cognitive ability. Additionally, we asked subjectsabout their understanding of the experimental tasks, about the reasoningbehind their decisions in the experiment as well as about the purpose ofthe experiment.
Before beginning the experiment, participants were instructed that therewould be several “decision situations” during the experiment and that eachseparate decision situation would be clearly labeled, but were given noother information about what these decision situations would entail. Im-portantly, they were instructed that only one of these situations would berandomly chosen by the computer to determine their entire earnings forthe experiment.12
10Using the normal-form version of the SPD was also suggested by the training our par-ticipants received on bi-matrix notation.
11For a recent overview of risk elicitation methods commonly used in economics, in-cluding the multiple-price list procedure we use which dates to at least Binswanger (1981),see Charness, Gneezy and Imas (2013).
12This is a commonly used procedure which theoretically ameliorates spillovers betweengames, such as using behavior in one decision to hedge against payoff uncertainty from
9
Experiment 1 was implemented using Z-tree (Fischbacher, 2007) in theXlab at the University of California, Berkeley. Because our hypotheses relyon the assumption that participants potentially identify with both a Chi-nese and US cultural identity, we aimed to recruit only students who origi-nated from a Chinese country (China, Taiwan, Hong Kong, and Singapore).Due to restrictions of the recruiting software and limited data availabilityscreening was done by common Chinese last names and the list of eligiblestudents was then hand selected from the entire subject pool. This ex-antescreening on nationality was not perfect and, consequently, we exclude ex-post from our analysis any participant who did not have at least one parentborn in a Chinese country, yielding a total of 102 subjects.13
The sessions took place from the fall of 2014 through the spring semesterof 2015. All participants in a session participated in the same treatment: 39participants in the US-prime treatment, 42 in the Chinese-prime treatmentand 21 who received the Neutral prime.Sessions lasted about 45 minutesand participants earned on average $16.
4 Results: experiment 1
4.1 Main result
We are primarily interested in how cultural identity affects cooperation asmeasured by behavior in the PD. Overall, we find higher cooperation ratesin the US-prime treatments (66.6%) than in the Chinese-prime treatments(42.9%), which is similar to cooperation in the Neutral-prime treatment(42.9%). The difference in cooperation rates between the US-prime andChinese-prime treatments is statistically significant at conventional levels(p = 0.04; two-tailed permutation test).14 Cooperation rates are summa-
other decisions.13Note that in the Chinese-prime treatment we inserted a filtering question before the
priming survey asking where the participants’ family originated from. Those who indicateda non-chinese country subsequently received the Neutral-prime survey to not confuse themwith Chinese-specific follow-up questions from the Chinese-prime survey (e.g., “what isyour favorite Chinese holiday?”). These participants are among the excluded participants.In total, we recruited 124 subjects (50 each in the US-prime and Chinese-prime treatmentand 24 in the Neutral-prime treatment) and excluded 22 subjects from the data analysis.This does not adversely affect sample sizes in the three treatments (Fisher’s exact test, p =0.65).
14The effect size is also substantial according to commonly-used guidelines (Hedges’ g =0.447). For example, the US Department of Education considers an effect size, or Hedges’g, larger than 0.25 to be “substantively important”(Institute of Education Sciences, 2014,
10
rized graphically in Figure 2.
Figure 2: Cooperation Rates in PD, Experiment 1
0.1
.2.3
.4.5
.6.7
.8.9
1
US Chinese NeutralPrime
Average Cooperation Rate in PD
It is possible that other variables confound this results. For example,demographics may not be exactly balanced across treatments so that whatappears to be a priming effect may simply reflect, e.g., a gender differencein cooperation. In the supplementary material, we provide balance checksof the three treatments to rule out systematic differences in an important,albeit limited, set of demographic variables. We run probit regressions ofthe probability of being randomized into one of the three treatments onindividual characteristics such as gender, age, major, etc. Omnibus testsreveal that these characteristics are not jointly significant predictors of thetreatments (joint p − value = 0.18 for US Prime, 0.15 Chinese Prime, and0.47 for Neutral Prime, see Table S3).
For more rigorous statistical evidence we estimate an econometric modelin which we explicitly control for participants’ observable characteristics.To account for arbitrary within-session correlations in behavior – for ex-ample, stemming from differences in demographic variables we did notcollect or traits we cannot observe – we cluster standard errors by session.Our econometric model will allow us to insert a full set of indicator vari-ables for our three treatments so that we can answer the important auxiliaryquestion of whether a US identity increases cooperation or a Chinese iden-tity reduces cooperation relative to the baseline in which neither of these
p. 23).
11
identities is primed.Table 1 reports marginal effects from a series of probit models. The pri-
mary explanatory variables common to all columns are a set of dummyvariables for our treatments. The Chinese-prime treatment serves as ourexcluded category. For comparison with our consideration of the raw dataabove, the first column includes no additional controls. The estimatedmarginal effect of the US prime relative to the Chinese prime, 23.3 percent-age points, matches the percentage point difference observed in the rawdata. The statistical significance of this difference is enhanced to p = 0.008.The zero marginal effect associated with the Neutral-prime treatment sug-gests that the Chinese-identity prime has little effect on cooperation.
In the second column, we include individual controls for gender, age,whether the participant reported being an economics major as some haveargued that economics students are less pro-social (see, e.g., Frank, Gilovichand Regan, 1996), and birthplace of participants, i.e., whether they wereborn in the US (second-generation immigrant). Controlling for these demo-graphics slightly increases the coefficient estimate of the US prime, whichremains highly statistically significant (p = 0.003). Moreover, none ofthe controls are individually statistically significant nor jointly significant(χ2(4) = 4.14; p = 0.39).
It is conceivable that some of our subjects did not believe in our pay-ment procedure (i.e., one randomly determined task) or identified the pur-pose of our experiment. As both would add noise and bias to our treatmentestimates, we aim to control for these issues in the regressions reported incolumns (3) and (4) of Table 1. In total 19 subjects (19 percent) reported inthe final questionnaire that they did not believe that their responses wouldmatter for payment. Excluding those subjects leads to an even strongertreatment effect, i.e., subjects in the US Prime treatment are 29 percentagepoints more likely to cooperate than subjects in the Chinese Prime treat-ment (column 3). Similarly, if we exclude the four subjects who identifiedthe purpose of the experiment, the treatment effect appears stronger with a28 percentage point higher likelihood of cooperation in the US Prime.
In the supplementary material, we present further robustness checksvalidating our results (see Table S4). In particular, we control for culturalvariation within Chinese countries. First, our results are robust to control-ling for participants with at least one parent born in Hong Kong or Taiwan(column 1) or restricting the sample to participants whose parents wereborn in mainland China (either father or mother or both – columns 2-3).Second, we address the concern that historical crop cultivation practicesin mainland China may have led to varying degrees of collectivism (rice-
12
Table 1: Regression: Cooperation rates in the PD – Experiment 1
Dependent variable: PD cooperation(1) (2) (3) (4)
US Prime 0.233*** 0.250*** 0.290*** 0.276***(0.088) (0.085) (0.109) (0.085)
Neutral Prime 0.000 -0.0113 0.0334 0.0209(0.126) (0.130) (0.125) (0.125)
Individual controls no yes yes yes
Observations 102 102 83 98Pseudo R2 0.04 0.06 0.07 0.07Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Each column reports marginal effects estimates from a Probitmodel. Columns 3 and 4 restrict the data to participants whobelieved our payment procedure and participants who have notidentified the purpose of our study, respectively. Controls in-clude age, gender, economics major, and birthplace of subject(born in US). Robust standard errors clustered by session appearin parentheses.
wheat cultivation divide). Specifically, we use information on birth citiesof parents to identify whether participants have ancestral roots in the re-gions north of the Yangtze river where wheat cultivation may have shapeda more individualistic culture (Talhelm et al., 2014). Including this informa-tion reduces our sample due to missing information on parents’ birth city,but does not change our result.
Moreover, we find no differential priming effect on cooperation for par-ticipants born in the US (column 5). That is, while second-generation par-ticipants are still more cooperative under the US prime (0.61) than underthe Chinese prime (0.50), the gap is much smaller than for non-US-bornparticipants with cooperation rates of 0.35 in the Chinese Prime and 0.69 inthe US Prime. As such, it is consistent with second-generation participantsbeing less conflicted about their two cultural identities and adopting val-ues, social norms, and political behavior of natives over time (Fernandezand Fogli, 2009; Luttmer and Singhal, 2011; Blau et al., 2013).
Summarizing the evidence so far, the data reveal that priming a USidentity substantially and significantly increases cooperation relative to thecounter-factual of priming a Chinese identity. Comparing behavior to theNeutral-prime treatment suggests that the comparative increase in cooper-
13
ation associated with a US cultural identity primarily works through theUS identity increasing cooperation rather than a Chinese identity decreas-ing cooperation.
4.2 Evidence on the beliefs channel
Before turning to our second experiment, we outline some of the evidencefrom our first experiment consistent with a prominent role for the beliefschannel in culture’s effect on cooperation. First of all, we examine behaviorin decisions where beliefs about others’ cooperativeness should be less cen-tral. This is the case in the SPD where second-movers can condition theirbehavior on their partner’s cooperation. Indeed, we find no evidence thatsecond-mover SPD strategies are affected by our prime. The distribution ofchoices is not statistically different in the US Prime from the Chinese Prime(Fisher’s exact test, p = 0.49). Another decision in which beliefs shouldplay little role is in the die-rolling task, as there is no strategic uncertaintyinvolved. Here, we also we find no evidence that our primes affected be-havior. For example, reports do not vary across treamtents in the die-rollingtask where an increased empathy for laboratory-mates might show up asless dishonesty when higher own-gains are pitted against others’ earnings(Mann-Whitney test, p = 0.24).
Figure 3: PD Cooperation: beliefs vs behavior
0.1
.2.3
.4.5
.6.7
.8.9
1
American Chinese NeutralPrime
Average Expected Cooperation Rate in PD
0.1
.2.3
.4.5
.6.7
.8.9
1
US Chinese NeutralPrime
Average Cooperation Rate in PD
As our second piece of initial evidence, at the end of our experimentwe explicitly asked participants for their beliefs about their fellow par-ticipants’ PD cooperation rates. This question was not incentivized. Wepresent participants’ beliefs side-by-side with actual cooperation rates inFigure 3. Qualitatively, it appears as though beliefs follow the same patternas actions – the US prime induced both the highest actual cooperation rates
14
and the most sanguine beliefs about others’ cooperation. The difference be-tween Chinese-Prime and US-Prime participants’ beliefs is not statisticallysignificant, however (Mann-Whitney test, p = 0.39). When interpretingthese beliefs we have to keep two things in mind. First, these beliefs werenot remunerated to keep the task as simple as possible.15 Second, a pri-mary issue with relying on elicited beliefs is that they may be tainted bybehavior. Previous research suggests not only that individuals do projecttheir own behaviors onto others when forming beliefs in a process knownas “false consensus” (Ross, Greene and House, 1977; Butler, Giuliano andGuiso, 2015), but that they should (Schelling, 1966, p.150).
All together, the evidence from Experiment 1 suggests culture may af-fect cooperation through both a preference channel and through a beliefschannel. To more cleanly get a handle on how much of culture’s effect onobserved behaviors operates through beliefs, we conducted a second ex-periment in which we minimize culture’s direct effect on beliefs formationby providing concrete and credible information about the likelihood of co-operation.
5 Experiment 2
Treatments in our second experiment followed exactly the same proceduresas in our first experiment except that immediately before participants madetheir decisions in the PD we provided them with information about co-operation rates in our previous experiments. Because recruiting Chinesesubjects proved difficult in Experiment 1, to economize on the number ofparticipants needed for adequate statistical power, in Experiment 2 we im-plemented only the Chinese-prime and the US-prime treatments.
5.1 Experimental design and procedures
To convey information about past behavior in the PD, participants vieweda screen immediately before submitting their decisions in the PD with thefollowing text (where decisions “option L” and “option T” refer to the co-operative outcome): “We have conducted many experimental sessions herein the X-lab over the last year that have involved exactly this same game.Participants in these previous experimental sessions were recruited from
15Some evidence suggests that while incentivizing belief elicitation improves consistencywith own behavior, non-incentivized beliefs are as predictive of others’ behavior as incen-tivized beliefs (e.g., Trautmann and Kuilen, 2015).
15
the same pool of people as participants for today’s experiment. To giveyou an idea of how previous participants in the role of the column [row]player played this game, a random selection of 10 participants reveals thatX out of these 10 participants chose ‘option L’ [‘option T’].”
To fill in the “X” for each participant, we drew a sequence of randomsamples of size ten from the Experiment 1 PD data. Within each treatment,each participant received information derived from a different element inthis sequence. To plausibly induce similar or, ideally, identical beliefs acrossour two treatments, we assigned elements of the sequence in parallel toparticipants across treatments.
To illustrate the process, suppose we drew 100 random samples of sizeten from our Experiment 1 PD data which resulted in a sequence {X1, ..., X100}of cooperation rates. The first participant in the Chinese-prime treatmentof Experiment 2 would receive the information X1, the second participantwould receive information X2 and so on. The first participant in the US-Prime treatment of Experiment 2 would also receive information X1, whilethe second US-prime participant would receive information X2, etc. Wedrew our sequence of subsamples before conducting any sessions of oursecond experiment. Our sequence was sufficiently long to ensure that wewould not exhaust its elements.
This process introduces the same information across treatments but dif-ferent information across individuals, permitting us to estimate the causalimpact of culture on preferences for cooperation holding (induced) beliefsfixed in both conditions. This is the primary advantage over the perhapsmore straightforward option of transmitting the same information to allparticipants. It also allows us to truthfully instruct participants that we“randomly selected” the information we were providing without runningthe risk of selecting unlikely information to transmit (e.g., all previous par-ticipants cooperated) raising concerns about experimenter demand effect.
We conducted the experiment at the X-lab of the University of Califor-nia, Berkeley. Again, we attempted to recruit only Chinese students. Be-cause we were restricted to the same recruiting process as in Experiment 1(inviting only participants with common Chinese last names), the recruit-ment was not perfectly successful. Therefore, we exclude from our analy-sis any participant who did not have at least one parent born in a Chinesecountry.16 In total, we have observations from 132 participants: 68 in the
16One might worry that excluding ex-post participants who were neither first- norsecond-generation Chinese may have introduced an imbalance in information transmit-ted within each treatment. However, this was not the case. On average, Chinese-primeparticipants were informed that 50% of previous participants cooperated in the PD, while
16
US-prime treatment and 64 in the Chinese-prime treatment. Sessions lastedabout 45 minutes and participants again earned 16 dollars on average.
6 Results: experiment 2
By exogenously introducing information about how previous participantsdrawn from the same subject pool played the PD, our goal was to greatlyameliorate the direct influence that our primes had on the beliefs-formationprocess, while leaving intact their direct influence on preferences. Thisallows us to more clearly reveal the effect of our cultural primes on pref-erences for cooperation. We start by examining the effect of the informa-tion provided. First, as expected the provided information about others’cooperation is positively related to cooperation (p = 0.041). Second, onemight expect that the beliefs we elicit at the end of the experiment wouldstrongly reflect the information we provided, since participants receivedno feedback during the experiment about actual behavior. While this wasthe case for US-prime participants (0.53 vs. 0.48, Wilcoxon signed-rank test,p = 0.41), participants in the Chinese prime treatment tended to state be-liefs higher than the information provided (0.65 vs 0.50, p = 0.01). Wheninterpreting these beliefs we have to keep in mind that they were not in-centivized and possibly tainted by participants’ behavior, not only in thePD but also in the other decisions made during the experiment.
Turning to PD cooperation, we see that the cooperation rate in the US-prime treatment was 41.2%, while 59.4% of participants in the Chinese-prime treatment chose to cooperate. This difference is statistically signifi-cant at a nearly conventional level (p = 0.053; two-tailed permutation test)and, again, substantial (Hedge’s g = 0.369). It is also the mirror image ofcooperation rates in Experiment 1.
In Table 2, we report marginal effects from several probit model esti-mates of PD cooperation. In our most basic specification (column 1) ouronly explanatory variable is an indicator for the US-prime treatment. Thenegative and statistically significant coefficient indicates that a US culturalidentity reduces cooperation substantially. In column 2 and 3, we control forthe information about previous cooperation the participant received andintroduce the same demographic controls as in our analysis of experiment1. None of these additional controls have much of an effect on the esti-mated magnitude, sign, nor statistical significance of the marginal effect
this figure was 48% for US-prime participants, a difference which was far from significant(χ2(6) = 2.59, p = 0.86).
17
Table 2: Regression: Cooperation rates in the PD – Experiment 2
Dependent variable: PD cooperation(1) (2) (3) (4) (5)
US Prime -0.179*** -0.170*** -0.176*** -0.106 -0.182***(0.059) (0.058) (0.064) (0.080) (0.068)
Information 0.049* 0.047* 0.066** 0.041*(0.025) (0.026) (0.027) (0.024)
Individual controls no no yes yes yes
Observations 132 132 132 92 118Pseudo R2 0.02 0.04 0.06 0.08 0.05Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Each column reports marginal effects estimates from a Probit model.Columns 4 and 5 restrict the data to participants who believed our pay-ment procedure and participants who have not identified the purposeof our study, respectively. Controls include age, gender, economics ma-jor, and birthplace of subject (born in US). Robust standard errors clus-tered by session appear in parentheses.
of a US identity. Consistent with our analysis above information is sig-nificantly and positively associated with cooperation, suggesting that bothbeliefs and preferences play a role in cooperation.
As before in experiment 1, none of the demographic controls are sig-nificant – neither jointly nor individually. In columns 4 and 5 we restrictour sample. Excluding participants who believed that their decisions didnot affect their payment reduces the coefficient estimate substantially andreduces statistical significance. In the last column we exclude participantswho identified the purpose of our experiment. This yields estimates andstatistical significance similar to the first three columns.
Finally, we present the same set of robustness checks as before in thesupplementary material (see Table S5). Again, our results are robust tocontrolling for variation in collectivsm stemming from the rice-wheat cul-tivation divide as well as whether the participant was a first- or second-generation immigrant. Moreover, in Table S6 we pool the data from ourUS and Chinese-Prime treatments in the two experiments and re-estimateour treatment effects while including a full set of controls as in the pre-vious regressions. We find that differences in the controls cannot explainthe behavioral differences in the two experiments. In particular, we findno indication that the behavioral differences in our second experiment are
18
driven by cultural differences as suggested by the cultivation of wheat.Although our results appear robust to various potential confounders,
one might still worry that finding the opposite results in our second exper-iment compared to our first experiment implies that the effects in both ex-periments are spurious. To partially address this concern, we compare co-operation rates among subsets of participants who received the same primeand held plausibly similar beliefs across experiments (see Figure S3 in thesupplementary material). Specifically, given that the Chinese prime in Ex-periment 1 induced pessimistic beliefs, we can compare these participantswith the subset of participants in Experiment 2 who received the Chineseprime and pessimistic information about others’ cooperation behavior (i.e.,that less than 5 in 10 subjects cooperated). Cooperation rates are virtuallyidentical in these two groups (43% vs. 46%, Fisher’s exact test, p = 1). Ina similar vein, participants in the US-prime treatment in Experiment 1 re-ported more optimistic beliefs about others’ behavior. Comparing them tothe subset of participants in Experiment 2 who received the US prime andoptimistic information about others’ behavior (i.e., that more than 5 in 10subjects cooperated) reveals no statistical difference (67% vs. 55%, Fisher’sexact test, p = 0.42).
The findings of Experiment 2 suggest that beliefs play a substantial rolein cultures’ impact on cooperation. After constraining culture’s capacityto affect beliefs directly, we find that a collectivist cultural identity has theeffect on the preference for cooperation that might be expected, enhancingthe attractiveness of cooperative behavior. This is consistent with findingsfrom LeBoeuf, Shafir and Bayuk (2010), who document that Chinese sub-jects cooperate more in a hypothetical PD if their Chinese identity (relativeto their US identity) is made salient. This contrasts sharply with our find-ings in Experiment 1, which represent the joint effect of culture’s effectson preferences and beliefs. Putting these two sets of results together sug-gests that culture’s direct effect on the beliefs-formation process in situa-tions involving strategic uncertainty is strong enough to overturn its effectson preferences.
7 Concluding Remarks
In this paper, we adopt the view that cultural identity is one aspect of in-dividuals’ typically multi-faceted self-concept. We borrow a technique thathas been used repeatedly successfully in the identity economics literatureto reveal the causal impact of particular identities on behavior: priming.
19
In an initial experiment, we primed either a collectivist (Chinese) identity,an individualist (US) identity or neither and found that, surprisingly, theindividualist cultural identity induced substantially more cooperation. Todig deeper into this unexpected result, we conducted a second experimentin which we attempted to ameliorate the direct effect of our cultural primeon beliefs by providing concrete information about previous participants’cooperation rates. Shutting down the beliefs channel, our data suggest thata collectivist cultural identity enhanced participants’ preference for coop-eration. Comparing results across experiments, our results imply that thepuzzling result from our first experiment largely stems from a relative pes-simism about strangers’ cooperation rates being induced by Chinese cul-ture relative to US culture. Consequently, overall our results suggest thatculture’s effect on the beliefs formation process in situations of strategicuncertainty warrants more than the scant research attention it has thus farreceived.
Taking our results at face value, one may wonder how they can be rec-onciled with important findings in the cultural economics literature docu-menting long-term persistence of cultural traits and behavior. However, itis important to note that there is a recent debate among prominent schol-ars on cultural persistence. Fernandez (2011, p. 484) argues forcefully that“... a definition of culture that considers [culture] to be slow-moving (see,e.g. Guiso, Sapienza, and Zingales (2006)) is rejected. The speed of culturalchange depends on how quickly social beliefs and preferences change overtime, which in turn depends on the environment broadly speaking ....” Ourresults contribute support to both sides of this debate. We demonstrate theimportance of both an ostensibly slow-moving component – preferences– as well as a cultural component that may move much more quickly –beliefs. Indeed, our second experiment suggests that particular culturalbeliefs may be readily updated to incorporate new information. Whetherand, if so, how quickly, these updated beliefs are subsequently absorbedand transmitted beyond the individual by culture is an important openquestion.
20
References
Akerlof, George A, and Rachel E Kranton. 2000. “Economics and identity.”The Quarterly Journal of Economics, 115(3): 715–753.
Akerlof, George A, and Rachel E Kranton. 2002. “Identity and schooling:Some lessons for the economics of education.” Journal of Economic Litera-ture, 40(4): 1167–1201.
Akerlof, George A., and Rachel E. Kranton. 2005. “Identity and the Eco-nomics of Organizations.” Journal of Economic Perspectives, 19(1): 9–32.
Akerlof, George A., and Rachel E. Kranton. 2010. Identity Economics: howour identities shape our work, wages, and well-being. Princeton, New Jer-sey:Princeton University Press.
Alesina, Alberto, and Paola Giuliano. 2015. “Culture and institutions.”Journal of Economic Literature, 53(4): 898–944.
Alesina, Alberto, Paola Giuliano, and Nathan Nunn. 2013. “On the ori-gins of gender roles: Women and the plough.” The Quarterly Journal ofEconomics, 128(2): 469–530.
Algan, Yann, and Pierre Cahuc. 2010. “Inherited trust and growth.” TheAmerican Economic Review, 100(5): 2060–2092.
Algan, Yann, and Pierre Cahuc. 2013. “Trust and growth.” Annual Reviewof Economics, 5(1): 521–549.
Asch, Solomon E. 1955. “Opinions and social pressure.” Scientific American,193(5): 31–35.
Battigalli, Pierpaolo, and Martin Dufwenberg. 2007. “Guilt in games.”American Economic Review, 97(2): 170–176.
Benabou, Roland, and Jean Tirole. 2003. “Intrinsic and extrinsic motiva-tion.” The Review of Economic Studies, 70(3): 489–520.
Benabou, Roland, and Jean Tirole. 2006a. “A cognitive theory of identity,dignity, and taboos.”
Benabou, Roland, and Jean Tirole. 2006b. “Incentives and prosocial behav-ior.” The American Economic Review, 96(5): 1652–1678.
21
Benjamin, Daniel J, James J Choi, A Joshua Strickland, et al. 2010. “SocialIdentity and Preferences.” The American Economic Review, 100(4): 1913–1928.
Bernheim, B Douglas. 1994. “A theory of conformity.” The Journal of Politi-cal Economy, 102(5): 841–877.
Besley, Timothey, and Torsten Persson. 2018. “Organizational Dynamics:Culture, Design, and Performance.” Mimeo, LSE and IIES.
Bicchieri, Cristina. 2006. The Grammar of Society. Cambridge UniversityPress.
Binswanger, Hans P. 1981. “Attitudes toward risk: Theoretical implicationsof an experiment in rural India.” The Economic Journal, 91(364): 867–890.
Bisin, Alberto, and Thierry Verdier. 2008. “The New Palgrave Dictionaryof Economics, Second Edition.” , ed. Steven N. Durlauf and Lawrence E.Blume, Chapter Cultural Transmission. Oxford:Palgrave MacMillan.
Blau, Francine D, Lawrence M Kahn, Albert Yung-Hsu Liu, and Kerry LPapps. 2013. “The transmission of women’s fertility, human capital, andwork orientation across immigrant generations.” Journal of PopulationEconomics, 26(2): 405–435.
Buggle, Johannes C. 2017. “Growing Collectivism: Irrigation, Group Con-formity and Technological Divergence.” mimeo.
Butler, Jeffrey V. 2018. “Identity and the theory of the firm.” Handbook of Be-havioral Industrial Organization, Cheltenham and Northampton, MA: EdwardElgar Publishing, 224–53.
Butler, Jeffrey V, Paola Giuliano, and Luigi Guiso. 2015. “Trust, values,and false consensus.” International Economic Review, 56(3): 889–915.
Butler, Jeffrey V, Paola Giuliano, and Luigi Guiso. 2016. “The rightamount of trust.” Journal of the European Economic Association, 14(5): 1155–1180.
Butler, Jeffrey V, , Paola Giuliano, and Luigi Guiso. 2016. “Trust andCheating.” The Economic Journal, 126: 1703–1738.
Cameron, Lisa, Nisvan Erkal, Lata Gangadharan, and Xin Meng. 2013.“Little emperors: behavioral impacts of China’s One-Child Policy.” Sci-ence, 339(6122): 953–957.
22
Charness, Gary, and Martin Dufwenberg. 2006. “Promises and partner-ship.” Econometrica, 74(6): 1579–1601.
Charness, Gary, Uri Gneezy, and Alex Imas. 2013. “Experimental meth-ods: Eliciting risk preferences.” Journal of Economic Behavior & Organiza-tion, 87: 43–51.
Chen, Yan, Sherry Xin Li, Tracy Xiao Liu, and Margaret Shih. 2014.“Which hat to wear? Impact of natural identities on coordination andcooperation.” Games and Economic Behavior, 84: 58–86.
Cohn, Alain, Ernst Fehr, and Michel Andre Marechal. 2014. “Business cul-ture and dishonesty in the banking industry.” Nature, 516(7529): 86–89.
Cohn, Alain, Michel Andre Marechal, and Thomas Noll. 2015. “Bad boys:How criminal identity salience affects rule violation.” The Review of Eco-nomic Studies, 82(4): 1289–1308.
Enke, Benjamin. 2017. “Kinship systems, cooperation and the evolution ofculture.” National Bureau of Economic Research.
Falk, Armin, Anke Becker, Thomas Dohmen, Benjamin Enke, DavidHuffman, and Uwe Sunde. 2017. “Global evidence on economic pref-erences.” The Quarterly Journal of Economics, forthcoming.
Fernandez, Raquel. 2008. “Culture and economics.” In The New PalgraveDictionary of Economics, 2nd edition. , ed. Steven N Durlauf and LawrenceBlume, 333–40. Palgrave Macmillan Basingstoke.
Fernandez, Raquel. 2011. “Does Culture Matter?” In The Handbook of SocialEconomics, Vol 1A. , ed. Jess Benhabib, Matthew O. Jackson and AlbertoBisin, 481–510. North-Holland.
Fernandez, Raquel, and Alessandra Fogli. 2006. “Fertility: The role of cul-ture and family experience.” Journal of the European Economic Association,4(2-3): 552–561.
Fernandez, Raquel, and Alessandra Fogli. 2009. “Culture: An empiricalinvestigation of beliefs, work, and fertility.” American Economic Journal:Macroeconomics, 1(1): 146–77.
Fischbacher, Urs. 2007. “z-Tree: Zurich toolbox for ready-made economicexperiments.” Experimental Economics, 10(2): 171–178.
23
Frank, Robert H, Thomas D Gilovich, and Dennis T Regan. 1996. “Doeconomists make bad citizens?” The Journal of Economic Perspectives,10(1): 187–192.
Gaechter, Simon, Jonathan Schulz, and Christian Thoeni. 2017. “CulturalOrigins of Cooperation.” mimeo.
Geanakoplos, John, David Pearce, and Ennio Stacchetti. 1989. “Psycho-logical games and sequential rationality.” Games and Economic Behavior,1(1): 60–79.
Giuliano, Paola. 2007. “Living arrangements in western europe: Doescultural origin matter?” Journal of the European Economic Association,5(5): 927–952.
Gorodnichenko, Yuriy, and Gerard Roland. 2011. “Which dimensions ofculture matter for long-run growth?” The American Economic Review,101(3): 492–98.
Gorodnichenko, Yuriy, and Gerard Roland. 2017. “Culture, institutions,and the wealth of nations.” Review of Economics and Statistics, 99(3): 402–416.
Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2006. “Does Culture Af-fect Economic Outcomes?” The Journal of Economic Perspectives, 20(2): 23–48.
Hajikhameneh, Aidin, and Erik O Kimbrough. 2018. “Individualism, col-lectivism, and trade.” Experimental Economics, forthcoming.
Heine, Steven J. 2015. Cultural Psychology. WW Norton & Company.
Henrich, J., J. Ensminger, R. McElreath, A. Barr, C. Barrett, A. Bolyanatz,J. C. Cardenas, M. Gurven, E. Gwako, N. Henrich, C. Lesorogol, F. Mar-lowe, D. P. Tracer, and J. Ziker. 2010. “Market, religion, community sizeand the evolution of fairness and punishment.” Science, 327(5972): 1480–1484.
Henrich, J., R. Boyd, S. Bowles, C. Camerer, E. Fehr, H. Gintis, andR. McElreath. 2001. “In search of Homo economicus: Behavioral ex-periments in 15 small-scale societies.” The American Economic Review,91(2): 73–78.
24
Henrich, J., R. Boyd, S. Bowles, C. Camerer, E. Fehr, H. Gintis, R. McEl-reath, M. Alvard, A. Barr, J. Ensminger, N. S. Henrich, K. Hill, F. Gil-White, M. Gurven, F. W. Marlowe, J. Q. Patton, and D. Tracer. 2005.“”Economic man” in cross-cultural perspective: Behavioral experimentsin 15 small-scale societies.” Behavioral and Brain Sciences, 28(6): 795–855.
Hofstede, Geert, Gert Jan Hofstede, and Michael Minkov. 2010. Culturesand organizations: software of the mind: 3rd edition. McGraw-Hill Profes-sional.
Institute of Education Sciences. 2014. “What Works Clearinghouse: Proce-dures and Standards Handbook Version 3.0.”
Knack, Stephen, and Philip Keefer. 1997. “Does social capital have an eco-nomic payoff? A cross-country investigation.” The Quarterly journal ofEconomics, 112(4): 1251–1288.
LeBoeuf, Robyn A, Eldar Shafir, and Julia Belyavsky Bayuk. 2010. “Theconflicting choices of alternating selves.” Organizational behavior and hu-man decision processes, 111(1): 48–61.
Lowes, Sara, Nathan Nunn, James A Robinson, and Jonathan L Weigel.2017. “The evolution of culture and institutions: Evidence from the KubaKingdom.” Econometrica, 85(4): 1065–1091.
Luttmer, Erzo FP, and Monica Singhal. 2011. “Culture, context, and thetaste for redistribution.” American Economic Journal: Economic Policy,3(1): 157–79.
Nisbett, Richard. 2010. The Geography of Thought: How Asians and WesternersThink Differently... and Why. Simon and Schuster.
Nisbett, Richard E, Kaiping Peng, Incheol Choi, and Ara Norenzayan.2001. “Culture and systems of thought: holistic versus analytic cogni-tion.” Psychological review, 108(2): 291.
Oyserman, Daphna, Heather M Coon, and Markus Kemmelmeier. 2002.“Rethinking individualism and collectivism: evaluation of theoretical as-sumptions and meta-analyses.” Psychological Bulletin, 128(1): 3.
Ross, Lee, David Greene, and Pamela House. 1977. “The “false consen-sus effect”: An egocentric bias in social perception and attribution pro-cesses.” Journal of Experimental Social Psychology, 13(3): 279–301.
25
Sapienza, Paola, Anna Toldra-Simats, and Luigi Zingales. 2013. “Under-standing trust.” The Economic Journal, 123(573): 1313–1332.
Schelling, Thomas C. 1966. “Arms and influence.” New Haven: Yale.
Schram, Arthur, and Gary Charness. 2015. “Inducing social norms in lab-oratory allocation choices.” Management Science, 61(7): 1531–1546.
Tabellini, Guido. 2008. “The scope of cooperation: Values and incentives.”The Quarterly Journal of Economics, 123(3): 905–950.
Tabellini, Guido. 2010a. “Culture and Institutions: Economic Develop-ment in the Regions of Europe.” Journal of the European Economic Asso-ciation, 8(4): 677–716.
Tabellini, Guido. 2010b. “Culture and institutions: economic developmentin the regions of Europe.” Journal of the European Economic Association,8(4): 677–716.
Tajfel, H., C. Flament, M.G. Billig, and R.F. Bundy. 1971. “Social Catego-rization and Intergroup Behavior.” European Journal of Social Psychology,1: 149–177.
Talhelm, Thomas, Xuemin Zhang, Shige Oishi, Chen Shimin, DechaoDuan, Xiaoli Lan, and Shinobu Kitayama. 2014. “Large-scale psycho-logical differences within China explained by rice versus wheat agricul-ture.” Science, 344(6184): 603–608.
Trautmann, Stefan T, and Gijs Kuilen. 2015. “Belief elicitation: A horserace among truth serums.” The Economic Journal, 125(589): 2116–2135.
Triandis, Harry C. 2001. “Individualism-collectivism and personality.”Journal of Personality, 69(6): 907–924.
Yamagishi, Toshio, Karen S Cook, and Motoki Watabe. 1998. “Uncer-tainty, trust, and commitment formation in the United States and Japan.”The American Journal of Sociology, 104(1): 165–194.
26
Supplementary Material
Figures
Figure S1: Prisoners’ Dilemma
Cooperate DefectCooperate 11, 11 0, 12Defect 12, 0 5, 5
Figure S2: Sequential Prisoners’ Dilemma
C DCC $11, $11 $0, $12DD $12, $0 $5, $5CD $11, $11 $5, $5DC $12, $0 $0, $12
Figure S3: Cooperation Rates across same Prime and Beliefs
0.1
.2.3
.4.5
.6.7
.8.9
1
Exp1 US Exp2 US/high info Exp1 CHN Exp2 CHN/low info
Average Cooperation Rate in PD in Experiment 1 & 2
27
Balance checks
Table S1: Summary statistics: Experiment 1
Variable Chinese prime US prime p-value
N mean sd mean sd
Male 81 0.36 0.07 0.26 0.07 0.35
Age 81 19.9 0.35 19.8 0.28 0.81
Econ 81 0.14 0.05 0.15 0.06 1.00
Born in US 81 0.52 0.08 0.33 0.08 0.12
Taiwan 81 0.02 0.02 0.15 0.06 0.05
Hongkong 81 0.10 0.05 0.05 0.04 0.68
Wheat cultivation 61 0.41 0.14 0.48 0.13 0.35
Table S2: Summary statistics: Experiment 2
Variable Chinese prime US prime p-value
N mean sd mean sd
Male 132 0.39 0.06 0.32 0.06 0.47
Age 132 20.2 0.25 20.1 0.28 0.51
Econ 132 0.14 0.04 0.24 0.05 0.86
Born in US 132 0.44 0.06 0.41 0.06 0.55
Taiwan 132 0.20 0.05 0.18 0.05 0.83
Hongkong 132 0.05 0.03 0.06 0.03 1.00
Wheat cultivation 91 0.65 0.13 0.84 0.13 0.075
28
Table S3: Regression: Balance check Experiment 1 & 2
Experiment 1 Experiment 2US prime Chinese prime Neutral prime
(1) (2) (3) (4)
Male -0.087 0.122 -0.054 -0.056(0.106) (0.113) (0.090) (0.094)
Age -0.007 0.005 -0.000 -0.008(0.024) (0.023) (0.020) (0.021)
Economics major -0.085 -0.107 0.212 0.146(0.125) (0.126) (0.133) (0.112)
Born in US -0.262*** 0.160 0.099 -0.030(0.098) (0.103) (0.089) (0.093)
Taiwan 0.355** -0.375*** 0.0631 -0.011(0.161) (0.096) (0.153) (0.119)
Hongkong 0.027 0.231 0.076(0.228) (0.224) (0.197)
Observations 102 102 96 132Pseudo R2 0.07 0.08 0.05 0.02Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Columns 1–3 report estimates from a Probit model of the probability of being randomizedinto one of the three treatments in Experiment 1 on individual characteristics. Column 4reports the probability of being randomized into one of the three treatments in Experiment2.
29
Robustness Checks
Table S4: Robustness checks: Experiment 1
Dependent variable: PD cooperation
Sample: restricted full restricted restricted full full(1) (2) (3) (4) (5) (6)
US Prime 0.294*** 0.282*** 0.297*** 0.260*** 0.317*** 0.224**(0.0856) (0.104) (0.101) (0.0804) (0.106) (0.0955)
Neutral Prime 0.0418 0.0329 0.108 0.120 -0.0141 -0.0153(0.125) (0.149) (0.169) (0.192) (0.131) (0.132)
Male 0.0778 0.0201 -0.00936 0.0860 0.0847 0.0895(0.121) (0.129) (0.123) (0.141) (0.122) (0.129)
Age 0.0132 0.00171 0.0307 0.0543* 0.0176 0.0204(0.0241) (0.0272) (0.0303) (0.0278) (0.0259) (0.0283)
Economics major 0.105 0.0405 -0.0122 0.131 0.131 0.115(0.118) (0.133) (0.136) (0.145) (0.112) (0.120)
Born in US 0.0719 0.0840 0.143 0.177 0.107 0.107(0.114) (0.107) (0.112) (0.123) (0.155) (0.222)
Taiwan -0.217(0.227)
Hongkong 0.323*(0.171)
Wheat cultivation -0.0484(0.0577)
US Prime x Born in US -0.165(0.241)
Years lived in US -0.00454(0.0128)
Joint sig. controls (p-value) 0.21 0.93 0.73 0.19 0.50 0.66
Observations 102 86 81 60 102 99Pseudo R2 0.09 0.05 0.06 0.12 0.06 0.05Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Each column reports marginal effects estimates from a Probit model. Column 1 excludes participantswith at least one parent born in Hong Kong and Taiwan. Columns 3 and 4 restrict the data setto participants whose parents were both born in China. Column 5 and 6 use the full data set ofexperiment 1. Robust standard errors clustered by session appear in parentheses.
30
Table S5: Robustness checks: Experiment 2
Dependent variable: PD cooperation
Sample: full restricted restricted restricted full full(1) (2) (3) (4) (5) (6)
US Prime -0.177*** -0.203*** -0.190** -0.260*** -0.143* -0.180***(0.0641) (0.0675) (0.0749) (0.0884) (0.0849) (0.0657)
Information 0.0429* 0.0597** 0.0625** 0.0858** 0.0444* 0.0464*(0.0259) (0.0264) (0.0310) (0.0335) (0.0241) (0.0244)
Male -0.116 -0.0238 0.0378 0.0951 -0.120 -0.120(0.101) (0.0912) (0.0993) (0.104) (0.0997) (0.0982)
Age -0.00240 -0.0144 -0.0149 -0.0168 -0.00275 -0.001(0.0228) (0.0214) (0.0217) (0.0173) (0.0236) (0.0237)
Economics major 0.0175 0.0406 0.0625 0.0605 0.0144 0.004(0.0925) (0.0938) (0.101) (0.0924) (0.0993) (0.096)
Born in US 0.0764 0.0321 0.0672 -0.0510 0.132 0.130(0.0915) (0.0973) (0.0812) (0.0645) (0.158) (0.179)
Taiwan 0.0613(0.101)
Hongkong 0.0645(0.170)
Wheat cultivation -0.0167(0.0665)
US Prime x Born in US -0.0815(0.183)
Years lived in US -0.00278(0.00975)
Joint sig. controls (p-value) 0.69 0.96 0.76 0.45 0.58 0.60
Observations 132 100 94 69 132 132Pseudo R2 0.06 0.06 0.07 0.13 0.06 0.06Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Columns 1–6 report marginal effects estimates from a Probit model. Column 1 excludes participantswith at least one parent born in Hong Kong and Taiwan. Columns 3 and 4 restrict the data setto participants whose parents were both born in China. Column 5 and 6 use the full data set ofexperiment 2.Robust standard errors clustered by session appear in parentheses.
31
Table S6: Robustness checks: Experiment 1 & 2
Dependent variable: PD cooperation
(1) (2) (3)
US Prime 0.235*** 0.248*** 0.248***(0.0899) (0.0849) (0.0870)
Experiment 2 0.161* 0.174** 0.201*(0.0945) (0.0883) (0.112)
US Prime x Experiment 2 -0.412*** -0.432*** -0.526***(0.105) (0.0999) (0.114)
Male -0.0843 0.0388(0.0743) (0.0813)
Age 0.0149 0.0183(0.0186) (0.0169)
Economics major 0.0256 0.0446(0.0785) (0.0943)
Born in US 0.0710 0.0424(0.0736) (0.0915)
Taiwan 0.0156(0.117)
Hongkong 0.210* 0.114(0.121) (0.143)
Wheat cultivation -0.0124(0.0994)
Joint sig. controls (p-value) 0.18 0.89
Observations 213 213 130Pseudo R2 0.03 0.05 0.07Notes: * p < 0.10, ** p < 0.05, *** p < 0.01Each column reports marginal effects estimates from a Probitmodel using data from both experiments (except the Neutral-Prime treatment in Experiment 1). Robust standard errors clus-tered by session appear in parentheses.
32