Trust games: A meta-analysis

https://doi.org/10.1016/j.joep.2011.05.007Get rights and content

Abstract

We collect data from 162 replications of the Berg, Dickhaut, and McCabe Investment game (the “trust” game) involving more than 23,000 participants. We conduct a meta-analysis of these games in order to identify the effect of experimental protocols and geographic variation on this popular behavioral measure of trust and trustworthiness. Our findings indicate that the amount sent in the game is significantly affected by whether payment is random, and whether play is with a simulated counterpart. Trustworthiness is significantly affected by the amount by which the experimenter multiplies the amount sent, whether subjects play both roles in the experiment, and whether the subjects are students. We find robust evidence that subjects send less in trust games conducted in Africa than those in North America.

Highlights

► Data are aggregated from 162 trust games involving over 23,000 participants. ► Subjects playing for random payments, playing with a simulated counterpart, or playing in Africa send less in the trust game. ► Receivers who are students, playing both roles, or playing the game with a lower multiplier send back less. ► We find no significant impact of the double blind protocol on behavior.

Introduction

Trusting in others and reciprocating that trust with trustworthy actions are everyday aspects of life. In most neighborhoods, most of the time, unlocked doors remain unopened, lost wallets containing cash are returned, and the vast majority of contracts, thankfully, remain incomplete. This is fortunate, since all of this trust and trustworthiness is good for the economy. Trust within organizations increases efficiency by lower monitoring costs (e.g. Frank, 1988), lowering turnover (Dirks & Ferrin, 2002), and increasing uncompensated positive behavior from employees (Dirks and Ferrin, 2002, Konovsky and Pugh, 1994). On a higher level of aggregation, scholars have linked a shared willingness to engage in trusting or trustworthy behavior to better economic outcomes (Arrow, 1972, Fukuyama, 1995, Putnam, 1993). Knack and Keefer (1997) use cross-country data to show a link between higher trust and higher GDP per capita. Higher levels of trust have been associated with more efficient judicial systems, higher quality government bureaucracies, lower corruption, and greater financial development (Guiso et al., 2004, La Porta et al., 1997).

Over the last two decades the measurement of trust and trustworthiness has been revolutionized through the use of laboratory experiments. One of the earliest moves in this direction was a game constructed by Camerer and Weigelt (1988). A simplified version of their experiment designed by Berg, Dickhaut and McCabe (BDM, 1995) has come to dominate the field. The BDM two stage trust game has become a popular and frequently replicated measure of behavioral trust and trustworthiness.1 It involves a sequential exchange in which there is no contract to enforce agreements. Subjects are endowed with $10, anonymously paired and assigned to either the role of sender or receiver. At stage one of the game, the sender may either pass nothing, or any portion x of the endowment (0  x  10) to the receiver. The sender then keeps 10  x, and the experimenter triples the remaining money so that 3x is passed onto the receiver. In stage two, the receiver may either pass nothing, or pass any portion y of the money received (0  y  3x) back to the sender. The amount passed by the sender is said to capture trust, “a willingness to bet that another person will reciprocate a risky move (at a cost to themselves),” and the amount returned to the trustor by the trustee to capture trustworthiness (Camerer, 2003, p. 85).

Berg and colleagues identified a considerable willingness to trust and reciprocate trust among subjects engaging in the one-time, anonymous, and controlled exchange setting – a result which deviated substantially from predictions assuming rational and self-interested opportunistic behavior. Some recent work has examined the limitations of this measure, arguing that the amount sent in the game confounds trust with altruism (Cox, 2004) and betrayal aversion (Bohnet et al., 2008, Fehr, 2009) and omits other important facets of trust (Ben-Ner and Halldorsson, 2010, Ermisch et al., 2009). Nevertheless, the game remains a popular choice among trust researchers. The trust game has been replicated across numerous countries, often using slightly different experimental protocols than the original BDM game. We take advantage of this diversity to perform a meta-analysis of the trust game in order to identify the effect of different experimental protocols and unobserved factors correlated with geographic region on this behavioral measure of trust and trustworthiness. Our data set covers 162 replications of the game across 35 countries. On average, there were 148 players in each of these replications and a total of 23,924 individuals.

While frequently used in fields such as medicine and psychology, meta-analyses are less common in economics (Van den Bergh, Button, Nijkamp, & Pepping, 1997). Given the relative youth of the field of experimental economics, it should come as no surprise that there were few meta-studies before 2000 or so.2 However, in the last few years there has been a proliferation of informal surveys of experimental results (Bowles, 2008, Chaudhuri, 2010, Croson and Gneezy, 2009, Kagel and Roth, 1995), many of which are focused on the trust game (Camerer, 2003, chap. 2; Chaudhuri, 2009, Fehr, 2009). Our paper differs from these in that we systematically identify experimental and geographic variables common to as many iterations of the trust game as possible, create a consistent data set, and employ formal econometric tools to identify the effect of these variables on the amount sent and returned in the trust game. In this sense, our study most resembles those performed by Zelmer (2003) for the public goods game, Oosterbeek, Sloof, and Van de Kuilen (2004) for the ultimatum game, and Jones (2008) for the prisoner’s dilemma game.

The most important reason to conduct a meta-analysis of the trust game is to verify the generalizability of its findings.3 It has been well known for some time that small changes to experimental protocols can have dramatic effects on behavior in the lab. For example, “framing effects” involving the alteration of a single word in subject instructions, like those demonstrated by Burnham, McCabe, and Smith (2000) for the public goods game, can significantly affect behavior (see also Bohnet and Cooter, 2005, Hertwig and Ortmann, 2001, Roth, 1995). Similar observations have been made with respect to frequently used procedures in the trust game such as whether or not the experiment is double blind (Berg et al., 1995, Hoffman et al., 1994) or if the strategy method is used to elicit second mover decisions (Casari & Cason, 2009). By looking at a large number of games, we identify which experimental protocols matter and by how much.

In addition to asking whether the rules used in the lab bias trust game behavior, we also investigate whether the average subject pool is representative. Most experiments conducted in the West use cheap, readily available, student subjects. There is evidence, however, that students are less trusting and exhibit less trustworthiness than adults (Fehr & List, 2004). More generally, even if one argues that experimental protocols and subject pools do not generate significant biases, it is possible that there are unobserved characteristics of the local population which make inferences about behavior difficult to generalize into other geographic regions (for examples see Henrich et al., 2010). Indeed, given the likelihood that trust and trustworthiness are, at least partly, endogenous to local institutions, we would be surprised if there are not systematic differences across geographic regions (Fehr, 2009). There is evidence for geographic variation in performance in ultimatum games (Oosterbeek et al., 2004), public goods games (Gächter, Herrmann, & Thöni, 2010), as well as dictator games (Henrich et al., 2001) and several studies show variation in trust and trustworthiness across populations as well (Bohnet et al., 2008, Naef and Schupp, 2008). While these trust game studies are based on relatively small sample sizes, the advantage of our meta-analysis is that it is based on data collected from a much larger population and drawn from countries covering five distinct geographic areas. We document our data, which readers may also like to consult as a source of information about what has been done in the literature (see Appendices A and B).

Section snippets

Meta-analytic procedure

The basic purpose of a meta-analysis is to apply methodological rigor to a literature review through the use of statistical techniques (Glass, 1976, Stanley, 2001). By combining multiple studies, a meta-analysis can reduce the impact of sampling error and improve estimates of the effects on trust game behavior (Hunter, 1997). Meta-analyses are most commonly used to conduct a quantitative literature review of a research finding or “effect” that has been investigated in primary research under a

Analysis and results

Our measure of behavioral trust is the amount of money passed by the sender divided by endowment. Our measure of trustworthiness is the amount returned by the receiver as a proportion of the amount available to return. When measured this way, both trust and trustworthiness are proportions, falling between zero and one. In order to avoid inefficient coefficient estimates and an inappropriately specified model using OLS, we apply the logit transformation to both trust and trustworthy in order to

Discussion

Our results have a variety of important implications for the potential impact of methodological and geographic variations on trust game behavior. We find that when only one or ‘some’ out of a group of subjects are paid the actual stakes described in the experiment, participants pass significantly less money to their counterpart. Random payment schemes appear to be motivating participants to behave differently, possibly due to the added risk associated with the final payment (Bottom, 1998).

We

Conclusion

Given the increasing importance of experimental methods in economics, we must continue to test the boundary conditions and generalizability of the results generated by these powerful techniques. Meta-analyses are a useful tool to uncover what matters and what does not in experimental technique. That is not to say that the results we report necessarily invalidate any experiments, or even any specific experimental protocol. As pointed out by Falk and Heckman (2009), well designed lab studies

Acknowledgements

We gratefully acknowledge Omar Al-Ubaydli, William Bottom, Marco Castillo, Kurt Dirks, Dan Houser, Courtney LaFountain, Kevin McCabe, Judi McLean Parks, Gary Miller, Jackson Nickerson, Nathan Nunn, Ragan Petrie, Alex Tabarrok, seminar participants at ISNIE 2009, as well as Associate Editor Simon Gaechter, and two anonymous reviewers for constructive comments on earlier versions of this manuscript. We also thank Ayal Chen-Zion for his assistance with data collection, and the many individuals who

References (97)

  • S. Burks et al.

    Playing both roles in the trust game

    Journal of Economic Behavior and Organization

    (2003)
  • J. Carpenter et al.

    The effect of stakes in distribution experiment

    Economics Letters

    (2005)
  • M. Casari et al.

    The strategy method lowers measured trustworthy behavior

    Economics Letters

    (2009)
  • J.C. Cox

    How to identify trust and reciprocity

    Games and Economic Behavior

    (2004)
  • W. Güth et al.

    The relevance of equal splits in ultimatum games

    Games and Economic Behavior

    (2001)
  • E. Hoffman et al.

    Preferences, property rights, and anonymity in bargaining games

    Games and Economic Behavior

    (1994)
  • G. Jones

    Are smarter groups more cooperative? Evidence from prisoner’s dilemma experiments, 1959–2003

    Journal of Economic Behavior and Organization

    (2008)
  • R. Rosenthal

    The file drawer problem and tolerance for null results

    Psychological Bulletin

    (1979)
  • R. Slonim et al.

    Gender selection discrimination: Evidence from a trust game

    Journal of Economic Behavior and Organization

    (2010)
  • M. Sutter et al.

    Trust and trustworthiness across different age groups

    Games and Economic Behavior

    (2007)
  • K. Swope et al.

    Personality preferences in laboratory economics experiments

    The Journal of Socio-Economics

    (2008)
  • L.F. Ackert et al.

    An experimental examination of the effect of potential revelation of identity on satisfying obligations

    New Zealand Economic Papers

    (2011)
  • P. Aghion et al.

    Regulation and distrust

    Quarterly Journal of Economics

    (2010)
  • Akai, K., & Netzer, R. J. (2009). Trust and reciprocity among international groups: Experimental evidence from Austria...
  • K. Arrow

    Gifts and exchanges

    Philosophy and Public Affairs

    (1972)
  • N. Ashraf et al.

    Decomposing trust and trustworthiness

    Experimental Economics

    (2006)
  • N. Bardsley et al.

    Experimental economics: Rethinking the rules

    (2010)
  • Bellemare, C., & Kröger, S. (2003). On representative trust. Working...
  • R.J. Bies et al.

    Beyond distrust: ‘Getting even’ and the need for revenge”

  • H.P. Binswanger

    Attitude toward risk: Experimental measures in rural India”

    American Journal of Agricultural Economics

    (1980)
  • I. Bohnet et al.

    Betrayal aversion: Evidence from Brazil, China, Oman, Switzerland, Turkey, and the United States

    American Economic Review

    (2008)
  • Bohnet, I., & Cooter, R. D. (2005). Expressive law: Framing or equilibrium selection? John F. Kennedy School of...
  • W.P. Bottom et al.

    Building a pathway to cooperation: Negotiation and social exchange between principal and agent

    Administrative Science Quarterly

    (2006)
  • S. Bowles

    Policies designed for self-interested citizens may undermine “the moral sentiments”: Evidence from economic experiments

    Science

    (2008)
  • J. Brandts et al.

    Hot vs. cold: Sequential responses and preference stability in experimental games

    Experimental Economics

    (2000)
  • Brandts, J. & Charness, G. (2010). The strategy versus the direct-response method: A survey of experimental...
  • C. Camerer

    Behavioral game theory: Experiments in strategic interaction

    (2003)
  • C. Camerer et al.

    Experimental tests of a sequential reputation model

    Econometrica

    (1988)
  • L. Cameron

    Raising the stakes in the ultimatum game: Experimental evidence from Indonesia

    Economic Inquiry

    (1999)
  • A. Chaudhuri

    Experiments in economics: Playing fair with money

    (2009)
  • A. Chaudhuri

    Sustaining cooperation in laboratory public goods experiments: A selective survey of the literature

    Experimental Economics

    (2010)
  • J. Coleman

    Foundations of social theory

    (1990)
  • R. Croson et al.

    Gender differences in preferences

    Journal of Economic Literature

    (2009)
  • K.T. Dirks et al.

    Trust in leadership: Meta-analytic findings and implications for research and practice

    Journal of Applied Psychology

    (2002)
  • R. Dornbusch et al.

    Macro economics

    (1994)
  • J. Ermisch et al.

    Measure people’s trust

    Journal of the Royal Statistical Society

    (2009)
  • A. Falk et al.

    Lab experiments are a major source of knowledge in the social sciences

    Science

    (2009)
  • Falk, A., Meier, S., & Zehnder, C. (2011). Did we overestimate the role of social preferences? The case of...
  • Cited by (647)

    • Evolution of trust in structured populations

      2024, Applied Mathematics and Computation
    • Smartphone use decreases trustworthiness of strangers

      2024, Journal of Economic Psychology
    View all citing articles on Scopus
    View full text