Review of Economic Studies (2006) 73, 357–380 0034-6527/06/00130357$02.00c2006 The Review of Economic Studies LimitedSocial Learning and Norms in aPublic Goods Experiment withInter-Generational Advice1ANANISH CHAUDHURIUniversity of AucklandSARA GRAZIANOCharles River AssociatesandPUSHKAR MAITRAMonash UniversityFirst version received May 2003; final version accepted October 2005 (Eds.)We study a linear public goods game using an inter-generational approach. Subjects in one gen-eration leave advice for the succeeding generation via free-form messages. Such advice can be privateknowledge (advice left by one player in generation t is given only to his or her immediate successor ingeneration t+ 1), public knowledge (advice left by players of generation t is made available to all mem-bers of generation t + 1), and common knowledge (where the advice is not only public but is also readaloud by the experimenter). Common knowledge of advice generates a process of social learning thatleads to high contributions and less free-riding. This behaviour is sustained by advice that is generallyexhortative, suggesting high contributions, which in turn creates optimistic beliefs among subjects aboutothers’ contributions. We suggest that socially connected communities may achieve high contributions toa public good even in the absence of punishment for norm violators.1. INTRODUCTIONThere is now a voluminous experimental literature that uses voluntary contributions mechanismsto capture the tension between contributing to a public good or free-riding on others’ contribu-tions. Ledyard (1995) provides a review of much of this literature. Prior studies have documentedthat (1) in a one-shot version of the public goods game groups of subjects on average contributebetween 40% and 60% of the optimal level with wide variations in individual contributions rang-ing from 100% contribution by some to 0% by others and (2) if the players interact repeatedlyover a number of rounds then contributions often start out at between 40% and 60% of the socialoptimum and decline steadily over time as more and more players choose to “free-ride”. Suchfree-riding has been the subject of intense study (see, for example, Andreoni, 1988; or Andreoniand Croson, 1998).The linear public goods game is an excellent vehicle for understanding the inherent ten-sion between cooperative and competitive behaviour in social dilemmas. The linear public goods1. A longer version of this paper containing detailed discussions about the implications of our results is avail-able from the Social Science Research Network at http://ssrn.com/abstract=580481. The instructions for the exper-iments, the original data, and the STATA program used to analyse the data are available on the journal’s websitehttp://www.restud.org.uk. All errors are the responsibility of the authors.357358 REVIEW OF ECONOMIC STUDIESgame is really an n-person prisoner’s dilemma where free-riding is the dominant strategy Nashequilibrium in a one-shot version of the game and also the subgame perfect outcome in finitelyrepeated versions of the game. It is also the evolutionarily stable strategy in such situations.See the discussion in Miller and Andreoni (1991) among others. As is well documented in theliterature, however, while free-riding does occur in finitely repeated versions of such games,the game-theoretic prediction of complete free-riding is clearly refuted even after a number ofrounds.In this paper we develop an inter-generational approach to a linear public goods game.A group of subjects is recruited into the laboratory and play the public goods game for 10 rounds(the exact experimental design and parameters are explained in Section 2). After his or her par-ticipation is over, each player is replaced by another player, his or her laboratory descendant,who then plays the game for another 10 rounds as a member of a fresh group of subjects. Thegenerations are non-overlapping. Advice from a member of one generation can be passed alongto his or her successor via free-form messages that generation t players leave for their generationt + 1 successors. Pay-offs span generations in the sense that the pay-off to a generation t playeris equal to what he or she has earned during his or her lifetime plus 50% of what his or her labo-ratory descendant earns. This provides an incentive to the subjects to pass on meaningful adviceto their successors.This inter-generational approach was pioneered by Schotter and Sopher (2001a,b, 2003),who point out that in real life people approach such social dilemmas in a manner that is differentfrom those captured in previous experiments. When confronted with such situations, they haveaccess to the wisdom of the past, in the sense that those who have faced the same dilemma beforethem (or at least immediately before them) are there to give them advice as to how to addressthe problem. Our conjecture is that playing a public goods game using such an inter-generationaldesign will, over time, lead to the evolution of norms of cooperation, with later generations notonly achieving higher levels of contribution but also managing to mitigate problems of free-riding. Norms or conventions of behaviour that arise during one generation may be passed onto the successors. Hence, we expect that outcomes in our inter-generational game will be moreefficient than those reported in previous experimental work.We believe that the concept of an evolutionarily stable strategy does not adequately capturethe way in which social evolution, as opposed to biological evolution, might function. This is pri-marily because the concept of evolutionarily stable strategies does not allow for social learning,and therefore it might not be able to explain large patterns of human cooperation. To do so weneed theories of cultural evolution or gene culture co-evolution along the lines of Cavalli-Sforzaand Feldman (1981) and Boyd and Richerson (1985). Human societies create and pass on socialconventions from one generation to the next and the inter-generational framework that we use inthis paper is an attempt to capture the evolution of such social norms.We incorporate three separate mechanisms for passing advice from one generation to thenext. The first is the “private knowledge of advice” treatment, where advice from one subject ingeneration t is given to his or her immediate successor in generation t + 1. The second is the“public knowledge of advice” treatment where