[Author Affiliation]Stephen Atlas, , , , saa2140@columbia.eduLouis Putterman, , , , Louis_Putterman@brown.edu; corresponding author[Acknowledgment]The authors wish to thank Enrico Spolaore and David Garman for their support of and comments on Atlas's master's thesis, and Vero Testa for assistance with a trust game literature review. We are grateful for expertise provided by Stefan Bornhofen, who coded the experimental apparatus in Second Life using Linden Scripting Language, and William D. Hascall, who built the lobby and treatment rooms.1. IntroductionIn recent decades, laboratory decision-making experiments have become a standard part of the economist's toolkit for testing theories of economic behavior, with the majority of experiments being conducted in university computer labs using students as subjects. While some results have been replicated using more diverse subject pools, and while many replications suggest that inferences from student behaviors are not necessarily misleading, it is widely agreed that including more broadly representative subject pools in experiments would be desirable.1One reason most experiments continue to study students is that a properly incentivized experiment is costly, and most students already on a university campus have far lower opportunity costs of getting to an on-campus lab and participating in an experiment there than do individuals engaged in full-time employment in other locations. For experimenters, too, the costs of setting up experiments in other locations can be high relative to conducting them on campus. Thus, experimenting in university labs with student subjects carries substantially lower monetary and logistical costs.The possibility of reaching a larger and broader subject population with modest logistical costs and cost per subject sparked our interest in conducting an exploratory experiment in a virtual world. Participants in the life-like on-line platform Second Life come from a broad demographic spectrum, and it has been estimated that about 30,000 people are engaged in some activity in Second Life (including holding paid jobs, selling products, furnishing virtual homes, and socializing) at any given moment. We wanted to investigate the potential of this medium and population as venues for conducting economics experiments with diverse participants at reasonable cost.The experiment we chose is the Trust Game introduced by Berg, Dickhaut, and McCabe (1995, hereafter BDM) and subsequently replicated, with or without modification, in many settings (a partial survey is provided by Camerer 2003). One reason for this choice is the game's simplicity: Only two players need be matched, each makes a single decision, and the choices can be consecutive rather than simultaneous, so that no subject group as such need be assembled. Another reason is the game's intrinsic interest as a prominent illustration of how rational calculations intersect with social dispositions, and as an important tool for studying a problem lying at the heart of much economic interaction (Arrow 1974). How the social element in the trust game will play out in the anonymity of the virtual world is a matter of both theoretical and practical interest, the latter owing to the fact that trust in e-commerce and virtual environments has commercial and other applications (Benbasat, Gefen, and Pavlou 2008; Chesney, Chuah, and Hoffmann 2009a).In addition to conducting a standard trust game in order to compare results of an otherwise similar protocol with different environments and subject pools, we decided to experiment with our virtual lab by varying two dimensions and performing exercises in comparative statics. First, we varied the physical environment by placing one of two photographs, or none, on the virtual lab room's wall, following a recent psychology experiment suggesting strong susceptibility of social behaviors to subtle but economically irrelevant cues (Bateson, Nettle, and Roberts 2006). …