In the trust game (see previous posting “Cheap? Maybe. Irrelevant? No!”), two players each have $10, the “first mover” can send the “second mover” any or none of his dollars, the amount sent is tripled, and the second mover can return any or none of this tripled amount. Outcomes include the “selfish/rational equilibrium” of zero trust, zero cooperation in which each player earns only her initial $10, the “fair and efficient equilibrium” of full trust and full reciprocity in which each earns $20, and the “sucker’s equilibrium” in which the first mover is trusting but the second is selfish, with payoffs of $0 vs. $40.
Greeter and prospective subject in lab entry room
Stephen Atlas and I modified the trust game slightly and let anonymous subjects play it in their avatar forms in the virtual world Second Life. Instead of $10, each player was given 100 units of the Second Life currency, Linden dollars. They arrived at the virtual lab individually, with a time delay to assure that no player could learn the identity of his or her counterpart. The first player to arrive was automatically assigned the “first mover” role, the next one to arrive after a sufficient delay was assigned that of “second mover.” A total of 309 subjects learned of the game in the virtual world through a message board, word of mouth, or by noticing the sign on the front of the lab, and these subjects spent an average of around twenty minutes to play the game and to answer a set of survey questions with the hope of earning some currency (useable “within world” or exchangeable for real cash).
How did Second Life denizens play the trust game? First movers chose to send an average of 52 of their 100 Lindens to their anonymous counterparts (with 94% sending at least some money), and the average second mover who received some money sent back 47% of it (26 Lindens). (Returning anything above 33.3% rewards the first mover’s trust.) Despite the prediction of traditional economic theory based on pure self-interest and rationality, second movers typically returned enough money so that it was profitable for a first mover to make the trusting move and send money.
Atlas and Putterman's lab in Second Life
We used the malleability of the virtual setting to investigate how subjects might be influenced by ostensibly irrelevant aspects of their environments. Without knowing of the differences, some subjects’ avatars were sent to a room with a framed photograph of cuddly giant pandas on the wall, others to a room with a framed photograph of a fish being filleted by a chef, and a third set were sent to a room in which no photo was displayed. All were shown the same instructions for the game, with no mention of the photos. According to rational choice theory, presence or absence of photos should have had no effect on decisions. But in actuality, first movers who saw pandas sent an average of 70 of their 100 Lindens, whereas ones who saw fish on plate with chef’s knife sent about 49 Lindens, about the same as those in the no photo condition. Second movers returned more the more that they had been sent by their counterpart, but there was no additional influence of photos or of their absence on second mover behavior. Panda people and their counterparts earned some 43% more money on average than did fish and no photo treatment subjects.
When most people are fair minded or reciprocating, as were most second movers in our game, subjects can earn more by acting in a trusting manner towards their unknown counterparts. Our experiment suggested that subtle cues in the environment that should be irrelevant from the standpoint of rationality can have a substantial effect on the inclination to trust. Cuddly and cute set off feelings of safety, while cold, dead, and in the process of being gutted, might reinforce feelings of concern in some (though one colleague told me it just made him feel hungry). Perhaps most interesting is that the experiment demonstrated the power of cuddly and cute in the distinctly non-cuddly medium of an imaginary, virtual world.
In two other treatments, there were no photos but instead a few words were added in the instructions. In one of these treatments, the added words were: “By working together and acting fairly, both participants can double their money.” In the other, they were “assuming that [the] second person wants to earn as much as possible, he/she will send none of the money received back to the first person. So if the first person sends L$100 and the second person keeps the largest amount of money, he/she will earn L$400 and the first person will earn zero.” Subjects exposed to the first wording sent an average of 59 of their 100 Lindens, whereas those exposed to the second wording sent an average of 33.
Like the pictures, the words should have had no impact, since the rules were the same for all groups and it shouldn’t have been difficult for participants to realize what kinds of outcomes were possible. Yet spelling out the more pessimistic, “selfishly rational” course of action for a second mover made first movers significantly more wary. The suggestive power of words was as great as that of pictures, but the difference appeared in this case when the words suggested a note of caution, not optimism.
A conjectural explanation of the fact that words mentioning cooperation had no ostensible effect while ones warning of “being stiffed” altered behaviors is that the possibility of cooperation occurred naturally to most participants, whereas many weren’t particularly drawn to the thought of potential treachery until explicitly reminded of it. Both the high average proportion returned by second movers and the large amounts sent by first movers, implying trusting default beliefs, indicate a baseline level of sociability among many players. The stark economic hypothesis of strict rationality and self-interest appears not to have been the default expectation of most, though a substantial number could be nudged towards it by explicit suggestion. Subjects’ behaviors thus suggest that expectations of a modicum of sociability are a more reasonable starting hypothesis in the analysis of human behavior than is the starker version of the theory of self-interest found in the economics textbooks of yesteryear.