Ten Isn’t Large! Group Size and Coordination in a Large-Scale Experiment

Economic activities typically involve coordination among a large number of agents. These agents have to anticipate what other agents think before making their own decisions. Agents may fail to arrive at the best outcome simply because they believe that the others will fail to reach the same outcome. These situations are not uncommon: bank runs, liquidity runs in financial markets or currency attacks may all result from depositors and investors simply believing that others will start a run or attack. Bad coordination becomes even worse if agents start believing information that does not provide an objective financial valuation of assets—what we call non-fundamental information—and synchronize their decisions accordingly.

These coordination issues have been studied only in small groups because, according to theory, the size of a group (if more than two agents) does not influence coordination. This paper challenges this view and compares coordination among small groups of 10 agents with that among large groups of more than 80 agents. These large groups could proxy real-world markets and organizations. We use a bank run game with common signals—the non-fundamental information—that agents could choose to follow when deciding whether to run on the bank.

We find that small groups do not provide information about the behaviours of much larger groups when uncertainty about others’ behaviours is high. Small groups coordinate on a wide range of factors, including the best outcome and the non-fundamental information. In contrast, large groups systematically coordinate on the safest but not necessarily the best outcome and ignore non-fundamental information. This finding calls into question the relevance of non-fundamental information in theoretical models. Our results have both theoretical and experimental implications.