Learning and Contagion in Teams
By David J. Cooper & John H. Kagel
Extended Abstract:There have been many economic experiments comparing the choices of individuals and teams. For settings, like many games, where there is an unambiguously optimal choice that can be reached via a reasoning process, the usual finding is that teams outperform individuals (e.g. Cooper and Kagel 2005, 2009, and 2016; Kocher and Sutter, 2005; and Feri et al, 2009. However there also exist several studies where teams fail to outperform individuals (e.g. Casari et al, 2016; Charness et al, 2016) and beating the demanding “truth wins” benchmark is unusual (Cooper and Kagel, 2005).
All of these studies ignore a fundamental fact of organizational life: things change. In particular, membership in groups is not fixed. Employees come and go as they are hired, fired, or quit. Governments are voted into and then out of power. New team members can bring new ideas and innovation. But they can also disrupt existing norms and communication patterns (Weber and Camerer, 2003). Even if a new team member knows a valuable idea, previous research (Cooper and Kagel, 2016) suggests that they are likely to fail at transmitting this idea to their new teammates. The central question of our paper is whether changing team membership will accelerate learning in games via new teammates transmitting valuable insights gained from their old teams.
We study learning inZeck, a sequential game that has an optimal (winning) strategy for the first player. Learning the optimal strategy is analogous to solving a “eureka” type logic problem, as it requires grasping an insight that is hard to learn but, once learned, can easily be explained to other people. Individuals only learn slowly to use the optimal strategy, and two subject teams with fixed members perform no better than individuals. Rematching teammates into new two subject teams immediately results in a large and persistent increase in use of both the optimal first move and, more generally, the optimal strategy. This increase can easily be seen in the figure shown at the end of this extended abstract. The graph shows the proportion of individuals or teams, depending on the treatment, using the optimal first move. The switch in teammates takes place between Stage 1 and Stage 2 for the Teams, Switching treatment. There is an immediate sharp jump in use of the optimal first move. The results are similar for use of the optimal strategy, although the response is somewhat slower.
The spike in use of the optimal strategy is driven by “mixed” teams, where one teammate was previously in a team that did not play optimally and one was in a team that did play optimally. These mixed teams disproportionately adopt the optimal strategy, consistent with a truth wins model. (A truth win model posits that a team will play optimally if any of its members would have played optimally when acting as individuals.) This is surprising, as the truth wins model performs quite poorly for teams with fixed membership. Underlying the transmission of optimal play when team membership is switched, there is a burst of communication when the switch takes place. This includes both an uptick in discussing what strategy to use and frequent attempts to determine which teammate has performed better in the past. The latter suggests that successful transmission is at least as much due to the less able teammate deferring to the more able teammate as actualtransmission of ideas between teammates.
Our results have important implications for settings where decision makers are groups rather than individuals, as reshuffling team membership can be an important source of learning how to play strategically. Studies comparing fixed teams with individuals have yielded mixed results about the performance of teams relative to individuals (as does this one), but this systematically underestimates the advantage of teams over individuals. More generally, changes in team membership can be extremely valuable for organizationsas a mechanism by which new ideas, strategies, and innovations can enter the group.