For instructors
Teaching notes
Two airlines on the same route would both earn more with high fares, yet each gains by undercutting the other. Members of a cartel agree on quotas and each is tempted to exceed its own. In both cases the outcome that is best for the pair is not what either side would choose alone.
What the game shows
The game tests how much cooperation survives when defection is individually profitable, and what repetition, a known final round, communication and anonymity do to it. With the presets for the stag hunt and for chicken, the same screen also shows coordination problems, in which the difficulty is trust or precedence and no dominant strategy exists.
The theory
In a single round defection is a dominant strategy, and mutual defection is the unique Nash equilibrium although both players prefer mutual cooperation. The first experiment with this structure was run at RAND in 1950 and is reported in Flood (1958). When the number of rounds is known, backward induction extends the argument to every round, since defection in the last round removes the reason to cooperate in the round before.
Two lines of theory explain why cooperation is nevertheless common. Kreps et al. (1982) show that a small doubt about whether the partner is strictly self-interested makes it rational to cooperate until close to the end. When the horizon is indefinite, cooperation can be sustained as an equilibrium by strategies that answer defection with defection, provided that the future matters enough, and Axelrod and Hamilton (1981) made reciprocal strategies of this kind widely known. In the stag hunt both common choices are equilibria, one better for both players and the other safer, a tension studied by Cooper et al. (1990) and by Battalio, Samuelson and Van Huyck (2001).
What to expect
Across the experiments reviewed by Sally (1995), participants cooperated in a little under half of their choices, and the opportunity to talk raised cooperation substantially. Andreoni and Miller (1993) found that players who kept the same partner for ten rounds cooperated significantly more than players who met a new partner in every round, which is what reputation building predicts.
Within a game of known length, cooperation is common in the early rounds and collapses near the end. Embrey, Fréchette and Yuksel (2018) show that the payoffs and the horizon determine how much cooperation forms at the start, and that with experience the first defection moves earlier, although this unraveling can be very slow. For indefinite horizons, Dal Bó and Fréchette (2011) report that cooperation fell to about 5 percent where it was not an equilibrium, and that it rose on average, though not in every session, where it was an equilibrium and also risk dominant.
Tips and things to look out for
- The most informative comparison is between a session with fixed partners and one with a new partner in every round. The difference between the two measures what reputation contributes.
- Watch the last two rounds of a session with fixed partners. A fall in cooperation there, after stable cooperation earlier, is the end-game effect that backward induction predicts for every round.
- The number of rounds can be announced to students or kept from them. When it is announced, the end-game effect that backward induction predicts can appear in the last rounds. When it is kept hidden, students know only that the game will end at some point, and a comparison between a section told the number and a section not told it shows how much the known horizon contributes to the decline. To approximate an indefinite horizon more closely, keep the number hidden, plan more rounds than you intend to play, tell the class that the session ends after any round with a stated probability, and stop by a visible random draw.
- Chat can be switched on from the host page in the middle of a session. A few rounds without it followed by a few rounds with it show the effect of communication within the same pairs.
- Partners remain anonymous unless you choose to reveal names. Revealing them brings in reputation outside the game, which raises cooperation for reasons that the model does not contain.
- A leaderboard invites students to compare themselves with their partner. A student who wants to finish ahead of the partner, and not to earn as much as possible, faces a different game in which defection is even more attractive. Every student screen therefore states that the aim is the largest possible total of one's own points over all rounds and that comparisons with others do not count. Repeating this aloud before play, or hiding the names on the leaderboard, further reduces the confusion.
- For the debrief, ask each pair that sustained cooperation what would have happened after a single defection. Their answers usually describe a trigger strategy without naming it.
References
- Andreoni, J., & Miller, J. H. (1993). Rational cooperation in the finitely repeated prisoner's dilemma: Experimental evidence. The Economic Journal, 103(418), 570–585.
- Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211(4489), 1390–1396.
- Battalio, R., Samuelson, L., & Van Huyck, J. (2001). Optimization incentives and coordination failure in laboratory stag hunt games. Econometrica, 69(3), 749–764.
- Cooper, R. W., DeJong, D. V., Forsythe, R., & Ross, T. W. (1990). Selection criteria in coordination games: Some experimental results. American Economic Review, 80(1), 218–233.
- Dal Bó, P., & Fréchette, G. R. (2011). The evolution of cooperation in infinitely repeated games: Experimental evidence. American Economic Review, 101(1), 411–429.
- Embrey, M., Fréchette, G. R., & Yuksel, S. (2018). Cooperation in the finitely repeated prisoner's dilemma. The Quarterly Journal of Economics, 133(1), 509–551.
- Flood, M. M. (1958). Some experimental games. Management Science, 5(1), 5–26.
- Kreps, D. M., Milgrom, P., Roberts, J., & Wilson, R. (1982). Rational cooperation in the finitely repeated prisoners' dilemma. Journal of Economic Theory, 27(2), 245–252.
- Sally, D. (1995). Conversation and cooperation in social dilemmas: A meta-analysis of experiments from 1958 to 1992. Rationality and Society, 7(1), 58–92.