For instructors

Teaching notes

A visitor to an unfamiliar town chooses the fuller of two restaurants. An investor buys because earlier investors bought, and a manager adopts a technology because competitors have adopted it. Each follows the choices of predecessors on the reasonable view that those choices reflect information.

What the game shows

The game tests whether people weigh their own private information correctly against what the public actions of others reveal. It shows that imitation can be rational for each individual and still leave the group poorly informed, because a person who imitates adds nothing to what later decision makers can learn. Students also see the difference between what others know and what others do, which is the source of the problem.

The theory

The model is due to Bikhchandani, Hirshleifer and Welch (1992) and, in a related form, to Banerjee (1992). Each person receives a private signal that is correct with a known probability and observes the choices of all predecessors. Once the choices observed so far outweigh any single signal, a rational person follows them whatever the own signal says. From that point every later choice is uninformative, and the cascade continues even though the group as a whole holds enough information to identify the truth.

With the urns used here a draw is correct with probability two thirds, and a person who is indifferent is taken to follow the own draw. A cascade then starts as soon as two more informative predictions favor one urn than the other. The probability that a long sequence settles on the wrong urn is one fifth, which follows from comparing the chance that the first two draws are both wrong, one ninth, with the chance that they are both right, four ninths. Since a cascade rests on very little evidence, it is also fragile, and a small amount of new public information can reverse it.

What to expect

Anderson and Holt (1997) ran the design used here with six participants in each sequence. Cascades formed in 41 of the 56 rounds in which the early predictions made one possible, and cascades on the wrong urn occurred about half as often as cascades on the right one.

People nevertheless rely on their own information more than the theory prescribes. In the combined data of thirteen experiments, Weizsäcker (2010) finds that participants who ought to contradict their own signal do so in fewer than half of the cases. One consequence, documented by Goeree et al. (2007) in sequences of twenty and forty decisions, is that cascades are often broken and that mistaken cascades tend to be corrected over time.

Tips and things to look out for

References