Summary | Full Article (JSTOR)

This article shows how Information Cascades can be demonstrated in a very controlled setting. The example is in a classroom with 3 urns, 3 blue marbles and 3 red marbles. The teacher puts 2 blue marbles and 1 red marble in the first urn (urn A) and 1 blue marble and 2 red marbles in the second urn (urn B). Then the teacher randomly selects one of the first 2 urns (by flip of a coin, for instance) and dumps it into the third urn without the students knowing which urn was chosen. Students are then allowed, one by one, to draw a marble from the third urn with replacement, and then predict which of the urns A or B was chosen and announce it to the rest of the class. Though the students knew the contents of urns A and B, they don’t know which urn was chosen to be dumped into the urn they are choosing from. Nor do the other students in the class know what color the marble is that each student draws, they only know the prediction that student makes.

After following these rules, it tends to be the case that Information Cascading occurs and after the first 2 students, and the subsequent predictions tend to agree with the initial predictions regardless of the color of the marble drawn. Without information sharing (students don’t announce their predictions), this result would be surprising, as if the student picks a blue marble it is twice as likely that the chosen urn was urn A. With information sharing (students do announce their predictions), Information Cascading begins to come into play when additional predictions add no new information to the subsequent drawers. For example, if the first 2 predictions are for urn A, no matter which color of marble the third student draws, he will choose urn A because the first 2 predictions will outweigh the new information from drawing the marble. Therefore, subsequent students do not gain any new information because they cannot infer whether the third student drew a red or blue marble. Information Cascading does not always arise. According to the article, cascades happen only about 80% of the time.