## Information Cascade Experiment by the Virginia University

This is a lab report by Virginia University to explore how people follow the information cascade in reality. It starts to offer a definition of “information cascade” that “when the information implied by early decisions outweighs any one person’s private information” (Anderson and Holt, 1). More specifically, the third person’s decision depends on the matched first two predictions but not his personal information, and the fourth person continues the pattern. The process is fairly similar to the red ball majority or blue ball majority guessing game that we go over in the Network class. The experience with three labeled balls “a” “a”, “b” or “b” “b”, “a” in two urns labeled A and B. The chance of getting the two runs is identical. And the experiment is run to test whether information cascade works and a payoff of \$2 is used as an incentive. Most of the time the information cascade occurred as predicted but there are so other interesting findings. First, a third person made a B prediction when he draws a “b” signal while there are two A decisions before him. Second, a second player predicted A when he actually sees a “b”, “which is inconsistent with Bayes’ rule and private information, is probably the result of confusion or carelessness.” (pg. 4). Overall, this type of inconsistency in about one-fourth of the cases where the optimal Bayesian decision differed from the decision implied by private information, and it’s not that uncommon.

The result of the experiment reflects how the model works in reality. There are some interesting new lights about information cascade that I have learned from this paper. First, we always talk about how the second person will choose his private information because the probability is a tie in class. There is a real-life example that person acts against this predicted behavior; as a result, we need to be careful how accurate the model is and accounts for its inconsistency. Second, the designer of the experiment uses \$2 as motivation but this motivation may not be strong enough. In our class example, we go up to \$1,000 so every player will try their best to get the right answer. The authors explain how some mistakes are caused by confusion or carelessness of this game. Its counterargument is that “increasing payoffs from \$0 to \$2 resulted in a decrease in the number of errors, but the increase from \$2 to \$4 had no significant effect on errors”(pg.6). But in fact, \$2 and \$4 is still in a similar price range and I wonder what will happen if the price is raised at least ten times higher. The main takeaway is that although there is information cascade, people are complicated and behaviors are still somehow unpredictable.

Citation:

Anderson, Lisa R., and Holt, Charles A. Information Cascade Experiments. Virginia University, 2 Aug. 2018, www.people.virginia.edu/~cah2k/cashbktr.pdf.