Skip to main content



Self-Correcting Information Cascades

Link: Self-Correcting Information Cascades

In class we studied information cascades, using the urn experiment to show how the choice the first two people make can determine how the rest of participants will make their decisions. Particularly, if the first two people drew the same marble color and thus guessed that the urn was also that color, then everyone else would also guess the same color, regardless of what their private information was (the marble they draw). This model however, is theoretical, and assumes that all participants in the experiment have perfect knowledge. Humans, of course, are prone to making errors, and this can have effects on what we see when the urn experiment and other like it are performed, especially if they are allowed to proceed for long enough. From a theoretical view, we would expect to an information cascade to always occur if given enough time. This paper examines the errors that humans have in their thinking, using a concept called the Quantal Response Equilibrium (QRE). It is similar to a Nash equilibrium, but it accounts for human error. In this view, players will tend to make choices with high payoffs, but not necessarily the highest payoffs.

The paper is rather technical, but the results of the experiment were clear.  Using the concept of QRE, the researchers were able to explain why multiple temporary cascades often occur when the urn experiment or others are like are performed for longer periods of time. Single cascades were almost entirely nonexistent for extended experiments. Not surprisingly, cascades are most often broken when a participant received a signal contrary to what the cascade suggested. What is surprising is that participants sometimes still went against what the cascade said even when they received the same signal. These temporary cascades allowed for self-correction too, in that once a wrong cascade is broken another can start again where the majority guess is actually correct. Using QRE as a model, it was seen that participants either overweighted the significance of their own signal or doubted the signs of those before them. The breaking of cascades also allows for new information to be gained for the next participants, and this allows for self-correction of a cascade. It is possible then that if an experiment is allowed to proceed for long enough, the public belief will end up being correct more often than not.

Theory and experiment tend to differ from each other, and this paper shows us another example of this. I imagine that similar phenomenon occur in other games, and the study of human error in game theory definitely has useful applications.

Comments

Leave a Reply

Blogging Calendar

November 2018
M T W T F S S
 1234
567891011
12131415161718
19202122232425
2627282930  

Archives