Bayesian Learning in Social Networks

The short story is that this is a model of "herd" behavior, where observing enough (wrong) decisions by other before will make you choose the wrong thing as well, even if you know better. The next person after you will do the same, resulting in an "information cascade" of wrong decisions, even though everyone "knows better." The paper studies theoretical conditions for whether such information cascades happen or not.

The slightly longer story about these types of models is that you have a bunch of agents that decide A or B one after another. It is assumed that either A or B is "better." Each agent gets to see the decisions of some subset of the previous agents. Each agent also has their own belief. The agent will then make a decision based on observing the decisions of others (assuming something about their decision making process to try to estimate what might be the true answer).

The punchline is that even if your personal belief is biased toward the better answer, if enough people make the wrong decision by chance, everyone after them also will, despite knowing better, because it turns out to be optimal.