The Impact of Experts and Error in Observation on Informational Cascades

Session Number

F08

Advisor(s)

Randall Berry, Northwestern University

Location

A-115

Start Date

28-4-2016 11:05 AM

End Date

28-4-2016 11:30 AM

Abstract

Models are often used to analyze observational learning. Many of which study a decision making process in which Bayesian agents make binary decisions based upon observations of previous agents. These observations may mislead an agent from their initial decision causing what is known as an Informational Cascade. Formally, an Information Cascade refers to a time in which every subsequent agent relinquishes their own private information and follows the previous agents. The model discussed in this investigation explores the effect of a more educated population, called experts, and error in an agent’s observation, defined as noise. Specifically, our main focus was studying the probability that agent’s would cascade correctly. The investigation showed a non-monotonic relationship between the probability of a correct cascade and noise. This relationship was also characterized by spikes, later explained throughout the investigation. The relationship between expert concentration and the probability of a correct cascade also showed a non-monotonic relationship. Overall, we presented a new model studying informational cascades with respect to noise and expert concentration. With this, we disproved the notion that a lower noise and higher expert concentration increases the probability of a correct cascade.


Share

COinS
 
Apr 28th, 11:05 AM Apr 28th, 11:30 AM

The Impact of Experts and Error in Observation on Informational Cascades

A-115

Models are often used to analyze observational learning. Many of which study a decision making process in which Bayesian agents make binary decisions based upon observations of previous agents. These observations may mislead an agent from their initial decision causing what is known as an Informational Cascade. Formally, an Information Cascade refers to a time in which every subsequent agent relinquishes their own private information and follows the previous agents. The model discussed in this investigation explores the effect of a more educated population, called experts, and error in an agent’s observation, defined as noise. Specifically, our main focus was studying the probability that agent’s would cascade correctly. The investigation showed a non-monotonic relationship between the probability of a correct cascade and noise. This relationship was also characterized by spikes, later explained throughout the investigation. The relationship between expert concentration and the probability of a correct cascade also showed a non-monotonic relationship. Overall, we presented a new model studying informational cascades with respect to noise and expert concentration. With this, we disproved the notion that a lower noise and higher expert concentration increases the probability of a correct cascade.