http://www3.interscience.wiley.com/cgi-bin/abstract/112726409/ABSTRACT 

 In the paper “Hierarchical Reporting, Aggregation, andInformation Cascades”, the authors (Anil Arya, Jonathan Glover and Brian Mittendorf ) illustrate how a cascade can decide the outcome of a decision in a hierarchical organization and how that can be prevented. In their model, a firm consists of 3 tiers – there are 3 lowest level managers, a middle manager, and a top manager. The firm needs to make a Yes/No decision (whether it should accept a project, etc.). Each of the 3 lowest level managers make private decisions (according to their own signals), which are forwarded to the middle manager. He updates his decision after considering their recommendations and his own signal. The top manager then makes the final decision after observing the decisions of the tiers below him.

The authors argue that rather than letting the middle and top manager have the knowledge of the votes of each lowest level manager, the lowest tier should submit an aggregated vote, i.e. a single decision representing the majority of the lowest tier. This prevents a cascade from forming as the middle and top managers, uncertain about whether the aggregated vote was mixed or unanimous, harbor a certain level of doubt about the recommendation. As a result, they rely more on their own discretion instead of merely rubber stamping the decision of the lower tiers.

By modifying their model, the authors investigated the consequences when the opinion of one of the lowest managers is deemed more valuable than those of the other two. This simulates the real-life situation in which the opinions of prominent voters who are regarded as more knowledgeable hold more weight. They found that an intermediate aggregation of information, in which the final decision makers are told of the number of votes, but not the identities of the voters, is the optimal solution to prevent excessive influence of a few individuals.

The practical application of their findings is that restriction of statistical information or identities of voters not only safeguards the privacy of individuals, but also prevents information cascades from forming during the decision making process.