My first essay reexamines experimental results suggesting that pooling of information by decision making groups is often incomplete, leading to a suboptimal decision. In my main experimental treatment, I calibrate decision makers' information load so that discovering the optimal choice is likely, given natural memory constraints. However, I show that errors recalling public information are typically mitigated by within-group correction, while comparable errors with private information cannot be corrected, biasing the pool of information toward the option favored by public information. This "corrective force bias" accounts for the majority of sub-optimal group decisions in my experiment. This bias is unacknowledged in prior experiments, and suggests that the observed information pooling failures have more to do with the structure of the information distribution than any inadequacies in group performance.
In my second essay, I provide a theoretical model which illustrates why experts advising a decision maker (DM) with a limited ability to process information might rationally bias their messages toward the policy they believe to be correct when the advisors' and DM's interests are aligned. In the most informative perfect Bayesian equilibrium of a game where advice is given sequentially, the first advisor sends a sincere message if his information is moderate. However, given information sufficiently favorable to a policy, advisor 1 biases his message in an attempt to ensure this policy is chosen even when the DM receives a noisy message. I describe this bias in the transmission of information as "paternalistic bias." My research suggests that the ex-post evaluation of the precision of experts should account for this paternalistic bias.
My third essay describes a rational choice model for a behavioral phenomenon known as biased recall, i.e., the fact that information which is consistent with an initial preference, expectation, or belief is more likely to be recalled. The model I provide is the first to endogenize the recall of information and ascribe utility functions to agents in order to solve for the optimal allocation of limited memory resources.