September 2023. [Working on major changes!]
Abstract
This paper presents a simple model of memory-based belief formation. An individual has a memory of signals, and upon receiving a prompt, recalls signals from his memory and forms belief using a naïve Bayes' formula. The recall probability of each signal depends on the cues of the signals and the prompt. The recall probability function can capture different stylized results of memory research, including how similarity, interference, salience, previous retrieval, etc. affect memory retrieval. I analyze three adaptations of the model and show that, first, varying recall probabilities w.r.t. likelihood ratio can lead to asymptotic erroneous belief. Second, the persuader prefers to send a less informative but more associative signal when the receiver has a large memory. Third, individuals overreact to new signals when they have a large memory, and will strengthen their strong prior beliefs even after receiving a belief-challenging signal. Paper
