Base-rate neglect and neural computations for subjective weight in probabilistic inference

Abstract

AO_SCPLOWBSTRACTC_SCPLOWHumans show systematic biases when estimating probability of uncertain events. Base-rate neglect is a well-known bias that describes the tendency to underweight information from the past relative to the present. In this study, we characterized base-rate neglect at the computational and neural implementation levels. At the computational level, we established that base-rate neglect arises from insufficient adjustment to weighting prior information in response to changes in prior variability. At the neural implementation level, we found that orbitofrontal cortex (OFC) and medial prefrontal cortex (mPFC) represent subjective weighting of information that reflects base-rate neglect. Critically, both subjective-weight and subjective-value signals that guide choice were found in mPFC. However, subjective-weight signals preceded subjective-value signals. These results indicate that when facing multiple sources of information, estimation bias such as base-rate neglect arises from information weighting computed in OFC and mPFC, which directly contributes to subjective-value computations that guide decisions under uncertainty.\n\nSO_SCPLOWIGNIFICANCEC_SCPLOW SO_SCPLOWTATEMENTC_SCPLOWFacing uncertainty, estimating the probability of different potential outcomes carries significant weight in affecting how we act and decide. Decades of research show that humans are prone to giving biased estimation but it remains elusive how these biases arise in the brain. We focus on base-rate neglect, a well-known bias in probability estimation and find that it is tightly associated with activity in the medial prefrontal cortex and orbitofrontal cortex. These regions represent the degree to which human participants weigh different sources of information, suggesting that base-rate neglect arises from information-weighting computations in the brain. As technology provides us the opportunity to seek and gather information at an ever-increasing pace, understanding information-weighting and its biases also carry important policy implications.