The influence of initial beliefs on judgments of probability

• Main Authors: Erica Catherine Yu, David Albert Lagnado
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2012-10-01
• Series: Frontiers in Psychology
• Subjects: Gambling; causality; belief revision; model-based learning; priors; Probability judgment
• Online Access: http://journal.frontiersin.org/Journal/10.3389/fpsyg.2012.00381/full

This study aims to investigate whether experimentally-induced prior beliefs affect processing of evidence including the updating of beliefs under uncertainty about the unknown probabilities of outcomes and the structural, outcome-generating nature of the environment. Participants played a gambling task in the form of computer-simulated slot machines and were given information about the slot machines' possible outcomes without their associated probabilities. One group was induced with a prior belief about the outcome space that matched the space of actual outcomes to be sampled; the other group was induced with a skewed prior belief that included the actual outcomes and also fictional higher outcomes. In reality, however, all participants sampled evidence from the same underlying outcome distribution, regardless of priors given. Before and during sampling, participants expressed their beliefs about the outcome distribution (values and probabilities). Evaluation of those subjective probability distributions suggests that all participants' judgments converged toward the observed outcome distribution. However, despite observing no supporting evidence for fictional outcomes, a significant proportion of participants in the skewed priors condition expected them in the future. A probe of the participants' understanding of the underlying outcome-generating processes indicated that participants’ judgments were based on the information given in the induced priors and consequently, a significant proportion of participants in the skewed condition believed the slot machines were not games of chance while participants in the control condition believed the machines generated outcomes at random. Beyond Bayesian or heuristic belief updating, priors not only contribute to belief revision but also affect one's deeper understanding of the environment.