Decomposing Belief Updating: Likelihoods and Posteriors

How to cite

Saeed, Tashfeen. 2026. "Decomposing Belief Updating: Likelihoods and Posteriors." AEA RCT Registry. AEARCTR-0017734. https://doi.org/10.1257/rct.17734-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Tashfeen Saeed*

* Primary investigator

Registry ID
AEARCTR-0017734
DOI
https://doi.org/10.1257/rct.17734-1.0
First registered
Last updated
Status
In development
Trial dates
2026-01-26 – 2026-04-30
Countries
Private
JEL codes
D83, C91
Registry record
http://www.socialscienceregistry.org/trials/17734

Abstract

This study proposes an experiment that elicits both likelihood judgments and posterior beliefs in a belief-updating task with a rich information structure. By directly measuring each phase of Bayesian updating, the experiment identifies whether deviations from Bayesian benchmarks arise from errors in likelihood assessment, Bayesian integration, or both. The study examines two questions: (i) whether eliciting or providing likelihoods improves posterior accuracy, and (ii) whether eliciting posterior beliefs affects the accuracy of likelihood judgments.

Intervention

Participants are randomly assigned to one of four experimental conditions in a belief-updating task. Across conditions, the study varies whether participants are asked to report likelihoods, are provided with likelihood information, and/or are asked to report posterior beliefs. All participants complete repeated rounds of a probabilistic inference task, where the information structure varies across rounds.

The Baseline (BL) group serves as the control condition, in which participants report posterior beliefs only. The three treatment groups (CU, PU, CO) vary in whether likelihoods are elicited or provided, and whether posterior beliefs are also elicited.

Experimental design

The study uses a belief-updating task based on a standard bookbag-and-poker-chip paradigm. Participants complete multiple independent rounds in which there are two bags containing different proportions of colored balls. In each round, one bag is selected at random, a small sample of balls is drawn with replacement, and participants report probabilities based on the observed sample.

Participants are randomly assigned to one of four treatment conditions that vary in whether they report likelihoods, posterior beliefs, or both, and whether likelihood information is provided or must be computed by the participant. In some conditions, participants report the probability of observing the sample from each bag. In other conditions, participants report the probability that each bag generated the sample. In one condition, participants are shown the correct likelihoods before reporting posterior beliefs.

Across rounds, the number of draws and the composition of the bags vary, generating exogenous variation in the Bayesian likelihoods and posteriors. All questions are framed hypothetically, with probabilities fully determined by the stated bag compositions and samples.

Additional details on procedures, analysis, and exclusions are provided in the uploaded Pre-Analysis Plan.

Design details

Randomization method
Participants are randomly assigned to one of four treatment conditions using Qualtrics’ built-in randomization features.
Randomization unit
Individual participants
Clusters
800 participants
Observations
800 participants
Arms
200 participants in each of four treatment groups
Primary outcomes
The primary outcomes are participants’ reported posterior probabilities that a bag generated a given sample, and their reported likelihood probabilities of drawing the sample from each bag.

Cited by

No citation data available for this registration yet.

Keywords