Belief Repair After Neglected Events

This study uses two individual computer-based belief-updating tasks for adult members of the Cambridge Judge Business School participant pool who are eligible for in-person laboratory sessions. Across both tasks, participants observe information, report beliefs, and later receive clarifying information that may change how earlier information should be interpreted. The purpose of the study is to measure how people revise beliefs when later clarification affects the interpretation of earlier evidence.

2026-03-09

2026-03-31

Task 1 primary confirmatory endpoint: difference between the subtype share assigned after the informative refinement to a pre-registered subtype of interest and the analogous subtype share assigned after the relabel-only refinement to a pre-registered comparison subtype.

Task 2 co-primary confirmatory endpoint 1: immediate change in the reported probability assigned to the lowest state in Task 2 between the last report before clarification and the first report after clarification.

Task 2 co-primary confirmatory endpoint 2: immediate change in the reported probability assigned to the highest state in Task 2 between the last report before clarification and the first report after clarification.

Beliefs are reported as percentages summing to 100 over the states currently shown to the participant.

Task 1 primary endpoint: At the informative refinement, we calculate the share of belief assigned immediately after the refinement to a pre-registered subtype of interest among the two newly revealed subtypes of the parent condition being split. Call this TreatSubtypeShare. At the relabel-only refinement, we calculate the analogous share assigned immediately after the refinement to a pre-registered comparison subtype. Call this ControlSubtypeShare. We define:

DeltaSubtypeShare = TreatSubtypeShare - ControlSubtypeShare

A larger value means the participant places relatively more belief on the pre-registered subtype of interest at the informative refinement than on the comparison subtype at the relabel-only refinement.

The exact arm-specific subtype mapping is pre-registered in the hidden experimental design field and the uploaded analysis plan so that the public record does not reveal the scripted surprise while data collection is ongoing.

Task 2 co-primary endpoint 1: Let LowestStateBefore and LowestStateAfter denote the participant’s reported probability on the lowest state immediately before and immediately after clarification. Define:

DeltaLowestState= LowestStateAfter - LowestStateBefore

Task 2 co-primary endpoint 2: Let HighestStateBefore and HighestStateAfter denote the participant’s reported probability on the highest state immediately before and immediately after clarification. Define:

DeltaHighestState= HighestStateAfter - HighestStateBefore

Non zero values indicate movement in the predicted direction after clarification.

Task 1 secondary inferential endpoint: difference between the immediate drop in belief in the parent condition at the informative refinement and the corresponding drop at the relabel-only refinement.

Task 1 secondary descriptive endpoint: difference between the absolute distance from an even split across the two newly revealed subtypes at the informative refinement and the corresponding absolute distance at the relabel-only refinement.

Task 1 secondary descriptive endpoint: difference between the absolute change in belief in the parent condition at the informative refinement and the corresponding absolute change at the relabel-only refinement.

Task 2 secondary inferential endpoint: immediate drop in the expected value of the ordered state in Task 2 between the last report before clarification and the first report after clarification.

Task 2 secondary descriptive endpoint: total immediate belief-reallocation magnitude at the clarification moment.

Task 1 secondary inferential endpoint: Let ParentBeforeTreat and ParentAfterTreat denote the participant’s belief in the parent condition immediately before and immediately after the informative refinement. Let ParentBeforeControl and ParentAfterControl denote the analogous beliefs around the relabel-only refinement. Define:

ParentConditionDrop = (ParentBeforeTreat - ParentAfterTreat) - (ParentBeforeControl - ParentAfterControl)

A larger value means the participant reduces belief in the parent condition more at the informative refinement than at the relabel-only refinement.

Task 1 direction-free subtype-separation endpoint: Let TreatSubtypeShare and ControlSubtypeShare be defined as above. Define:

AbsSubtypeSeparation = abs(TreatSubtypeShare - 50) - abs(ControlSubtypeShare - 50)

A larger value means the informative refinement induces a larger immediate separation between the two newly revealed subtypes than the relabel-only refinement, regardless of direction.

Task 1 direction-free parent-change endpoint: Define:

AbsParentChange = abs(ParentAfterTreat - ParentBeforeTreat) - abs(ParentAfterControl - ParentBeforeControl)

A larger value means the informative refinement induces a larger immediate change in belief in the parent condition than the relabel-only refinement, regardless of direction.

Task 2 secondary inferential endpoint: Let Belief50Before, Belief55Before, and Belief60Before denote the reported probabilities on the three ordered states before clarification, and let Belief50After, Belief55After, and Belief60After denote the analogous probabilities after clarification. Define:

ExpectedRedBefore = 50*Belief50Before + 55*Belief55Before + 60*Belief60Before

ExpectedRedAfter = 50*Belief50After + 55*Belief55After + 60*Belief60After

ExpectedRedDrop = (ExpectedStateBefore - ExpectedStateAfter) / 100

A larger value means the clarification shifts beliefs toward lower states.

Task 2 total immediate movement endpoint: Define:

TotalJumpMagnitude = 0.5 * (abs(Belief50After - Belief50Before) + abs(Belief55After - Belief55Before) + abs(Belief60After - Belief60Before))

A larger value means the participant makes a larger immediate reallocation of belief at clarification, regardless of direction.

Additional projection-based magnitude measures, miss-distance measures, later-dynamics summaries, and cross-task consistency measures are pre-specified in the hidden design field and uploaded analysis plan and do not alter the main confirmatory claims.

This is an in-person laboratory experiment in which participants complete two individual tasks at separate computers with no interaction across participants. One task includes individual random assignment to one of two arms that determine the order in which an informative change versus a non-informative change occurs. The other task uses a common scripted design for all participants. The main confirmatory analyses use participant-level immediate belief changes at pre-specified moments. Eligible participants are members of the Cambridge Judge Business School participant pool with native-level English proficiency, age 60 or below, and eligibility for in-person laboratory sessions. Recruitment will continue until 80 participants with usable confirmatory outcome data are obtained or until a maximum of 120 participants is enrolled.

Not available

At the start of Task 1, each participant is randomly assigned to a workstation/computer token from a shuffled set. Each token is pre-linked to arm O1 or O2 according to a computer-generated 1:1 allocation list. This implements individual-level random assignment to O1 versus O2. There is no session-level treatment assignment and no manual reassignment after the token draw. Task 2 has no separate treatment-arm randomization because all participants complete the same scripted design.

Individual participant. In Task 1, assignment to O1 versus O2 occurs through random assignment to a pre-coded workstation/computer token. Task 2 has no separate randomized arms.

No

Individual participants. Target usable sample: 80 individuals. Maximum enrolled sample: 120 individuals.

Maximum 120 enrolled participants across all arms. Recruitment stops when 80 participants have usable confirmatory outcome data or when the cap of 120 enrolled participants is reached, whichever comes first

Task 1 has two arms with approximately 1:1 allocation. Target usable sample: approximately 40 participants in O1 and 40 participants in O2. Maximum enrolled sample: approximately 60 participants in O1 and 60 participants in O2, subject to minor imbalance from random assignment.

Task 2 is completed by all enrolled participants and has no additional randomized arms.