Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 109 of the 109 references it cites were found in a registry or an index; 19 of the 33 sources its claims cite were checked for support.

Social information creates self-fulfilling prophecies in judgments of pain, vicarious pain, and cognitive effort

Open access
Yazdanpanah, Aryan, Jung, Heejung, Soltani, Alireza, Wager, Tor D. · 2026 · Proceedings of the National Academy of Sciences

Abstract

Expectations can shape perception and potentially lead to self-fulfilling prophecies such as placebo effects that persist or grow over time. Nonetheless, whether and how unreinforced and unconditioned social cues (i.e., suggestions about future experiences that have not been reinforced with reward or punishment) can create and sustain such effects is unknown. We conducted a set of experiments in which participants (N = 111) experienced stimuli eliciting somatic pain (heat), vicarious pain (videos of others in pain), and cognitive effort (a mental-rotation task), at three intensity levels each. Before each stimulus, participants viewed a social cue that ostensibly indicated ratings from 10 other participants but was in fact randomized to a high or low mean aversiveness level independent of actual stimulus intensity. Across all tasks, participants' expectations and experience ratings shifted in line with the cues, with high-aversive cues leading to higher perceived aversiveness. Computational modeling and behavioral analysis revealed lower learning rates for prediction errors inconsistent with the trial's cue value (e.g., better than expected for high-aversive cues) and higher learning rates for prediction errors consistent with the cue value (e.g., worse than expected for high-aversive cues). These findings reveal a confirmation bias in learning: people update more when outcomes align with expectations. Combined with expectation effects on perception, this bias helps sustain social cue effects. Together, these mechanisms show how social information can shape perception and learning, giving rise to self-fulfilling prophecies.

Overview

CoChat Summary

## TL;DR

This paper tests whether unreinforced social cues can bias judgments of pain, vicarious pain, and cognitive effort, and finds that they do. Across three tasks with 111 participants, the authors report self-reinforcing expectancy effects, confirmation bias in learning, and some cross-domain stability in cue effects.

## Key findings

  • Objective stimulus intensity strongly increased perceptual ratings in all three tasks: pain (P = 2.68×10^-17), vicarious pain (P = 5.69×10^-18), and cognitive effort (P = 4.02×10^-15), with large effect sizes from 0.78 to 0.86.
  • High social cues substantially increased expectation ratings compared with low social cues in all tasks: pain (P = 1.42×10^-16), vicarious pain (P = 6.43×10^-18), and cognitive effort (P = 2.68×10^-17), with effect sizes from 0.82 to 0.86.
  • High social cues also increased perceptual ratings, even controlling for objective stimulus intensity: pain (P = 2.62×10^-13), vicarious pain (P = 2.57×10^-16), and cognitive effort (P = 6.84×10^-16), with effect sizes from 0.73 to 0.81.
  • Confirmation bias in learning was observed across all three tasks: learning rates for cue-congruent trials were significantly higher than for cue-incongruent trials in pain (P = 8.97×10^-13), vicarious pain (P = 5.5×10^-15), and cognitive effort (P = 2.9×10^-15).
  • In cross-task correlations, confirmation bias in learning was moderately stable between pain and cognitive effort (rs = 0.28) and between vicarious pain and cognitive effort (rs = 0.29), but not between pain and vicarious pain (rs = 0.15, P = 0.15).
  • The best-fitting computational model was Model 3c, which included separate learning rates for confirmatory and disconfirmatory prediction errors plus perceptual carryover; the paper reports that this model fit better than the alternatives by model frequency and protected exceedance probability.

## Methods

  • Three-task experimental design: thermal pain, vicarious pain, and mental rotation cognitive effort, each with a 2 (high/low social cue) × 3 (low/medium/high stimulus intensity) factorial design and 12 trials per condition; the tasks were run across three sessions on different days.
  • Sample: 111 recruited participants total (111 unique individuals), average age 24.7 ± 5.5 years; analyses were restricted to participants completing at least 24 trials, yielding 95 in the pain task, 99 in vicarious pain, and 95 in cognitive effort.
  • Analysis combined participant-level general linear models, nonparametric tests, multilevel mediation analyses, cross-task rank correlations, and hierarchical Bayesian reinforcement-learning model comparison across nine models.

## Limitations

  • The paper states that some participants did not learn the values of high and low cues despite trial-by-trial feedback, and that some showed no carryover effect, especially in the pain task.
  • The paper notes that cue effects on perception had only low-to-moderate reliability (r = 0.32 to 0.50), and stimulus intensity effects were low for the cognitive effort task (r = 0.18), which the text says has implications for using these measures as individual-differences indices.
  • Evident from the text: the cue ratings were “mere suggestions” with no real-world reinforcement and the tasks were laboratory-based with healthy participants from Dartmouth College, so generalization beyond this setting is not directly tested.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1073/pnas.2513856123

Cites

109 of 109 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1. Experimental paradigm and behavioral results. (A) Trial design for the three tasks: pain, vicarious pain, and cognitive effort. In each task, a social cue is presented that indicates perceptual ratings of previous participants in the study (a fictitious cover story). Participants rate the expected intensity of pain, another individual’s pain, or effort, followed by presentation of a task stimulus and a subsequent experience rating. In the pain task, the stimulus is heat applied to the inner forearm (either 48, 49, or 50 °C). In the vicarious pain task, it is watching another person in pain (with three levels of prenormed intensity). In the cognitive effort task, it is mental rotation of objects with three levels of rotation (50, 100, and 150). (B) Perceptual ratings as a function of stimulus intensity and cue type. As expected, perceptual ratings increase when the stimulus intensity level goes up for all three tasks. Importantly, perceptual ratings are also higher for the high vs. low cues for all tasks and stimulus intensity levels. Error bars represent the within-subject SEM using the Cousineau–Morey method (59, 60). Note that the ranges of the y axes are not the same; we are interested in within-task effects. (C) Individual participant-level (dots) stimulus effects, cue effects, and their interaction on perceptual ratings in each task. Stimulus and cue effects reflect high–low stimulus and high–low cue differences. **P < 0.01; ***P < 0.001 for tests of average within-participant effects against zero using the Wilcoxon two-sided signed rank test.
Fig. 1. Experimental paradigm and behavioral results. (A) Trial design for the three tasks: pain, vicarious pain, and cognitive effort. In each task, a social cue is presented that indicates perceptual ratings of previous participants in the study (a fictitious cover story). Participants rate the expected intensity of pain, another individual’s pain, or effort, followed by presentation of a task stimulus and a subsequent experience rating. In the pain task, the stimulus is heat applied to the inner forearm (either 48, 49, or 50 °C). In the vicarious pain task, it is watching another person in pain (with three levels of prenormed intensity). In the cognitive effort task, it is mental rotation of objects with three levels of rotation (50, 100, and 150). (B) Perceptual ratings as a function of stimulus intensity and cue type. As expected, perceptual ratings increase when the stimulus intensity level goes up for all three tasks. Importantly, perceptual ratings are also higher for the high vs. low cues for all tasks and stimulus intensity levels. Error bars represent the within-subject SEM using the Cousineau–Morey method (59, 60). Note that the ranges of the y axes are not the same; we are interested in within-task effects. (C) Individual participant-level (dots) stimulus effects, cue effects, and their interaction on perceptual ratings in each task. Stimulus and cue effects reflect high–low stimulus and high–low cue differences. **P < 0.01; ***P < 0.001 for tests of average within-participant effects against zero using the Wilcoxon two-sided signed rank test.CC BY · PubMed Central
Fig. 2. Persistent effects of expectations and perceptual ratings in high and low cues across time. Dynamics of expectation (A–C) and perceptual (D–F) ratings in the three tasks (indicated on the top), showing the persistent cue effects. The error bars show the within-subject errors. Note that not all subjects had all the trials for low and high cues. Dashed lines indicate the start of a new session, which corresponds to a new day. Smoothed curves reflect spline fits, with shading and error bars reflecting within subject error based on the Cousineau–Morey method (59, 60). The ratings are from 0 to 180 scale.
Fig. 2. Persistent effects of expectations and perceptual ratings in high and low cues across time. Dynamics of expectation (A–C) and perceptual (D–F) ratings in the three tasks (indicated on the top), showing the persistent cue effects. The error bars show the within-subject errors. Note that not all subjects had all the trials for low and high cues. Dashed lines indicate the start of a new session, which corresponds to a new day. Smoothed curves reflect spline fits, with shading and error bars reflecting within subject error based on the Cousineau–Morey method (59, 60). The ratings are from 0 to 180 scale.CC BY · PubMed Central
Fig. 3. Reliability and cross-task correlations of cue and stimulus effects. Correlation and reliability matrix between the cue and stimulus effects within and between tasks. Cue effects are defined as [high cue – low cue] effects on perceptual ratings, and stimulus intensity effects as [high objective intensity – low objective intensity] effects on ratings. The diagonal axis (elements in the matrix with the same x and y label) represents the reliability of the effects (odd–even trial reliability with Spearman–Brown adjustment). Off-diagonal elements of the matrix represent the Spearman rank correlation between different effects. The correlations between cue effects are generally stronger than the correlation between the stimulus effects, pointing out the stronger stability of cue effects compared to stimulus intensity effects across tasks. N = 88 participants completed all three tasks.
Fig. 3. Reliability and cross-task correlations of cue and stimulus effects. Correlation and reliability matrix between the cue and stimulus effects within and between tasks. Cue effects are defined as [high cue – low cue] effects on perceptual ratings, and stimulus intensity effects as [high objective intensity – low objective intensity] effects on ratings. The diagonal axis (elements in the matrix with the same x and y label) represents the reliability of the effects (odd–even trial reliability with Spearman–Brown adjustment). Off-diagonal elements of the matrix represent the Spearman rank correlation between different effects. The correlations between cue effects are generally stronger than the correlation between the stimulus effects, pointing out the stronger stability of cue effects compared to stimulus intensity effects across tasks. N = 88 participants completed all three tasks.CC BY · PubMed Central
Fig. 4. Model-agnostic learning rates indicate confirmation bias in learning. (A–C) The model-agnostic learning rates calculated as the slope of a line fitted to the expectation update as a function of prediction error (PE) within each condition, separately for pain, vicarious pain, and cognitive effort tasks. High (red) and low (blue) cue correspond to appetitive (PE<0) and aversive (PE > 0) trials, respectively. Each dot represents a participant and N is the number of participants in each task. Upward arrows, downward arrows, and x represent the number of participants that are outlier above the distribution, outlier below the distribution, and participants with no trials in that specific condition, respectively. We included participants with at least 24 trials which led to 95, 99, and 95 in pain, vicarious pain, and cognitive tasks, respectively.
Fig. 4. Model-agnostic learning rates indicate confirmation bias in learning. (A–C) The model-agnostic learning rates calculated as the slope of a line fitted to the expectation update as a function of prediction error (PE) within each condition, separately for pain, vicarious pain, and cognitive effort tasks. High (red) and low (blue) cue correspond to appetitive (PE<0) and aversive (PE > 0) trials, respectively. Each dot represents a participant and N is the number of participants in each task. Upward arrows, downward arrows, and x represent the number of participants that are outlier above the distribution, outlier below the distribution, and participants with no trials in that specific condition, respectively. We included participants with at least 24 trials which led to 95, 99, and 95 in pain, vicarious pain, and cognitive tasks, respectively.CC BY · PubMed Central
Fig. 5. Model comparisons reveal that the model with two learning rates and carryover from previous trial provides the best fit to data across the three tasks. (A) Schematic of the best-fitting model (Model 3c). In this model, the stimulus (O) and expectation (E) together determine the perceptual rating or perceived outcome (O), i.e., experienced utility, on each trial. Crucially, the model assumes that O is both influenced by E and used to generate the prediction error that informs future values of E, creating a short-circuit in the learning process. The model includes six free parameters: two learning rates for congruent and incongruent trials (αc and αi), a weight on the relative contribution of S and E on O (ω), a carryover effect from current O to the next trial’s E (βO, “carryoverPrcpt”), and two scaling parameters common to all models (σE² and σO²). (B–D) Plots report models’ protected exceedance probabilities (pxp) equal to the probability a model is most prevalent, considering chance, and models’ frequencies, measuring how common each model is in the population.
Fig. 5. Model comparisons reveal that the model with two learning rates and carryover from previous trial provides the best fit to data across the three tasks. (A) Schematic of the best-fitting model (Model 3c). In this model, the stimulus (O) and expectation (E) together determine the perceptual rating or perceived outcome (O), i.e., experienced utility, on each trial. Crucially, the model assumes that O is both influenced by E and used to generate the prediction error that informs future values of E, creating a short-circuit in the learning process. The model includes six free parameters: two learning rates for congruent and incongruent trials (αc and αi), a weight on the relative contribution of S and E on O (ω), a carryover effect from current O to the next trial’s E (βO, “carryoverPrcpt”), and two scaling parameters common to all models (σE² and σO²). (B–D) Plots report models’ protected exceedance probabilities (pxp) equal to the probability a model is most prevalent, considering chance, and models’ frequencies, measuring how common each model is in the population.CC BY · PubMed Central
Fig. 6. Computational model captures the bidirectional effects between perception and expectations, as well as different learning rates in congruent vs. incongruent trials. (A–C) Estimated learning rates, separately for congruent (αc) and incongruent (αi) learning rates in pain, vicarious pain, and cognitive effort tasks. The higher learning rates for congruent trials compared to incongruent trials correspond to confirmation bias in expectation updates. The learning rate of one indicates that participants use the perceptual rating from the previous trial as their current expectation. (D) Estimated relative weights (ω) of expectation to stimulus, separately for pain, vicarious pain, and cognitive effort tasks. Each dot represents a participant. The weights are relatively small, pointing out the stronger effect of stimulus compared to expectations. (E) Estimated carryover parameter (βO).
Fig. 6. Computational model captures the bidirectional effects between perception and expectations, as well as different learning rates in congruent vs. incongruent trials. (A–C) Estimated learning rates, separately for congruent (αc) and incongruent (αi) learning rates in pain, vicarious pain, and cognitive effort tasks. The higher learning rates for congruent trials compared to incongruent trials correspond to confirmation bias in expectation updates. The learning rate of one indicates that participants use the perceptual rating from the previous trial as their current expectation. (D) Estimated relative weights (ω) of expectation to stimulus, separately for pain, vicarious pain, and cognitive effort tasks. Each dot represents a participant. The weights are relatively small, pointing out the stronger effect of stimulus compared to expectations. (E) Estimated carryover parameter (βO).CC BY · PubMed Central

Verification

Claims the paper makes

9 Valid · 0 Not valid · 5 Not checked · 82 Own results and interpretations

45 more claims

Create a free account to see all 96 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 9
  • It may state two separate claims as one. 41

Create a free account to see these claims.

Citations

Cites

Show all 104 references

Research works printed but not linked: 6

  1. N. A. Christakis, Blueprint: The Evolutionary Origins of a Good Society (Little, Brown Spark, 2019).
  2. M. Sherif, A study of some social factors in perception. Arch. Psychol. Columbia Univ. 187, 60–60 (1935).
  3. I. P. Pavlov, Conditional Reflexes: An Investigation of the Physiological Activity of the Cerebral Cortex (Oxford University Press, 1927).
  4. R. S. Sutton, A. G. Barto, Reinforcement Learning: An Introduction (The MIT Press, ed. 2, 2018).
  5. 10.1016/S0304-3959(97)00016-X · DOI · Paper page
  6. H. Helson, Adaptation-Level Theory: An Experimental and Systematic Approach to Behavior (Harper and Row, New York, NY, 1964).

Cited by

See all 1

Metadata

Authors
Yazdanpanah, A., Jung, H., Soltani, A., & Wager, T. D.
Affiliations
Dartmouth College (US)
Corresponding author
  • Soltani, Alireza
  • Wager, Tor D.
Venue
Proceedings of the National Academy of Sciences
Published
2026-02-09
Volume
123
Issue
7
Pages
e2513856123-e2513856123
ISSN
  • 0027-8424 (ISSN-L)
  • 1091-6490
OpenAlex
W7128442868
Cited by
1
References
104
Citations by year
2026: 1
FWCI
4.11
Type
journal article
Language
en
Open access
Locations
  • Proceedings of the National Academy of Sciences (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Pain Management and Placebo Effect
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
  • Health Sciences › Medicine › Physiology (medical) › Pain Mechanisms and Treatments
Keywords
self-fulfilling prophecy, social cues, vicarious pain, pain perception, cognitive effort, mental rotation, placebo effect, prediction error learning, social information
MeSH terms
Outcome Expectations, Adult, Cognition, Cues, Female, Humans, Judgment, Male, Pain, Placebo Effect, Young Adult, Pain Perception

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Social information creates self-fulfilling prophecies in judgments of pain, vicarious pain, and cognitive effort. https://app.cochat.ai/papers/10.1073/pnas.2513856123
The paper
Yazdanpanah, A., Jung, H., Soltani, A., & Wager, T. D. (2026). Social information creates self-fulfilling prophecies in judgments of pain, vicarious pain, and cognitive effort. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2513856123

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.