Joseph Heffner
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Joseph Heffner has a total of 39 co-authors. Here are the top co-authors by number of shared publications:
- Oriel FeldmanHall
Department of Cognitive and Psychological Sciences, Brown University
11 shared publications - Robb B. Rutledge
Department of Psychology, Yale University, New Haven, USA
3 shared publications - Aisulu Omar
Center for Computation and Visualization, Brown University, Providence, RI, United States
2 shared publications - Bradford Nazario Roarr
Center for Computation and Visualization, Brown University, Providence, RI, United States
2 shared publications - Chloe Zimmerman Gunsilius
Brown University
2 shared publications - Ellen Duong
Center for Computation and Visualization, Brown University, Providence, RI, United States
2 shared publications
Latest Publications
- Open Access
- 4 September 2026
- Published by Center for Open Science in PsyArXiv
Longitudinal computational phenotyping reveals a stable marker for depression symptoms
Major depression is associated with reinforcement learning deficits, but it is unclear whether model parameters are stable behavioral markers for specific symptoms. We collected longitudinal smartphone task and questionnaire data from 300 depressed and 124 non-depressed participants over six months (3,461 sessions) with diagnosis confirmed by clinical interviews. Orthogonal general depression and anhedonia dimensions were extracted from questionnaires. Using computational modeling, we found that reward sensitivity, quantifying the blunted impact of small rewards on learning, was associated with elevated anhedonia and not general depression or diagnosis. In contrast, low baseline mood was associated with increased general depression but not anhedonia. Longitudinal analysis showed that baseline mood tracks changes in general depression over time, while reward sensitivity is a stable trait-like marker that predicts anhedonia. Using smartphone-based longitudinal computational phenotyping, we find distinct relationships between reinforcement learning and heterogeneous symptoms, identifying behavioral markers that represent potential targets for personalized medicine.
- 4 September 2026
- Published by Center for Open Science in PsyArXiv
Open AccessLongitudinal computational phenotyping reveals a stable marker for depression symptoms
Major depression is associated with reinforcement learning deficits, but it is unclear whether model parameters are stable behavioral markers for specific symptoms. We collected longitudinal smartphone task and questionnaire data from 300 depressed and 124 non-depressed participants over six months (3,461 sessions) with diagnosis confirmed by clinical interviews. Orthogonal general depression and anhedonia dimensions were extracted from questionnaires. Using computational modeling, we found that reward sensitivity, quantifying the blunted impact of small rewards on learning, was associated with elevated anhedonia and not general depression or diagnosis. In contrast, low baseline mood was associated with increased general depression but not anhedonia. Longitudinal analysis showed that baseline mood tracks changes in general depression over time, while reward sensitivity is a stable trait-like marker that predicts anhedonia. Using smartphone-based longitudinal computational phenotyping, we find distinct relationships between reinforcement learning and heterogeneous symptoms, identifying behavioral markers that represent potential targets for personalized medicine.
- Open Access
- 30 June 2026
- Published by Center for Open Science in PsyArXiv
Goals remap values and not choices
The ability to pursue different goals is fundamental to adaptive decision-making, yet the mechanisms driving this capacity remain unclear. Decision-making is often thought of as having distinct valuation and choice stages. Goals could potentially act at either stage by remapping values or by remapping choices. To distinguish between these possibilities, participants completed an economic decision-making task choosing between a risky and a safe option under opposing goals to maximize or minimize points. If goals remap choices, people will simply switch to the opposite option under opposing goals. If goals remap values, people will treat losses like gains under opposing goals, leading to only sometimes switching to the opposite option. Across four studies (N=1096) and a preregistered replication (N=500), participant choices were best explained by goals remapping values. Computational modeling based on prospect theory confirmed two counterintuitive predictions of goals remapping value: first, that loss aversion acts on gains under the minimize goal, and second, that individuals systematically make identical choices despite opposing goals due to how they value options. We also found reduced overall outcome sensitivity under the novel minimize goal as measured both by choices and by happiness ratings collected periodically throughout the task. When goal changes were frequent, reduced affective outcome sensitivity predicted reduced choice outcome sensitivity, suggesting that emotional alignment across goals supports rational choice. Together, these findings reveal how goals change values, capturing the behavioral and emotional signatures of goal-directed decision-making.
- 30 June 2026
- Published by Center for Open Science in PsyArXiv
Open AccessGoals remap values and not choices
The ability to pursue different goals is fundamental to adaptive decision-making, yet the mechanisms driving this capacity remain unclear. Decision-making is often thought of as having distinct valuation and choice stages. Goals could potentially act at either stage by remapping values or by remapping choices. To distinguish between these possibilities, participants completed an economic decision-making task choosing between a risky and a safe option under opposing goals to maximize or minimize points. If goals remap choices, people will simply switch to the opposite option under opposing goals. If goals remap values, people will treat losses like gains under opposing goals, leading to only sometimes switching to the opposite option. Across four studies (N=1096) and a preregistered replication (N=500), participant choices were best explained by goals remapping values. Computational modeling based on prospect theory confirmed two counterintuitive predictions of goals remapping value: first, that loss aversion acts on gains under the minimize goal, and second, that individuals systematically make identical choices despite opposing goals due to how they value options. We also found reduced overall outcome sensitivity under the novel minimize goal as measured both by choices and by happiness ratings collected periodically throughout the task. When goal changes were frequent, reduced affective outcome sensitivity predicted reduced choice outcome sensitivity, suggesting that emotional alignment across goals supports rational choice. Together, these findings reveal how goals change values, capturing the behavioral and emotional signatures of goal-directed decision-making.
- Open Access
- 29 April 2026
- Published by Springer Nature in Communications Psychology
- Vol. 4 (1)
Affective valence as a computational signal for learning value
Prevailing accounts differ on affect's role during learning, arguing that affect is either a byproduct of choice or irrelevant to the process altogether. Across two experiments (N1 = 75; N2 = 95) plus a replication study (NR = 55), we combine behavioral tasks with computational modeling to describe affect's role in value-based learning and choice. Trial-by-trial valence predicts both choice and beliefs about future outcomes, beyond experienced rewards. Models incorporating a valence term that specifically shapes updating during learning fit the data better than those with only reward. Valence's contribution persists under heightened uncertainty and during passive learning, and is especially sensitive to socio-emotional information. By showing that valence tracks changes in people's expectations-their optimism/pessimism about the objective outcomes they would receive from their actions, and not just how they evaluate those outcomes during choice-these results advance a computational account of emotion in decision-making and clarify its potency in social contexts.
- 29 April 2026
- Published by Springer Nature in Communications Psychology
- Vol. 4 (1)
Open AccessAffective valence as a computational signal for learning value
Prevailing accounts differ on affect's role during learning, arguing that affect is either a byproduct of choice or irrelevant to the process altogether. Across two experiments (N1 = 75; N2 = 95) plus a replication study (NR = 55), we combine behavioral tasks with computational modeling to describe affect's role in value-based learning and choice. Trial-by-trial valence predicts both choice and beliefs about future outcomes, beyond experienced rewards. Models incorporating a valence term that specifically shapes updating during learning fit the data better than those with only reward. Valence's contribution persists under heightened uncertainty and during passive learning, and is especially sensitive to socio-emotional information. By showing that valence tracks changes in people's expectations-their optimism/pessimism about the objective outcomes they would receive from their actions, and not just how they evaluate those outcomes during choice-these results advance a computational account of emotion in decision-making and clarify its potency in social contexts.
- Open Access
- 22 August 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-11
Separable neural signals for reward and emotion prediction errors
Reinforcement learning models focus on reward prediction errors as the driver of behavior. However, recent evidence indicates that deviations from emotion expectations, termed affective prediction errors, also crucially shape behavior. Whether there is neural separability between emotion and reward signals remains unknown. We employ electroencephalography during social learning to investigate the neural signatures of reward and affective prediction errors. Behavioral results reveal that affective prediction errors are associated with choices when little is known about how a partner will behave. This behavioral evidence is mirrored neurally by engagement of separate event-related potentials. More specifically, the feedback-related negativity is largely and consistently indexed by reward prediction errors, while the P3b is more consistently tracked by affective prediction errors. The P3b in particular is linked to subsequent choices, highlighting the mechanistic influence of emotion during social learning. These findings present evidence for a neurobiologically viable emotion learning signal that is partially distinguishable, at both the behavior and neural levels, from reward.
- 22 August 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-11
Open AccessSeparable neural signals for reward and emotion prediction errors
Reinforcement learning models focus on reward prediction errors as the driver of behavior. However, recent evidence indicates that deviations from emotion expectations, termed affective prediction errors, also crucially shape behavior. Whether there is neural separability between emotion and reward signals remains unknown. We employ electroencephalography during social learning to investigate the neural signatures of reward and affective prediction errors. Behavioral results reveal that affective prediction errors are associated with choices when little is known about how a partner will behave. This behavioral evidence is mirrored neurally by engagement of separate event-related potentials. More specifically, the feedback-related negativity is largely and consistently indexed by reward prediction errors, while the P3b is more consistently tracked by affective prediction errors. The P3b in particular is linked to subsequent choices, highlighting the mechanistic influence of emotion during social learning. These findings present evidence for a neurobiologically viable emotion learning signal that is partially distinguishable, at both the behavior and neural levels, from reward.
- Open Access
- 16 September 2024
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 121 (39)
Language sentiment predicts changes in depressive symptoms
The prevalence of depression is a major societal health concern, and there is an ongoing need to develop tools that predict who will become depressed. Past research suggests that depression changes the language we use, but it is unclear whether language is predictive of worsening symptoms. Here, we test whether the sentiment of brief written linguistic responses predicts changes in depression. Across two studies (N= 467), participants provided responses to neutral open-ended questions, narrating aspects of their lives relevant to depression (e.g., mood, motivation, sleep). Participants also completed the Patient Health Questionnaire (PHQ-9) to assess depressive symptoms and a risky decision-making task with periodic measurements of momentary happiness to quantify mood dynamics. The sentiment of written responses was evaluated by human raters (N= 470), Large Language Models (LLMs; ChatGPT 3.5 and 4.0), and the Linguistic Inquiry and Word Count (LIWC) tool. We found that language sentiment evaluated by human raters and LLMs, but not LIWC, predicted changes in depressive symptoms at a three-week follow-up. Using computational modeling, we found that language sentiment was associated with current mood, but language sentiment predicted symptom changes even after controlling for current mood. In summary, we demonstrate a scalable tool that combines brief written responses with sentiment analysis by AI tools that matches human performance in the prediction of future psychiatric symptoms.
- 16 September 2024
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 121 (39)
Open AccessLanguage sentiment predicts changes in depressive symptoms
The prevalence of depression is a major societal health concern, and there is an ongoing need to develop tools that predict who will become depressed. Past research suggests that depression changes the language we use, but it is unclear whether language is predictive of worsening symptoms. Here, we test whether the sentiment of brief written linguistic responses predicts changes in depression. Across two studies (N= 467), participants provided responses to neutral open-ended questions, narrating aspects of their lives relevant to depression (e.g., mood, motivation, sleep). Participants also completed the Patient Health Questionnaire (PHQ-9) to assess depressive symptoms and a risky decision-making task with periodic measurements of momentary happiness to quantify mood dynamics. The sentiment of written responses was evaluated by human raters (N= 470), Large Language Models (LLMs; ChatGPT 3.5 and 4.0), and the Linguistic Inquiry and Word Count (LIWC) tool. We found that language sentiment evaluated by human raters and LLMs, but not LIWC, predicted changes in depressive symptoms at a three-week follow-up. Using computational modeling, we found that language sentiment was associated with current mood, but language sentiment predicted symptom changes even after controlling for current mood. In summary, we demonstrate a scalable tool that combines brief written responses with sentiment analysis by AI tools that matches human performance in the prediction of future psychiatric symptoms.
- Open Access
Dissociable neural signals for reward and emotion prediction errors
Reinforcement learning models focus on reward prediction errors (PEs) as the driver of behavior. However, recent evidence indicates that deviations from emotion expectations, termed affective PEs, play a crucial role in shaping behavior. Whether there is neural separability between emotion and reward signals remains unknown. We employ electroencephalography during social learning to investigate the neural signatures of reward and affective PEs. Behavioral results reveal that while affective PEs predict choices when little is known about how a partner will behave, reward PEs become more predictive overtime as uncertainty about a partner’s behavior diminishes. This functional dissociation is mirrored neurally by engagement of distinct event-related potentials. The FRN indexes reward PEs while the P3b tracks affective PEs. Only the P3b predicts subsequent choices, highlighting the mechanistic influence of affective PEs during social learning. These findings present evidence for a neurobiologically viable emotion learning signal that is distinguishable—behaviorally and neurally—from reward. Significance For nearly a century, scientists have asked how humans learn about their worlds. Learning models borrowed from computer science—namely, reinforcement learning—provide an elegant and simple framework that showcases how reward prediction errors are used to update one’s knowledge about the environment. However, a fundamental question persists: what exactly is ‘reward’? This gap in knowledge is problematic, especially when we consider the multiplicity of social contexts where external rewards must be contextualized to gain value and meaning. We leverage electroencephalography to interrogate the role of emotion prediction errors—violations of emotional expectations—during learning. We observe distinct neural signals for reward and emotion prediction errors, suggesting that emotions may act as a bridge between external rewards and subjective value.
Open AccessDissociable neural signals for reward and emotion prediction errors
Reinforcement learning models focus on reward prediction errors (PEs) as the driver of behavior. However, recent evidence indicates that deviations from emotion expectations, termed affective PEs, play a crucial role in shaping behavior. Whether there is neural separability between emotion and reward signals remains unknown. We employ electroencephalography during social learning to investigate the neural signatures of reward and affective PEs. Behavioral results reveal that while affective PEs predict choices when little is known about how a partner will behave, reward PEs become more predictive overtime as uncertainty about a partner’s behavior diminishes. This functional dissociation is mirrored neurally by engagement of distinct event-related potentials. The FRN indexes reward PEs while the P3b tracks affective PEs. Only the P3b predicts subsequent choices, highlighting the mechanistic influence of affective PEs during social learning. These findings present evidence for a neurobiologically viable emotion learning signal that is distinguishable—behaviorally and neurally—from reward. Significance For nearly a century, scientists have asked how humans learn about their worlds. Learning models borrowed from computer science—namely, reinforcement learning—provide an elegant and simple framework that showcases how reward prediction errors are used to update one’s knowledge about the environment. However, a fundamental question persists: what exactly is ‘reward’? This gap in knowledge is problematic, especially when we consider the multiplicity of social contexts where external rewards must be contextualized to gain value and meaning. We leverage electroencephalography to interrogate the role of emotion prediction errors—violations of emotional expectations—during learning. We observe distinct neural signals for reward and emotion prediction errors, suggesting that emotions may act as a bridge between external rewards and subjective value.
- Open Access
- 12 January 2024
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 12 (1) , e47177
SOMAScience: A Novel Platform for Multidimensional, Longitudinal Pain Assessment
Chronic pain is one of the most significant health issues in the United States, affecting more than 20% of the population. Despite its contribution to the increasing health crisis, reliable predictors of disease development, progression, or treatment outcomes are lacking. Self-report remains the most effective way to assess pain, but measures are often acquired in sparse settings over short time windows, limiting their predictive ability. In this paper, we present a new mobile health platform called SOMAScience. SOMAScience serves as an easy-to-use research tool for scientists and clinicians, enabling the collection of large-scale pain datasets in single- and multicenter studies by facilitating the acquisition, transfer, and analysis of longitudinal, multidimensional, self-report pain data. Data acquisition for SOMAScience is done through a user-friendly smartphone app, SOMA, that uses experience sampling methodology to capture momentary and daily assessments of pain intensity, unpleasantness, interference, location, mood, activities, and predictions about the next day that provide personal insights into daily pain dynamics. The visualization of data and its trends over time is meant to empower individual users’ self-management of their pain. This paper outlines the scientific, clinical, technological, and user considerations involved in the development of SOMAScience and how it can be used in clinical studies or for pain self-management purposes. Our goal is for SOMAScience to provide a much-needed platform for individual users to gain insight into the multidimensional features of their pain while lowering the barrier for researchers and clinicians to obtain the type of pain data that will ultimately lead to improved prevention, diagnosis, and treatment of chronic pain.
- 12 January 2024
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 12 (1) , e47177
Open AccessSOMAScience: A Novel Platform for Multidimensional, Longitudinal Pain Assessment
Chronic pain is one of the most significant health issues in the United States, affecting more than 20% of the population. Despite its contribution to the increasing health crisis, reliable predictors of disease development, progression, or treatment outcomes are lacking. Self-report remains the most effective way to assess pain, but measures are often acquired in sparse settings over short time windows, limiting their predictive ability. In this paper, we present a new mobile health platform called SOMAScience. SOMAScience serves as an easy-to-use research tool for scientists and clinicians, enabling the collection of large-scale pain datasets in single- and multicenter studies by facilitating the acquisition, transfer, and analysis of longitudinal, multidimensional, self-report pain data. Data acquisition for SOMAScience is done through a user-friendly smartphone app, SOMA, that uses experience sampling methodology to capture momentary and daily assessments of pain intensity, unpleasantness, interference, location, mood, activities, and predictions about the next day that provide personal insights into daily pain dynamics. The visualization of data and its trends over time is meant to empower individual users’ self-management of their pain. This paper outlines the scientific, clinical, technological, and user considerations involved in the development of SOMAScience and how it can be used in clinical studies or for pain self-management purposes. Our goal is for SOMAScience to provide a much-needed platform for individual users to gain insight into the multidimensional features of their pain while lowering the barrier for researchers and clinicians to obtain the type of pain data that will ultimately lead to improved prevention, diagnosis, and treatment of chronic pain.
- Open Access
- 13 March 2023
- Published by JMIR Publications Inc. in JMIR Preprints
SOMAScience: A Novel Platform for Multidimensional, Longitudinal Pain Assessment (Preprint)
UNSTRUCTURED Chronic pain is one of the most significant health issues in the United States, affecting more than 20% of the population. Despite its contribution to the increasing health crisis, reliable predictors of disease development, progression, or treatment outcomes are lacking. Self-report remains the most effective way to assess pain, but measures are often acquired in sparse settings over short time windows, limiting their predictive ability. In this paper, we present a new mobile health platform called SOMAScience. SOMAScience serves as an easy-to-use research tool for scientists and clinicians, enabling the collection of large-scale pain datasets in single- and multicenter studies by facilitating the acquisition, transfer, and analysis of longitudinal, multidimensional, self-report pain data. Data acquisition for SOMAScience is done through a user-friendly smartphone app, SOMA, that uses experience sampling methodology to capture momentary and daily assessments of pain intensity, unpleasantness, interference, location, mood, activities, and predictions about the next day that provide personal insights into daily pain dynamics. The visualization of data and its trends over time is meant to empower individual users’ self-management of their pain. This paper outlines the scientific, clinical, technological, and user considerations involved in the development of SOMAScience and how it can be used in clinical studies or for pain self-management purposes. Our goal is for SOMAScience to provide a much-needed platform for individual users to gain insight into the multidimensional features of their pain while lowering the barrier for researchers and clinicians to obtain the type of pain data that will ultimately lead to improved prevention, diagnosis, and treatment of chronic pain.
- 13 March 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessSOMAScience: A Novel Platform for Multidimensional, Longitudinal Pain Assessment (Preprint)
UNSTRUCTURED Chronic pain is one of the most significant health issues in the United States, affecting more than 20% of the population. Despite its contribution to the increasing health crisis, reliable predictors of disease development, progression, or treatment outcomes are lacking. Self-report remains the most effective way to assess pain, but measures are often acquired in sparse settings over short time windows, limiting their predictive ability. In this paper, we present a new mobile health platform called SOMAScience. SOMAScience serves as an easy-to-use research tool for scientists and clinicians, enabling the collection of large-scale pain datasets in single- and multicenter studies by facilitating the acquisition, transfer, and analysis of longitudinal, multidimensional, self-report pain data. Data acquisition for SOMAScience is done through a user-friendly smartphone app, SOMA, that uses experience sampling methodology to capture momentary and daily assessments of pain intensity, unpleasantness, interference, location, mood, activities, and predictions about the next day that provide personal insights into daily pain dynamics. The visualization of data and its trends over time is meant to empower individual users’ self-management of their pain. This paper outlines the scientific, clinical, technological, and user considerations involved in the development of SOMAScience and how it can be used in clinical studies or for pain self-management purposes. Our goal is for SOMAScience to provide a much-needed platform for individual users to gain insight into the multidimensional features of their pain while lowering the barrier for researchers and clinicians to obtain the type of pain data that will ultimately lead to improved prevention, diagnosis, and treatment of chronic pain.
- Open Access
- 31 March 2022
- Published by Springer Nature in Nature Communications
- Vol. 13 (1) , 1-11
A probabilistic map of emotional experiences during competitive social interactions
Theories of emotion and decision-making argue that negative, high arousing emotions—such as anger—motivate competitive social choice (e.g., punishing and defecting). However, given the long-standing challenge of quantifying emotion and the narrow framework in which emotion is traditionally examined, it remains unclear which emotions are actually associated with motivating these types of choices. To address this gap, we combine machine learning algorithms with a measure of affect that is agnostic to any specific emotion label. The result is a probabilistic map of emotion that is used to classify the specific emotions experienced by participants in a variety of social interactions (Ultimatum Game, Prisoner’s Dilemma, and Public Goods Game). Our results reveal that punitive and uncooperative choices are linked to a diverse array of negative, neutrally arousing emotions, such as sadness and disappointment, while only weakly linked to anger. These findings stand in contrast to the commonly held assumption that anger drives decisions to punish, defect, and freeride—thus, offering new insight into the role of emotion in motiving social choice.
- 31 March 2022
- Published by Springer Nature in Nature Communications
- Vol. 13 (1) , 1-11
Open AccessA probabilistic map of emotional experiences during competitive social interactions
Theories of emotion and decision-making argue that negative, high arousing emotions—such as anger—motivate competitive social choice (e.g., punishing and defecting). However, given the long-standing challenge of quantifying emotion and the narrow framework in which emotion is traditionally examined, it remains unclear which emotions are actually associated with motivating these types of choices. To address this gap, we combine machine learning algorithms with a measure of affect that is agnostic to any specific emotion label. The result is a probabilistic map of emotion that is used to classify the specific emotions experienced by participants in a variety of social interactions (Ultimatum Game, Prisoner’s Dilemma, and Public Goods Game). Our results reveal that punitive and uncooperative choices are linked to a diverse array of negative, neutrally arousing emotions, such as sadness and disappointment, while only weakly linked to anger. These findings stand in contrast to the commonly held assumption that anger drives decisions to punish, defect, and freeride—thus, offering new insight into the role of emotion in motiving social choice.
- 19 October 2021
- Published by Springer Nature in Nature Human Behaviour
- Vol. 5 (10) , 1391-1401
Emotion prediction errors guide socially adaptive behaviour
People make decisions based on deviations from expected outcomes, known as prediction errors. Past work has focused on reward prediction errors, largely ignoring violations of expected emotional experiences—emotion prediction errors. We leverage a method to measure real-time fluctuations in emotion as people decide to punish or forgive others. Across four studies (N = 1,016), we reveal that emotion and reward prediction errors have distinguishable contributions to choice, such that emotion prediction errors exert the strongest impact during decision-making. We additionally find that a choice to punish or forgive can be decoded in less than a second from an evolving emotional response, suggesting that emotions swiftly influence choice. Finally, individuals reporting significant levels of depression exhibit selective impairments in using emotion—but not reward—prediction errors. Evidence for emotion prediction errors potently guiding social behaviours challenge standard decision-making models that have focused solely on reward.
- 19 October 2021
- Published by Springer Nature in Nature Human Behaviour
- Vol. 5 (10) , 1391-1401
Emotion prediction errors guide socially adaptive behaviour
People make decisions based on deviations from expected outcomes, known as prediction errors. Past work has focused on reward prediction errors, largely ignoring violations of expected emotional experiences—emotion prediction errors. We leverage a method to measure real-time fluctuations in emotion as people decide to punish or forgive others. Across four studies (N = 1,016), we reveal that emotion and reward prediction errors have distinguishable contributions to choice, such that emotion prediction errors exert the strongest impact during decision-making. We additionally find that a choice to punish or forgive can be decoded in less than a second from an evolving emotional response, suggesting that emotions swiftly influence choice. Finally, individuals reporting significant levels of depression exhibit selective impairments in using emotion—but not reward—prediction errors. Evidence for emotion prediction errors potently guiding social behaviours challenge standard decision-making models that have focused solely on reward.