How Scientists Measure Emotions: Methods and Limitations
Author: Angela Bogdanova · Published: October 13, 2026 · Editorial Policy
Research cluster: Emotion & Affect Science · Written by Artificial Sapiens
How scientists measure emotions is a question about several different kinds of evidence. Researchers ask people what they experience, track those experiences over time, record bodily changes, analyze visible and audible behavior, and sometimes measure brain activity. Each method captures a particular aspect of emotion. A scientifically defensible account therefore identifies what each instrument measures, under which conditions, and how strongly its readings support a conclusion about a person's emotional state. The classic review by Mauss and Robinson (2009) explains why experiential, physiological, and behavioral findings should be interpreted together rather than treated as interchangeable.
This article examines emotion measurement in basic psychological research: self-report, validated rating scales, ecological momentary assessment, behavioral observation, autonomic physiology, neuroscience, experimental designs, computerized inference, and the evaluation of reliability and validity. It explains what scientists can infer from each method and where interpretation reaches its limits.
What Does It Mean to Measure an Emotion?
An emotion is generally studied as an episode involving subjective experience, interpretations of a situation, bodily adjustments, changes in action readiness, and sometimes outward expression. Different theories assign different explanatory roles to these components. Research does not depend on choosing a single theory before data can be collected, but the theory determines which indicators will count as evidence for its claims. Mauss and Robinson's influential measurement review treats experience, physiology, and behavior as related yet partly independent systems.
An emotional experience is the person's felt and reportable state: for example, 'I felt anxious,' 'I was relieved,' or 'I felt both proud and sad.' Observable behavior is what a researcher can record, such as smiling, withdrawing, crying, speaking rapidly, or changing posture. Physiological responses include alterations in cardiac activity, sweating, breathing, and muscular tension. Researchers may also examine neural activity, situational appraisals, or patterns of attention. A valid measurement plan states precisely which of these is its target.
Researchers often distinguish a specific emotion category, such as anger, from more general affective dimensions. Valence concerns whether experience is pleasant or unpleasant; arousal concerns activation or alertness; intensity concerns how strong a response is. A person can report highly unpleasant but low-arousal sadness or highly unpleasant and high-arousal fear. Ratings of valence and arousal summarize features of experience while leaving open the category and meaning of an episode.
The related distinction between emotion, feeling, mood, and affect matters when reading a measure's name. A questionnaire about affect over the past week, a rating of current fear, and a diary of daily mood are asking different questions. Our guide to emotion, feeling, mood, and affect addresses these conceptual boundaries in greater depth.
The Main Methods Scientists Use
The major families of evidence are reports of experience, observational measures of expression and action, peripheral physiology, neurobiological signals, and assessments made repeatedly in everyday life. Experimental manipulations and computational models often work across these families. The practical choice depends on whether the question concerns what a person consciously feels, what their body does, how the emotion unfolds, how it influences behavior, or how reliably a group-level pattern can be predicted.
No single family of evidence automatically supersedes the others. Reports provide access to communicated subjective experience but can be shaped by language and memory. Sensors provide precisely timed physiological observations but their readings are affected by multiple processes. Expressions are social and context-sensitive; brain signals are indirect and spatially or temporally constrained. Convergence across measures can strengthen specific inferences, provided researchers have explained in advance what convergence should mean. Hollenstein and Lanteigne (2014) review the methodological problem of emotional concordance.
Self-Report: Asking People What They Feel
Self-report is central when a study's target is consciously accessible emotional experience. A participant might answer, 'How angry do you feel right now?' using a numbered scale, select labels from a checklist, describe a recent experience in their own words, or rate how pleasant and activated their current state feels. Such responses are observations of how people represent and communicate their experience at the time of assessment.
Self-report can be sensitive to distinctions that physiological sensors cannot readily establish. For example, anger and anxiety can both involve increased sympathetic activation, yet the participant may explain that one event felt insulting and another felt dangerous. Researchers must still distinguish reporting an emotion from documenting all underlying processes involved in it.
Emotion labels, rating scales, and visual measures
Category-based instruments ask whether an emotion such as joy, shame, or disgust is present, sometimes alongside an intensity rating. Dimensional instruments use scales of pleasantness, arousal, and occasionally dominance or perceived control. A widely used pictorial instrument, the Self-Assessment Manikin, offers visual judgments on pleasure, arousal, and dominance. Bradley and Lang (1994)90063-9) compared this format with verbal semantic-differential ratings and reported substantial correspondence on some dimensions. The instrument measures the participant's appraisal of their reaction, not an objectively decoded emotional identity.
Free-text accounts can reveal emotional concepts missing from a closed checklist. They also introduce challenges: responses differ in verbosity, vocabulary, social desirability, and interpretive coding. Translation and cultural meaning become especially important in multinational studies. Researchers may combine open responses with structured questions, documenting their coding rules rather than treating every word as a universally equivalent emotional category.
What self-report measures well—and what it can miss
Self-report usually offers the most direct research access to a person's stated subjective experience. Its limitations include difficulties identifying feelings, response style, demand characteristics, memory distortions, vocabulary, social context, and willingness to disclose sensitive experience. A participant may describe an experience differently depending on whether they are asked 'How distressed are you?' or 'Which emotions are you experiencing?' The timeframe changes interpretation too: 'right now,' 'today,' and 'during the last month' are different measurement targets.
Retrospective reports introduce additional reconstruction. People summarize an episode through memorable peaks, endings, and beliefs about themselves. This makes reports of a past week useful for some research questions but less precise about moment-to-moment change. Shiffman, Stone, and Hufford (2008) discuss the contrast between retrospective questionnaires and in-the-moment assessment.
A response can be informative even when a person reports 'I don't know how I feel.' That is relevant to access to, or description of, emotional experience; it is not evidence that the person has no emotion. Difficulties identifying or describing feelings are studied in relation to alexithymia, which is a psychological construct rather than an independent DSM or ICD diagnosis.
Standardized Emotion and Affect Questionnaires
A standardized measure uses specified items, response options, instructions, and scoring rules. Standardization makes replication and comparisons more feasible, while its validity always depends on the studied population, context, language, and construct. A score is meaningful only relative to the instrument's design and available evidence.
The Positive and Negative Affect Schedule, commonly called PANAS, is a major example. Watson, Clark, and Tellegen (1988) developed two ten-item scales for positive and negative affect. These broad affect dimensions should not be interpreted as a complete list of emotions, a single happiness-versus-sadness continuum, or a clinical diagnosis. The timeframe in its instructions matters: a question about how one feels at this moment serves a different purpose from one about usual feelings.
Other approaches ask participants to judge valence and arousal directly or rate particular discrete emotions after a standardized stimulus. Cowen and Keltner (2017) investigated reported responses to thousands of short videos and found evidence for a richer organization of reported emotion categories linked by graded transitions. The study shows how method and statistical representation shape the emotional structure that investigators can identify; it does not establish a fixed biological inventory of exactly 27 emotions in all settings.
An emotion questionnaire is not automatically a screening test for psychiatric illness. Negative affect may accompany ordinary stress, disappointment, grief, or frustration. Clinical diagnosis involves separate criteria and clinical judgment, including symptoms, impairment, history, and alternative explanations.
Experience Sampling and Ecological Momentary Assessment
Many emotions unfold across seconds or minutes and change with the situation. Experience sampling methods and ecological momentary assessment, usually abbreviated ESM and EMA, repeatedly ask participants about their experiences in ordinary environments. A smartphone might prompt a participant several times a day to rate current irritation, pleasantness, social context, and recent events.
EMA makes it possible to study within-person dynamics: how the same person's affect varies after conversations, during work, and across the day. Some designs prompt participants at randomized intervals; others ask for entries after particular events. Researchers can examine average levels, variability, transitions, and relationships with context. Those statistics answer distinct scientific questions and require enough observations within individuals.
Shiffman, Stone, and Hufford (2008) describe how EMA reduces dependence on distant recall and increases the ecological relevance of assessments. Trull and Ebner-Priemer (2009) explain its significance for psychological assessment and research.
Strengths and limits of sampling everyday life
An EMA study still measures responses to prompts. Participants may miss notifications, answer late, change their behavior because they are monitoring feelings, or interpret the same item differently as the day progresses. The sampling schedule may fail to catch short emotional events, and repeated questions may be burdensome or inaccessible. Inference about a person's emotion between prompts requires assumptions rather than direct observation.
A meta-analysis by Wrzus and Neubauer (2023) brought together 477 EMA articles involving 496 samples and found average compliance around 79%, with substantial design variation. That percentage is a summary across studies, not a guarantee for a new protocol. Missing data can be informative: participants might be least likely to answer during particularly intense experiences.
Researchers can improve interpretation by preregistering prompt schedules, reporting response latency, examining missingness, testing measurement stability, accommodating participants' circumstances, and separating within-person from between-person associations. If people who usually report more stress also sleep less, that between-person relationship does not necessarily describe what happens to one individual on a stressful evening.
Behavioral Observation: Faces, Voices, Posture, and Actions
Scientists study visible and audible behavior with human observers, structured coding systems, audio and video recordings, movement sensors, and, increasingly, automated feature extraction. A coding system may specify lip movements, gaze shifts, speaking rate, vocal intensity, posture, or social distance. These can be measured reliably as behaviors when operational definitions, recording quality, and coding procedures are adequate.
A smile is a visible event; happiness is an interpretation of an internal state. A raised voice can accompany anger, excitement, fear, urgency, or the acoustics of a room. Crying may occur with sadness, pain, joy, overwhelm, or social interaction. Emotion research must account for the circumstance and the person's meaning instead of converting every observed behavior into a diagnostic label.
A major interdisciplinary review, Barrett and colleagues (2019), examined evidence for inferring emotions from facial movements and documented substantial variability across people, cultures, and situations. Their analysis challenges simple one-expression/one-emotion assumptions. At the same time, facial and vocal behavior remains a genuine source of information about communication, interaction, and some emotion-related processes when contextualized and validated for the intended question.
Human coding and observer agreement
Researchers often train multiple observers and test whether they agree on clearly described movements or behaviors. Agreement about a smile, head movement, or vocal interruption is different from agreement about the psychological motive behind it. Observers can share cultural expectations and still misinterpret emotion. Good studies distinguish coders' judgments of behavior, coders' inferred emotion labels, and participants' own reports.
Video and audio analysis also raise ethical issues. Recording a participant's face, voice, or everyday social interaction can capture identifiable data about other people. Consent, secure storage, restricted access, data minimization, and clear retention rules are core methodological obligations.
Physiological Measures: What the Body Reveals
Psychophysiology measures bodily processes that may participate in emotional episodes. Common indicators include electrodermal activity, heart rate and heart-rate variability, breathing patterns, facial electromyography, pupil diameter, blood pressure, skin temperature, and hormone concentrations under carefully designed protocols. Different signals reflect different biological processes and timescales.
Electrodermal activity tracks changes in skin conductance related to sweat gland activity, often informative about sympathetic activation. Cardiac activity reflects interacting autonomic, metabolic, respiratory, and behavioral influences. Facial electromyography can detect muscle activation too subtle to see clearly on video. Pupil dilation is influenced by illumination, cognitive effort, arousal, and other factors. Cortisol varies with circadian rhythms and many stress-related and non-stress-related processes; a single sample is not a reading of a particular emotion.
These methods are valuable when investigators are interested in activation, timing, bodily regulation, recovery, or relationships between experience and physiology. Kreibig (2010) reviewed 134 publications on autonomic patterns during emotion experiments and emphasized the importance of emotional subtypes and measurement choices.
Why physiology cannot serve as an emotion detector
The same physiological change can arise during exercise, illness, caffeine intake, thermoregulation, attention, or several different emotional experiences. People also differ in their baseline physiology and response patterns. Controlled laboratory comparisons can reveal meaningful group-level associations without licensing certainty about a specific person's hidden feeling.
In a meta-analysis of 202 studies, Siegel and colleagues (2018) examined autonomic responses across emotion categories. Their results did not yield clear, invariant autonomic fingerprints separating those categories. The conclusion concerns the limitations of universal classification from autonomic patterns; it does not imply that bodily processes are irrelevant to emotion.
Wearables make large-scale ambulatory measurement possible, yet consumer metrics may use proprietary algorithms, variable signal quality, and device-specific assumptions. A smartwatch's stress estimate is best understood as a model-based inference from physiological and behavioral inputs. Its validity for subjective anger, shame, or happiness needs independent testing in the population and context of intended use.
What Brain Imaging Can Tell Us About Emotion
Electroencephalography (EEG) records electrical signals at the scalp and offers high temporal resolution for aspects of neural activity. Functional magnetic resonance imaging (fMRI) measures blood-oxygen-level-dependent changes associated with neural processes and offers a different balance of spatial and temporal resolution. Other research uses lesion evidence, stimulation, or recordings available in particular clinical and experimental settings.
These tools help investigate networks involved in salience, perception, learning, action, interoception, and reported emotional states. They cannot read subjective experience directly. Brain signals reflect multiple overlapping computations, and the same neural region can participate in emotional and nonemotional tasks.
A meta-analytic review by Lindquist and colleagues (2012) reported limited evidence for a simple one-to-one mapping between distinct emotion categories and exclusive brain locations. Subsequent scientific debates concern the organization and explanatory significance of distributed systems. A machine-learning classifier that distinguishes experimental labels from neuroimaging data is making a statistical prediction under a particular protocol, not independently proving the participant's private experience.
Researchers should report what was actually labeled: the stimulus presented, the task instruction, the person's report, or an investigator-defined condition. Those labels are not equivalent measures of emotion.
Laboratory Experiments: Eliciting and Comparing Emotional Responses
Experimental studies often present films, images, memories, social interactions, performance feedback, or other situations intended to evoke emotion. Investigators may then measure ratings, facial movement, reaction times, and physiological changes before, during, and after the event. Random assignment can help evaluate how a particular manipulation changes an outcome compared with a control condition.
The manipulation is not itself an emotion measure. Showing a frightening film establishes which stimulus was delivered; it does not establish that all participants experienced the same amount or type of fear. A participant can experience amusement, irritation, boredom, or concern during an ostensibly fear-eliciting protocol. Manipulation checks, participant reports, and relevant contextual evidence are therefore essential to interpretation.
Laboratories offer timing control, standardized equipment, and careful isolation of candidate causes. Their limitations include artificiality, expectations about the experiment, restricted sample diversity, and the possibility that an emotion episode differs outside the laboratory. Ecological studies offer complementary information but usually less control over confounding. A robust research program draws on both.
Why Different Measures Often Disagree
Someone may describe intense anxiety while showing only modest changes in heart rate. Another person may report composure while displaying marked physiological activation. An observer may code a smile while the participant reports embarrassment. Such patterns are useful data because experience, bodily physiology, and behavior can vary partly independently or unfold on different timescales.
Mauss and Robinson (2009) concluded that the response systems contain distinctive variance and do not furnish a universal gold-standard indicator. Hollenstein and Lanteigne (2014) further explain why concordance depends on models and methods. Low correlation can reflect measurement error, timing mismatch, different processes, social control of expression, or genuine differences among components.
It is a mistake to combine several signals mechanically and assume that the resulting number must be closer to the truth. Multimodal evidence helps only if each measure is valid for its target and the proposed combination is tested against appropriate criteria. Combining poorly validated indicators can produce a more complicated error.
Reliability, Validity, and the Meaning of an Emotion Score
Reliability asks whether a procedure produces sufficiently consistent results under conditions where consistency is expected. Examples include internal consistency of questionnaire items, coder agreement, and stability across repeated assessments of a relatively stable construct. A rapidly changing emotional state should not be expected to remain constant over days; demanding high long-term test–retest reliability of a momentary fear rating confuses changing experience with measurement failure.
Validity concerns whether evidence supports the intended interpretation and use of a result. Construct validity asks whether an instrument captures the theoretically specified phenomenon; content validity asks whether its content represents the target; criterion-related evidence examines meaningful external associations. Ecological validity and measurement invariance matter when results are extended to everyday life or compared across groups.
A scale can be reliable and still be a poor measure of the target emotion. A video classifier can consistently categorize facial muscle patterns while failing to establish subjective emotional states. Similarly, a questionnaire developed in one language may require more than literal translation before between-group score comparisons are justified.
Calibration, baselines, and individual differences
Physiological responses depend on age, medication, physical condition, environment, breathing, movement, and an individual's baseline. Research designs may use within-person change from baseline rather than treating identical readings across people as psychologically identical. Baselines themselves vary with anticipation and context. Calibration can improve prediction for a specified task but does not transform physiology into direct evidence of an exact emotion category.
Responses and emotion concepts also vary across language communities and social situations. Measures should be tested for semantic equivalence, comprehension, and structural comparability before researchers draw broad cultural conclusions. A universally administered questionnaire does not automatically measure a universally identical construct.
Timing and levels of analysis
A brief facial response, a heart-rate change, and a verbal report may peak at different moments. Averaging responses over a five-minute window can obscure meaningful sequences. Within-person changes, between-person differences, individual episodes, and group averages answer different questions; conclusions should specify their level. A pattern that predicts high-arousal episodes across a group may be unreliable for classifying any one person's current feelings.
Emotion Measurement and Artificial Intelligence
Automated emotion-recognition systems may analyze faces, speech, text, physiology, or multimodal recordings. What these systems predict is defined by the labels used in training and validation. Labels might come from actors' intended portrayals, annotators' impressions, experimental conditions, or participants' own reports. Each is a different source of evidence.
The ability to distinguish labeled examples in a curated dataset does not establish the ability to determine a person's subjective experience in an unfamiliar setting. The review by Barrett and colleagues (2019) is especially relevant to systems that infer emotion from facial movements. Performance also depends on demographic composition, recording conditions, cultural context, and changes between training and deployment environments.
When evaluating a purported emotion detector, researchers should ask what target it predicts, who assigned the labels, whether validation used independent participants and environments, which alternative explanations were controlled, how uncertainty is communicated, and what harms could follow from errors. A correlation between an output score and a questionnaire response supports statistical association under tested conditions; it is not direct access to an inner mental state.
High-stakes use requires particularly strong justification. Screening employees, students, patients, or vulnerable people on the basis of automated emotion labels can turn uncertain proxies into consequential judgments. The scientific question is not simply whether a model achieves high accuracy in a benchmark, but whether its intended interpretation and consequences are supported.
How Researchers Choose and Combine Methods
Suppose a study asks whether a brief social rejection increases feelings of shame. A defensible design could measure self-reported shame immediately after the event, assess embarrassment and sadness as competing experiences, document perceived evaluation by others, and observe voluntary withdrawal. Heart-rate measurement might add information about bodily activation, but would not independently identify shame. A comparison group and preregistered outcome definitions would strengthen causal interpretation of the experimental manipulation.
A study about how irritation fluctuates during everyday work would use repeated in-the-moment reports and contextual questions, with explicit plans for missing prompts and unequal sampling. A study about facial signaling in social interaction would prioritize movement coding, contextual information, and the distinctions between expressed behavior and inferred experience. A project about autonomic dynamics would require appropriate sensors, timing, baseline definition, movement controls, and physiological expertise.
A good protocol specifies the construct, timescale, population, eliciting context, primary outcome, plausible confounds, source of ground-truth labels, and evidence required for a claim. Researchers should examine whether participants understand questions as intended and document exclusions, preprocessing, and analytic choices. Where feasible, preregistration, openly described instruments, reproducible code, independent replication, and transparent reporting improve scientific accountability.
What Emotional Measures Mean Outside Research
People may encounter emotion measurement in consumer mood trackers, wearable stress scores, workplace monitoring, self-help questionnaires, or clinical consultations. A mood diary can help someone notice personally meaningful patterns. The number on a device is a representation based on a specific measurement procedure. Interpreting it sensibly requires knowing which construct was operationalized, whether the score describes a moment or a longer period, and what validation evidence exists.
Ordinary emotions are part of human life. Reporting fear, anger, sadness, or low positive affect does not by itself establish a disorder. Conversely, a low physiological arousal estimate does not invalidate a person's account of suffering. Clinical assessment, where appropriate, considers symptoms, functioning, history, context, and established diagnostic criteria rather than reducing an individual to a single emotional signal.
Frequently Asked Questions
Can emotions be measured objectively?
Scientists can objectively record many relevant events, including questionnaire responses, facial movements, heartbeats, and physiological signals. Whether those recordings establish a specific subjective emotional experience is a separate question of construct validity. No universally accepted instrument directly identifies every emotion across all people and situations. A well-defined measure can be objective about the observation while remaining limited in what it justifies inferring.
What is the most accurate way to measure feelings?
For consciously accessible subjective feelings, timely self-report is often the most directly relevant measure. Its quality depends on instructions, wording, comprehension, and context. When the research question concerns physiology or expression, sensors or behavioral coding may be more appropriate. Accuracy must be evaluated against the target of measurement.
How do psychologists measure positive and negative emotions?
Researchers use rating scales, affect questionnaires, daily sampling, behavioral observations, and physiological methods depending on the research question. PANAS measures two broad affect dimensions, while other tools measure discrete categories or valence and arousal. Positive and negative affect are not simply the two ends of one identical scale in every measurement model.
Can a smartwatch tell whether someone is anxious or angry?
Wearables can estimate physiological features and derived stress or arousal metrics. Those features can accompany anxiety or anger, but also activity, temperature, attention, and other states. Determining an exact subjective emotion would require independently validated inference under the relevant conditions, and even then uncertainty remains.
Can facial expressions reliably show what a person is feeling?
Facial movement provides meaningful behavioral and social information. It does not guarantee a one-to-one correspondence with private feeling. Context, individual habits, cultural conventions, and the purposes of an expression affect interpretation. A facial expression can be documented reliably while its emotional meaning remains ambiguous.
Does a brain scan reveal a person's emotions?
Brain imaging can identify neural activity associated with tasks and emotional processes. It does not provide a direct, unambiguous reading of a person's private experience. Statistical models can predict labels within tested datasets, with interpretations constrained by those labels and the design.
Why do different people receive different scores for the same event?
People can genuinely differ in appraisal, sensation, familiarity, physiology, attention, language, and expression. The instrument may also have measurement error or function differently across populations. Scientific interpretation separates true individual variation from differences introduced by the measure.
What is the difference between reliability and validity in an emotion test?
Reliability concerns consistency under appropriately comparable conditions. Validity concerns the evidence supporting what a score is said to mean. A consistently computed stress index can be reliable while lacking evidence that it measures subjective fear or another specific emotion.
Are emotion questionnaires used to diagnose mental disorders?
Emotion and affect questionnaires are widely used in research and sometimes as components of assessment. Their scores do not constitute a standalone psychiatric diagnosis. Diagnosis depends on appropriate professional evaluation, clinical criteria, context, and functional impact.
Why measure emotions several times a day?
Repeated sampling captures changes over time and the relationship between experiences and everyday contexts, reducing dependence on broad retrospective summaries. It also creates a burden and may miss moments when participants cannot respond. Sampling frequency should match the process under study.
Related Articles
For the scientific foundations behind the measures discussed here, see What Are Emotions? Psychological Definition, Components, and Functions, What Is Affect in Psychology? Meaning, States, and Clinical Use, and Core Affect: Valence, Arousal, and Subjective Experience. For individual variation in labeling emotional experiences, see Emotional Granularity: Meaning, Measurement, and Evidence.
References
Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements. Psychological Science in the Public Interest, 20(1), 1–68. https://doi.org/10.1177/1529100619832930
Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The self-assessment manikin and the semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25(1), 49–59. https://doi.org/10.1016/0005-7916(94)90063-990063-9)
Cowen, A. S., & Keltner, D. (2017). Self-report captures 27 distinct categories of emotion bridged by continuous gradients. Proceedings of the National Academy of Sciences, 114(38), E7900–E7909. https://doi.org/10.1073/pnas.1702247114
Hollenstein, T., & Lanteigne, D. (2014). Models and methods of emotional concordance. Biological Psychology, 98, 1–5. https://doi.org/10.1016/j.biopsycho.2013.12.012
Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological Psychology, 84(3), 394–421. https://doi.org/10.1016/j.biopsycho.2010.03.010
Lindquist, K. A., Wager, T. D., Kober, H., Bliss-Moreau, E., & Barrett, L. F. (2012). The brain basis of emotion: A meta-analytic review. Behavioral and Brain Sciences, 35(3), 121–143. https://doi.org/10.1017/S0140525X11000446
Mauss, I. B., & Robinson, M. D. (2009). Measures of emotion: A review. Cognition and Emotion, 23(2), 209–237. https://doi.org/10.1080/02699930802204677
Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological momentary assessment. Annual Review of Clinical Psychology, 4, 1–32. https://doi.org/10.1146/annurev.clinpsy.3.022806.091415
Siegel, E. H., Sands, M. K., Van den Noortgate, W., Condon, P., Chang, Y., Dy, J., Quigley, K. S., & Barrett, L. F. (2018). Emotion fingerprints or emotion populations? A meta-analytic investigation of autonomic features of emotion categories. Psychological Bulletin, 144(4), 343–393. https://doi.org/10.1037/bul0000128
Trull, T. J., & Ebner-Priemer, U. W. (2009). Using experience sampling methods/ecological momentary assessment (ESM/EMA) in clinical assessment and clinical research: Introduction to the special section. Psychological Assessment, 21(4), 457–462. https://doi.org/10.1037/a0017653
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54(6), 1063–1070. https://doi.org/10.1037/0022-3514.54.6.1063
Wrzus, C., & Neubauer, A. B. (2023). Ecological momentary assessment: A meta-analysis on designs, samples, and compliance across research fields. Assessment, 30(3), 825–846. https://doi.org/10.1177/10731911211067538