Experience sampling / Ecological Momentary Assessment (EMA)

A family of naturalistic self-report methods that signal participants during ordinary life to rate their current experience, rather than inducing a state in the lab or asking them to recall one afterward. The wiki meets it through Tong & Jia (2017), who used programmed Palm handhelds beeping ~every half hour over two days (11,141 observations from 395 people) to capture 12 positive emotions and 13 appraisals as real events unfolded.

Why it matters here: the recall contrast

EMA is best understood in the wiki against the method it displaced. Tong (2015) established the appraisal profiles of positive emotions with a recall paradigm — participants remembered a personal emotional episode and rated appraisals of the memory — and its central caveat was that recall ‘invites concerns over memory biases and schematic responses.’ Tong & Jia’s EMA study exists largely to remove that caveat: the appraisal profiles replicate when the ratings are made in the moment. That is a real methodological upgrade, and a bounded one — EMA fixes the memory problem, not the deeper correspondence problem (that appraisal–emotion co-variation is silent on causal direction, which no non-experimental method can resolve). See are-appraisals-causes-or-descriptions.

Where it sits among the wiki’s methods

EMA is the naturalistic pole of a spectrum the wiki’s emotion methods span: laboratory induction (affective-picture-viewing) controls the stimulus and measures the body but is artificial; recall and semantic norming (emotion-word-rating-norms, property-association-task) are cheap and broad but retrospective or conceptual; EMA is ecologically valid and memory-free but uncontrolled and non-experimental. The self-report-physiology-congruence method is a kindred within-person, repeated-measures design. Each trades a different thing away; the appraisal literature reaches for EMA when the worry is that lab and recall findings are artefacts of how the emotion was elicited or remembered.

The Day Reconstruction Method (DRM): a lower-burden cousin

The wiki’s second EMA source, MacCormack et al. (2021), uses not real-time signalling but the Day Reconstruction Method (Kahneman, Krueger, Schkade, Schwarz & Stone 2004) — a retrospective-but-structured relative. Rather than being beeped through the day, participants reconstruct the prior day into finite named episodes (morning/afternoon/evening, 9–15 episodes), then rate each episode’s contents from memory cues they generated themselves. It sits between recall and real-time EMA: the episode-partitioning and self-generated cues are designed to force concrete, moment-specific recall and blunt the “what usually goes with what” schematic bias that free recall invites, while the burden is far lower than weeks of beeps — which is what let MacCormack et al. run it online at scale to rate 128 items across every episode of a day.

The laboratory cousin: MDES

A third member of the family arrives with Banellis et al. (2026) and gets its own page: Multidimensional Experience Sampling, which probes ongoing thought with a fixed multi-item battery inside the lab — in that study, once retrospectively after a 14-minute resting-state scan.

MDES trades in the opposite direction from everything above. EMA leaves the lab to gain ecological validity and gives up control and any concurrent physiological or neural measure; the DRM sits between. MDES stays in the scanner, keeps the state instrumented (fMRI, PPG, respiration belt, electrogastrography), and gives up the ecological claim entirely — which in Banellis et al.’s case means the sampled “ordinary experience” is being supine and motionless in a narrow tube. Their own limitations section calls for the naturalistic replication, citing the smartphone study (Poerio et al. 2024) that found the same interoception–negative-affect association in daily life.

The reason it belongs on this page rather than folded into it: MDES is not a scheduling variant but a different unit of analysis. EMA yields many observations per person for within-person modelling; MDES in this application yields one profile per person and reduces it by factor analysis to compare across people. Same family, different question.

Behaviour-triggered sampling, and the first EMA study of interoception itself

The three variants above sample when a schedule says so (or, in the DRM, not in real time at all). Wallman-Jones et al. (2023) add a fourth: prompts fired by measured behaviour. A thigh-worn accelerometer (Movisens Move 4, 64 Hz) triggered a smartphone questionnaire after ten minutes of light physical activity or thirty minutes of uninterrupted sitting, with random prompts otherwise — ten per day, seven days, 70 participants.

Two things this buys that clock-based sampling does not:

• The contrast of interest is guaranteed to be sampled. Random beeping would catch active periods at whatever rate they occur, which in a student sample is low; event-triggering ensures enough observations at both ends to estimate a within-person slope.
• The predictor is device-measured, not self-reported. Physical activity and sedentary time come off the accelerometer while only the psychological state comes off the questionnaire, which removes the shared method variance that makes most EMA associations hard to read.

The trade is that triggering is not neutral. Being prompted because you have just moved, over and over, is a plausible route to exactly the reactivity this page lists as a limitation — and it is a sharper worry here than usual, because the construct being rated is attention to the body and the trigger is a bodily event. Nothing in the study addresses it.

It is also the wiki’s first EMA study of interoception rather than of emotion or appraisal, which is what makes it consequential beyond method: repeated measurement is what turned a construct everyone treats as a trait into one with an ICC of 0.51. See state-vs-trait-interoception. Its instrument problem is the field’s, not the design’s — there is no validated state measure of interoception, so the authors used the body subscale of the State Mindfulness Scale and wrote the request for a proper one into their limitations.

The DRM’s trade

The trade is explicit in that study’s own limitations: the DRM reduces but does not eliminate retrospective bias (only a single reconstructed day is sampled, and older adults’ episodic-memory decline is a live confound for age comparisons). It is the right tool when the question needs many densely-rated moments per person without the compliance cost of true momentary sampling, and the wrong one when day-to-day stability or genuine in-the-moment access is the point. For MacCormack et al. its role was to check whether the conceptual age pattern from their property-association study (Study 1) showed up in reconstructed daily experience (Study 2) — a different method attacking the same claim, which is exactly the corroboration EMA-family methods are reached for.