Mobile Experience Sampling — Ecological Momentary Assessment in Daily Life
- Introduced by
- Mihaly Csikszentmihalyi & Reed Larson
- Year
- 1983
- Type
- Intensive longitudinal data collection technique
- Family
- Data collection
What does it do?
Mobile Experience Sampling (ESM) is an intensive longitudinal data-collection technique in which participants respond to brief, repeated questionnaires delivered to their smartphones at random or scheduled intervals throughout the day.
By capturing thoughts, feelings, behaviors, and context at or near the moment they occur, ESM minimises retrospective recall bias and provides a high-resolution picture of psychological and behavioral fluctuations in everyday life.
The intuition behind the method
Traditional surveys ask people to reflect back over days or weeks — a process prone to memory distortion and the dominance of peak or recent experiences. ESM sidesteps this by asking short, targeted questions multiple times a day, right when and where daily life unfolds. Think of it as repeated 'snapshots' of experience rather than a single panoramic photo taken after the fact. The result is data that tracks within-person variation across time and context, enabling researchers to ask not only 'what is your average mood?' but 'how does your mood shift across different activities, people, and locations?'
History
The Experience Sampling Method was developed by Mihaly Csikszentmihalyi and Reed Larson at the University of Chicago in the early 1980s as a way to study optimal experience and daily-life psychology. Their 1987 validation paper in the Journal of Nervous and Mental Disease established foundational psychometric evidence. Shiffman, Stone, and Hufford's 2008 Annual Review of Clinical Psychology paper systematized the parallel tradition of Ecological Momentary Assessment in health research. The widespread adoption of smartphones from the late 2000s onward transformed ESM from a pager-based protocol into a scalable mobile methodology.
When to use it
Use ESM when your research question concerns within-person variability, momentary states, or the dynamics of behavior in natural contexts — questions such as how affect fluctuates across daily activities, how stress accumulates across a workday, or how social interactions shape mood in real time. It is well-suited to psychology, health sciences, organizational research, and consumer behavior. Avoid ESM when the construct of interest requires extended reflection (e.g., life-story meaning), when participants lack the literacy or motivation for repeated short questionnaires, when the study context makes frequent phone use impractical or unsafe, or when the research budget and timeline cannot support the infrastructure, compliance monitoring, and multilevel analysis that ESM demands.
Limitations and considerations
- Participant burden is high; repeated interruptions across days can cause fatigue, reactivity to assessment, and attrition.
- Non-compliance is systematic (assessments missed during demanding or private moments), potentially biasing findings.
- Data management, app infrastructure, and multilevel modeling require specialized technical expertise and resources.
- Momentary measures are brief by necessity; complex constructs requiring rich reflection cannot be adequately captured in 1–3-minute snapshots.
- Findings are specific to the sampled period and context; generalizing to other life phases or populations requires replication.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Captures momentary states with minimal retrospective recall bias, dramatically improving ecological validity.
- Enables within-person analyses that cross-sectional designs cannot provide — tracking fluctuations, lagged effects, and person-level variability.
- Produces a large number of observations per participant, giving substantial statistical power for within-person hypotheses even with moderate sample sizes.
- Naturalistic setting: data are collected in participants' real environments rather than artificial laboratory conditions.
- Flexible delivery via smartphone apps allows passive data augmentation (GPS, accelerometry, screen use) alongside self-reports.
Applications
- Tracking mood and affect variability in clinical populations (depression, bipolar disorder, anxiety) to identify within-person risk patterns.
- Studying how social interactions, physical activity, and sleep quality co-vary with well-being in daily life.
- Investigating momentary craving, substance use triggers, and relapse precursors in addiction research.
- Measuring work engagement, stress, and cognitive load fluctuations across a workday in organizational psychology.
- Capturing consumer decision-making and media use as they occur in natural consumption contexts.
- Assessing symptom dynamics and treatment response in patients with chronic pain or fatigue conditions.
How it works
How it works
The researcher defines the construct of interest (mood, stress, cravings, social interaction, etc.) and decides on a signaling scheme: random (signals occur at unpredictable moments within preset time windows), interval-contingent (fixed times each day), or event-contingent (participant triggers the assessment when a defined event occurs). The number of daily assessments and study duration are set to balance data density against participant burden; 3–10 signals per day over 7–14 days is typical.
Participants are recruited and given a dedicated mobile app or SMS-based system that delivers signals and records responses. A thorough briefing session explains the protocol, and a short practice day is usually run before the study proper begins. Scales must be brief — responses are intended to take 1–3 minutes — and validated for repeated within-person measurement.
Signals are delivered at the pre-specified times or intervals. When a signal arrives, the participant opens the app and answers the momentary questionnaire, capturing their current state, ongoing activity, social context, and location (if permitted). Responses are timestamped and uploaded automatically. Compliance and response latency are recorded for later quality control.
Response rates are tracked throughout the study period. Participants who fall below an acceptable threshold (commonly 70–75 % of signals) may be contacted with reminders or excluded from analysis. High non-compliance is not random — it tends to occur during demanding or private moments — and can introduce systematic bias that must be addressed in the analysis plan.
Because each participant contributes many observations, the data have a nested structure (assessments within persons, persons within groups). Multilevel modeling (HLM / mixed-effects models) is the standard analytic framework. Within-person effects (how a variable fluctuates around each person's mean) are separated from between-person effects (how people differ on average), enabling fine-grained hypotheses about state dynamics, lagged effects, and person-level moderators.
Common pitfalls
- Too many items per signal: questionnaires longer than 10–15 items sharply reduce compliance; keep each signal brief and validated for within-person use.
- Ignoring intraclass correlations: treating ESM data as independent observations inflates Type I error; always use multilevel or mixed-effects models that account for the nested structure.
- Confusing within-person and between-person effects: a variable that predicts outcomes at the between-person level need not do so at the within-person level (and vice versa); these effects must be explicitly separated in the analysis.
- Neglecting reactivity: the act of self-monitoring can alter the behavior or experience being studied; include a habituation period or account for reactivity in sensitivity analyses.
Frequently asked questions
How many signals per day and for how many days is reasonable?
A common protocol delivers 4–8 random signals per day over 7–14 consecutive days, yielding 28–112 assessments per participant. The optimal number depends on the intraclass correlation of your outcome (higher ICC = fewer repeated measures needed), participant burden tolerance, and the time-scale of the phenomenon. Very brief protocols (3 days) may miss weekly rhythms; very long ones (>21 days) risk substantial fatigue-related dropout.
What sample size do I need for an ESM study?
Power in ESM is a function of both the number of participants and the number of assessments per person. For typical within-person effect sizes (r ≈ 0.15–0.30), simulation studies suggest 50–100 participants with 30–50 valid assessments each provides adequate power for multilevel models. Fewer participants can be partially compensated by more assessments, but below ~30 participants between-person variance estimates become unstable.
Is ESM suitable for qualitative research?
Predominantly ESM generates quantitative data (ratings, frequencies). However, event-contingent designs can incorporate open text fields for brief narrative responses, and ESM data can be used to purposively select moments for follow-up qualitative interviews — a design sometimes called 'ESM-enriched interviews'. Pure qualitative ESM is uncommon.
What software and apps are commonly used for ESM?
Popular research platforms include m-Path, ExperienceSampler, MetricWire, and movisensXS. For clinical or larger-scale studies, custom REDCap or Qualtrics SMS integrations are also used. Choice depends on budget, need for passive sensing, and the level of programming expertise available.
How does ESM differ from the diary method?
Diary studies typically ask participants to complete one end-of-day record, capturing a daily summary. ESM captures multiple assessments throughout the day at or near the moment of experience. ESM provides finer temporal resolution and avoids daily-summary recall bias, but imposes higher participant burden. For phenomena that unfold over days (e.g., daily stressor accumulation) a diary design may suffice; for intraday fluctuations ESM is preferable.
Related methods
Sources
Csikszentmihalyi, M., & Larson, R. (1987). Validity and reliability of the Experience-Sampling Method. Journal of Nervous and Mental Disease, 175(9), 526–536.
Open DOI referenceCopied.Stone, A. A., Shiffman, S., Atienza, A. A., & Nebeling, L. (Eds.). (2007). The Science of Real-Time Data Capture: Self-Reports in Health Research. Oxford University Press.
ISBN: 978-0195178715Copied.