Mobile Experience Sampling Method — ESM via Smartphone
- Introduced by
- Mihaly Csikszentmihalyi & Reed Larson
- Year
- 1983–1987
- Type
- Intensive longitudinal data collection technique
- Family
- Data collection
What does it do?
The Mobile Experience Sampling Method (ESM) collects repeated, time-stamped self-reports from participants in their natural environment using a smartphone app.
By signaling participants multiple times per day over days or weeks, researchers capture psychological states, behaviors, and contexts as they occur — eliminating retrospective bias and revealing within-person dynamics that single-session surveys cannot detect.
The intuition behind the method
Traditional surveys ask 'How did you feel this week?' and rely on imperfect memory. Mobile ESM instead sends short prompts to a participant's phone several times a day — like a text saying 'What are you doing right now? How do you feel?' — and collects answers in the moment. Over many signals and many days, the researcher builds a dense picture of how mood, behavior, or context fluctuates and co-varies in real life. The mobile phone makes this feasible at scale; the repeated-measures design makes the insights ecologically valid and fine-grained.
History
The Experience Sampling Method was introduced by Mihaly Csikszentmihalyi and Reed Larson in the early 1980s, initially using electronic pagers to signal participants. Their 1987 paper in the Journal of Nervous and Mental Disease established the reliability and validity of the approach. The term Ecological Momentary Assessment, emphasizing clinical and health applications, was popularized by Saul Shiffman and Arthur Stone in the 1990s. The widespread adoption of smartphones after 2008 transformed ESM from a logistically demanding laboratory tool into a scalable mobile data collection method, giving rise to the Mobile ESM variant with its dedicated app ecosystems.
When to use it
Mobile ESM is the method of choice when the research question concerns within-person fluctuation, real-time context, or ecological validity of psychological or behavioral constructs. It is especially valuable for studying mood, stress, substance use, social interaction, or symptoms in daily life. Use it when retrospective self-report is likely to be biased by memory or social desirability. Do not use it when the population cannot use a smartphone reliably, when only between-person differences are of interest (a standard survey suffices), or when the study duration required exceeds participant tolerance.
Limitations and considerations
- High participant burden — responding to multiple daily prompts for weeks can cause fatigue and dropout.
- Requires multilevel analytic expertise; standard regression models are inappropriate for nested ESM data.
- Signal reactivity: the act of prompting participants may itself alter the construct being measured.
- Data loss from missed responses is non-random and can bias results if not handled carefully.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Captures within-person dynamics and real-time context that cross-sectional surveys miss entirely.
- Eliminates retrospective memory bias — data are collected at or near the moment of experience.
- High ecological validity: findings reflect behavior and affect in natural settings rather than artificial laboratory conditions.
- Smartphones enable passive data augmentation (GPS, accelerometer) at no additional participant burden.
- Allows separation of stable trait-level effects from transient state-level effects through multilevel analysis.
Applications
- Tracking daily fluctuations in depressive mood, anxiety, or psychotic symptoms in clinical psychology and psychiatry research.
- Studying substance use, craving, and relapse triggers in addiction research by capturing real-time antecedents.
- Measuring physical activity, sedentary behavior, and dietary choices in public health and behavioral medicine.
- Examining social interaction quality, loneliness, and emotion regulation in everyday interpersonal contexts.
- Assessing work engagement, stress, and recovery in occupational health research across the working day.
- Capturing learning behaviors, motivation, and context in educational psychology studies outside the classroom.
How it works
How it works
The researcher defines the study protocol: the constructs to measure (e.g., mood, stress, activity), the signal schedule (random within fixed intervals, fixed-time, or event-contingent), the number of daily signals (typically 4–10), and the study duration (typically 7–30 days). Brief, low-burden questionnaires — usually 5–15 items per prompt — are prepared to fit the response window (commonly 2–10 minutes).
Participants are recruited and onboarded using a dedicated mobile ESM app (e.g., m-Path, Experience Sampler, or custom builds). They complete baseline measures and receive training on the response procedure. Compliance expectations and data privacy arrangements are communicated explicitly.
At each scheduled time point, the app sends a notification. The signal schedule can be (a) random-interval: prompts delivered at unpredictable times within fixed windows so participants cannot anticipate and prepare answers; (b) fixed-interval: prompts at the same times each day; or (c) event-contingent: participant-initiated after a predefined event occurs.
The participant opens the app and completes the brief questionnaire. Responses are time-stamped and, optionally, augmented with passive sensor data (GPS location, accelerometer, screen usage). Researchers monitor compliance rates in real time; a compliance rate above 70–75% is generally considered adequate for analysis.
Data are compiled into a multilevel structure: observations nested within persons, which are often further nested within days. Multilevel modeling (MLM) or dynamic structural equation modeling (DSEM) is used to decompose within-person variability from between-person differences, estimate lagged associations, and test person-level moderators.
Common pitfalls
- Too many items per prompt: exceeding 15 items sharply reduces compliance rates; keep prompts short and focused.
- Ignoring non-compliance: analyzing only completed responses without checking whether missingness is systematic can introduce serious bias.
- Using single-level analysis on multilevel data: applying ordinary regression to ESM data inflates Type I error rates; multilevel models are required.
- Failing to account for time-of-day effects: mood and behavior vary systematically across the day; models should include time controls or within-day covariates.
- Under-powered designs: ESM requires sufficient signals per person AND sufficient persons; consult power analysis software (e.g., PANGEA) before finalising protocol.
Frequently asked questions
How many signals per day and how many days are typical?
Most ESM studies use 4–10 signals per day over 7–14 days, yielding 40–140 observations per participant. Fewer signals reduce statistical power for within-person estimates; more signals increase participant burden and attrition. Power analysis tools such as PANGEA are recommended to determine the optimal design.
What is the difference between ESM, EMA, and the daily diary method?
These terms overlap considerably. ESM (Csikszentmihalyi) typically uses random-interval signaling and emphasizes real-time capture of experience. EMA (Shiffman/Stone) is the preferred clinical-health term and includes both random-interval and event-contingent designs. Daily diary studies use a single end-of-day report rather than multiple within-day prompts. Mobile ESM specifically denotes delivery via a smartphone app, enabling passive sensor augmentation.
Is a compliance rate of 60% acceptable?
Compliance below 70% is generally considered a risk to data quality. More critically, non-compliance that is systematically linked to the construct being studied will bias estimates. Always examine compliance patterns by time-of-day, day-of-study, and participant characteristics before proceeding to the primary analysis.
What statistical model should I use to analyze ESM data?
Multilevel modeling (MLM), also called hierarchical linear modeling, is the standard approach because it correctly handles the non-independence of repeated observations within persons. For dynamic, lagged associations dynamic structural equation modeling (DSEM) is increasingly used. Standard OLS regression is not appropriate for ESM data.
Which mobile ESM apps are commonly used in research?
Widely used platforms include m-Path (KU Leuven), Experience Sampler (open source), and iESP. Commercial platforms such as MetricWire and LifeData are also used. The choice depends on budget, required passive sensing capabilities, and whether open-source auditability matters.
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.Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological momentary assessment. Annual Review of Clinical Psychology, 4, 1–32.
Open DOI referenceCopied.