Dr. Egon Dejonckheere
Dr. Egon Dejonckheere is an Assistant Professor at Tilburg University, affiliated with KU Leuven, specializing in ambulatory assessment and intensive longitudinal research. His work combines methodological innovation to improve these methods with applications in mental health and emotion dynamics, and includes dozens of smartphone-based studies in daily life.
Choosing a sampling scheme in ESM or EMA
A sampling scheme determines when participants complete assessments in experience sampling methodology (ESM) or ecological momentary assessment (EMA) studies. Time-based schemes schedule surveys at fixed, random, or semi-random moments. Event-based schemes link assessments to an experience, behavior, or context, with participants or passive sensors initiating the survey. Burst designs concentrate assessments into particular periods, while hybrid designs combine approaches.
The choice matters because different schedules give you access to different parts of daily life. As such, the same momentary survey may answer different research questions depending on when it is administered.
This guide compares the main ESM and EMA sampling schemes and explains their practical and analytical trade-offs. The best choice depends on the process you want to understand, how quickly it changes, and what participants can realistically complete.
The choice matters because different schedules give you access to different parts of daily life. As such, the same momentary survey may answer different research questions depending on when it is administered.
This guide compares the main ESM and EMA sampling schemes and explains their practical and analytical trade-offs. The best choice depends on the process you want to understand, how quickly it changes, and what participants can realistically complete.
What is a sampling scheme?
In Experience Sampling Methodology (ESM) or Ecological Momentary Assessment (EMA), your sampling strategy usually refers to determining the assessment moments within participants. It concerns the question of when and how momentary surveys should be delivered.
Defining a sampling scheme in ESM or EMA is structured around three design decisions.
Defining a sampling scheme in ESM or EMA is structured around three design decisions.
- A trigger: What initiates an assessment? A time rule, a participant action, or a detected event?
- The frequency: How many assessments are planned, or how many eligible events do you aim to sample?
- The study structure: Are assessments collected evenly over consecutive days, or structured in sequences around certain events?
In practice, you can mix and match the answers to these three questions to create an ESM or EMA sampling scheme that fits your research question. The sampling schemes discussed below should therefore not be seen as mutually exclusive, but as approaches that can be combined.
Time-based sampling
Time-based Experience Sampling Methodology (ESM) or Ecological Momentary Assessment (EMA) schedules the assessments according to a clock or elapsed-time rule. Surveys are sent whether or not a particular experience has occurred. This makes it useful for observing continuous processes or frequently present states such as mood, fatigue, or the intensity of psychological complaints in daily life.
The timing can be fixed, random, or semi-random.
The timing can be fixed, random, or semi-random.
Fixed sampling
Fixed sampling in ESM or EMA sends surveys at predetermined times, such as 09:00, 13:00, 17:00, and 21:00 each day. Another option is an assessment every two hours during a defined waking period (sometimes called interval-contingent sampling).
For example, a workplace study may compare concentration at 10:00 and 15:00 throughout two working weeks. Those two moments have a clear meaning within the research question.
For example, a workplace study may compare concentration at 10:00 and 15:00 throughout two working weeks. Those two moments have a clear meaning within the research question.
Advantages
Predictable prompts can make assessments easier for participants to anticipate and incorporate into their daily routine, which may reduce burden and support adherence to the ESM or EMA study.
Repeatedly assessing participants at the same times also facilitates comparisons across days because observations are aligned to the same points in the daily cycle. This can be especially useful when the research question in ESM or EMA concerns systematic diurnal patterns, such as differences between morning, afternoon, and evening experiences.
When prompts are equally spaced, fixed sampling also produces observations with a consistent time lag, which can be advantageous for statistical models that rely on lagged associations. In practice, however, the actual interval between observations may still slightly vary because ESM or EMA participants do not always respond immediately, prompts may be missed, and the overnight interval is typically much longer than within-day intervals.
Repeatedly assessing participants at the same times also facilitates comparisons across days because observations are aligned to the same points in the daily cycle. This can be especially useful when the research question in ESM or EMA concerns systematic diurnal patterns, such as differences between morning, afternoon, and evening experiences.
When prompts are equally spaced, fixed sampling also produces observations with a consistent time lag, which can be advantageous for statistical models that rely on lagged associations. In practice, however, the actual interval between observations may still slightly vary because ESM or EMA participants do not always respond immediately, prompts may be missed, and the overnight interval is typically much longer than within-day intervals.
Disadvantages
The predictability of fixed sampling schemes is also their main disadvantage. Participants may anticipate upcoming prompts and alter their behavior, activities, or attention in response, creating a potential source of measurement interference with natural behavior.
Fixed schedules can also produce systematic selection or sampling biases in the situations that are observed. Because the same times are sampled repeatedly, the ESM or EMA design may disproportionately capture recurring routines while underrepresenting experiences that occur at other or less predictable moments. For example, a prompt scheduled every day around lunch time may repeatedly assess similar eating, work, or social contexts.
The same issue applies to nonresponse: if an ESM or EMA participant has a recurring class, meeting, commute, or caregiving obligation at a scheduled assessment time, missingness may become systematically associated with particular contexts rather than occurring at random. This can bias your conclusions. As a result, fixed sampling is less well suited when the primary aim is to obtain a broadly representative sample of experiences across the full range of everyday situations.
Fixed schedules can also produce systematic selection or sampling biases in the situations that are observed. Because the same times are sampled repeatedly, the ESM or EMA design may disproportionately capture recurring routines while underrepresenting experiences that occur at other or less predictable moments. For example, a prompt scheduled every day around lunch time may repeatedly assess similar eating, work, or social contexts.
The same issue applies to nonresponse: if an ESM or EMA participant has a recurring class, meeting, commute, or caregiving obligation at a scheduled assessment time, missingness may become systematically associated with particular contexts rather than occurring at random. This can bias your conclusions. As a result, fixed sampling is less well suited when the primary aim is to obtain a broadly representative sample of experiences across the full range of everyday situations.
When to use
Fixed sampling schemes are best suited for ESM or EMA studies in which the timing of assessments is itself substantively meaningful. Examples include comparing experiences at specific points in the day, such as morning versus evening mood, or examining processes over a predefined interval between assessments.
Fixed sampling is also appropriate when researchers want all participants to be assessed according to the same temporal structure, and when the impact of potential measurement interference is not of great concern. Before choosing this scheme, make sure that the selected assessment times correspond to the moments that are most relevant to the research question.
Fixed sampling is also appropriate when researchers want all participants to be assessed according to the same temporal structure, and when the impact of potential measurement interference is not of great concern. Before choosing this scheme, make sure that the selected assessment times correspond to the moments that are most relevant to the research question.
Random sampling
Random sampling in ESM or EMA selects assessment times by chance within an eligible period.
For example, an app could independently draw six survey times uniformly between 08:00 and 22:00 each day. Assessments are scheduled across the full eligible waking period without first dividing that period into smaller blocks.
For example, an app could independently draw six survey times uniformly between 08:00 and 22:00 each day. Assessments are scheduled across the full eligible waking period without first dividing that period into smaller blocks.
Advantages
Because participants do not know exactly when a prompt will arrive, random sampling reduces the opportunity to anticipate upcoming assessments and adjust behavior accordingly. Assessment times also vary from day to day, allowing ESM or EMA studies to sample experiences across a wider range of moments and situations than schedules that repeatedly assess participants at the same clock times.
When assessment times are drawn uniformly from the eligible period, each eligible moment has the same probability of being selected. This can help avoid systematically prioritizing particular times of the day. Across a sufficiently large number of observations, random sampling can therefore provide broad coverage of participants' daily experiences.
When assessment times are drawn uniformly from the eligible period, each eligible moment has the same probability of being selected. This can help avoid systematically prioritizing particular times of the day. Across a sufficiently large number of observations, random sampling can therefore provide broad coverage of participants' daily experiences.
Disadvantages
Because assessment times are selected independently, ESM or EMA prompts can cluster together by chance. Several prompts may occur relatively close to one another, while another part of the day contains few or no assessments. Random sampling therefore does not guarantee that observations will be evenly distributed within each day.
The unpredictability of prompts may also make assessments more disruptive or difficult to accommodate than a predictable schedule. In addition, the time intervals between consecutive observations vary, which needs to be accounted for when examining temporal or lagged relations.
Random sampling also does not guarantee that all relevant situations will be observed. Rare or short-lived events can still occur entirely between prompts. Moreover, completed observations may still become selective if participants are systematically less likely to respond in certain situations.
The unpredictability of prompts may also make assessments more disruptive or difficult to accommodate than a predictable schedule. In addition, the time intervals between consecutive observations vary, which needs to be accounted for when examining temporal or lagged relations.
Random sampling also does not guarantee that all relevant situations will be observed. Rare or short-lived events can still occur entirely between prompts. Moreover, completed observations may still become selective if participants are systematically less likely to respond in certain situations.
When to use
Random sampling is most appropriate when researchers want assessment moments to be selected randomly from the full eligible period and do not require each part of the day to be represented within a given day.
In practice, this is a relatively specific use case in ESM or EMA research, where researchers often want to ensure that assessments are distributed across different parts of the day. In those cases, semi-random sampling is typically the more appropriate choice.
In practice, this is a relatively specific use case in ESM or EMA research, where researchers often want to ensure that assessments are distributed across different parts of the day. In those cases, semi-random sampling is typically the more appropriate choice.
Semi-random sampling
Semi-random sampling in ESM or EMA divides the day into predefined blocks and randomly schedules assessments within each block. It is also called stratified random or blocked random time sampling.
For example, divide 08:00 to 22:00 into seven two-hour blocks and schedule one survey per block. This distributes planned assessments somewhat evenly across the day while retaining uncertainty about their exact timing.
For example, divide 08:00 to 22:00 into seven two-hour blocks and schedule one survey per block. This distributes planned assessments somewhat evenly across the day while retaining uncertainty about their exact timing.
Advantages
Semi-random sampling combines structured coverage across the day (analogous to fixed sampling) with variation in exact assessment times (analogous to random sampling). Because each predefined block contributes an assessment, researchers can ensure that different parts of the ESM or EMA day are sampled rather than leaving their coverage entirely to chance.
At the same time, the random timing within each block makes prompts less predictable than in fixed sampling. This can reduce repeated sampling of exactly the same routines and limit participants' ability to anticipate precisely when an assessment will occur.
Compared with fully random sampling, the predefined blocks also reduce the chance that several prompts cluster within one part of the day while other periods remain unsampled. Adding a minimum time gap can further prevent closely spaced interruptions.
At the same time, the random timing within each block makes prompts less predictable than in fixed sampling. This can reduce repeated sampling of exactly the same routines and limit participants' ability to anticipate precisely when an assessment will occur.
Compared with fully random sampling, the predefined blocks also reduce the chance that several prompts cluster within one part of the day while other periods remain unsampled. Adding a minimum time gap can further prevent closely spaced interruptions.
Disadvantages
Semi-random sampling does not eliminate predictability completely. If an ESM or EMA participant knows that one prompt must occur within a particular block and none has arrived yet, they may increasingly anticipate the assessment toward the end of that block.
Prompts can also occur close together across adjacent block boundaries. For example, one assessment could be scheduled at 09:59 and the next at 10:01. A minimum time gap can prevent this, but it also changes the sampling procedure and should therefore be reported clearly.
As with random sampling, the intervals between consecutive assessments remain unequal. Temporal or lagged analyses should therefore account for the actual elapsed time rather than automatically assuming that consecutive observations are equally spaced.
Finally, similar to other time-based sampling schemes, brief or rare events may occur entirely between assessments and would then have to be recalled retrospectively. Semi-random sampling is consequently less suitable when the research question focuses on discrete events whose timing cannot be predicted in advance.
Prompts can also occur close together across adjacent block boundaries. For example, one assessment could be scheduled at 09:59 and the next at 10:01. A minimum time gap can prevent this, but it also changes the sampling procedure and should therefore be reported clearly.
As with random sampling, the intervals between consecutive assessments remain unequal. Temporal or lagged analyses should therefore account for the actual elapsed time rather than automatically assuming that consecutive observations are equally spaced.
Finally, similar to other time-based sampling schemes, brief or rare events may occur entirely between assessments and would then have to be recalled retrospectively. Semi-random sampling is consequently less suitable when the research question focuses on discrete events whose timing cannot be predicted in advance.
When to use
Semi-random sampling is particularly suitable for ESM or EMA studies that involve states or processes that are continuously present, when researchers want broad coverage across the day without making exact prompt times predictable.
Choose the number and width of the sampling blocks, as well as any minimum interval between prompts, based on the time scale of the process being studied. Our complete guide on how to run an ESM or EMA study provides a more in-depth discussion of these design choices.
Choose the number and width of the sampling blocks, as well as any minimum interval between prompts, based on the time scale of the process being studied. Our complete guide on how to run an ESM or EMA study provides a more in-depth discussion of these design choices.
Event-based sampling
Event-contingent sampling in Experience Sampling Methods (ESM) or Ecological Momentary Assessment (EMA) links an assessment to the occurrence of a predefined event, such as a social interaction, a meal, an argument, physical activity, or arrival at a particular location. Rather than time-based sampling, assessments are scheduled in relation to specific events that are relevant to the research question.
Event frequency is not under the researcher’s control and may differ substantially between participants. This introduces practical and statistical considerations: researchers need to decide whether to fix the study duration, the number of events per participant, or some combination of both. Your decision may affect power calculations, participant burden, and compensation, especially when some participants need much longer than others to provide enough event reports.
Another important design choice is when the ESM or EMA assessment occurs relative to the event: before, during, immediately after, or after a delay. These timings capture different information. For example, a report after an argument can assess the event and its aftermath, but studying what preceded it requires measurements collected beforehand rather than retrospective reconstruction.
Event-contingent sampling designs can be participant-initiated or sensing-triggered.
Event frequency is not under the researcher’s control and may differ substantially between participants. This introduces practical and statistical considerations: researchers need to decide whether to fix the study duration, the number of events per participant, or some combination of both. Your decision may affect power calculations, participant burden, and compensation, especially when some participants need much longer than others to provide enough event reports.
Another important design choice is when the ESM or EMA assessment occurs relative to the event: before, during, immediately after, or after a delay. These timings capture different information. For example, a report after an argument can assess the event and its aftermath, but studying what preceded it requires measurements collected beforehand rather than retrospective reconstruction.
Event-contingent sampling designs can be participant-initiated or sensing-triggered.
Participant-initiated sampling
Participant-initiated sampling asks people to start an ESM or EMA survey when a predefined event occurs. For example, participants who are asked to log social interactions in daily life may tap a button after a conversation lasting at least five minutes. The event definition should be sufficiently clear that participants can recognize when a report is required.
Advantages
Participant-initiated sampling allows ESM or EMA researchers to target specific events that may be too rare or brief to capture with time-based prompts. It is particularly useful for delineated events or episodes that are meaningful to the participant but difficult to detect automatically, such as interpersonal conflicts, binge episodes, or subjective emotional experiences.
Because participants can report shortly after the event, the delay between the experience and the assessment can remain relatively short, thereby reducing recall demands compared with asking about the same event much later.
Because participants can report shortly after the event, the delay between the experience and the assessment can remain relatively short, thereby reducing recall demands compared with asking about the same event much later.
Disadvantages
Participant-initiated sampling places considerable responsibility and burden on participants. They must recognize that the event has occurred, remember the study instructions and event definition, and initiate the assessment themselves. This may be especially difficult during stressful, emotionally intense, or otherwise demanding situations.
Participants’ interpretation ofthe event definition may differ or change over time, which can introduce selection bias in which events are reported. Moreover, reports completed after an event remain retrospective. Even with a short delay, participants already know how the event unfolded, which can influence how they evaluate what they experienced during it. For example, after a panic attack has passed, knowing that the feared outcome did not occur may affect reports of perceived threat during the episode.
A further challenge is that unreported events are usually invisible to the researcher. An empty event log may mean that no event occurred, but it may also mean that an event occurred and was not reported. This makes compliance harder to determine than in time-based sampling.
Participants’ interpretation ofthe event definition may differ or change over time, which can introduce selection bias in which events are reported. Moreover, reports completed after an event remain retrospective. Even with a short delay, participants already know how the event unfolded, which can influence how they evaluate what they experienced during it. For example, after a panic attack has passed, knowing that the feared outcome did not occur may affect reports of perceived threat during the episode.
A further challenge is that unreported events are usually invisible to the researcher. An empty event log may mean that no event occurred, but it may also mean that an event occurred and was not reported. This makes compliance harder to determine than in time-based sampling.
When to use
Participant-initiated sampling is particularly suitable for clearly defined, discrete events that participants can recognize and report shortly after they occur, especially when such events would be captured too infrequently by time-based sampling. If the research question also concerns what happens before the event or how participants recover afterwards, combine event reports with time-based scheduled assessments or follow-up measurements using burst designs.
Because the absence of an event report is ambiguous, ESM or EMA researchers may consider adding scheduled checks to assess whether relevant events occurred without being reported. Alternatively, you can provide time-based reminders that repeat the event definition, so participants are periodically reminded which events they should report and are less likely to forget to initiate an assessment.
Because the absence of an event report is ambiguous, ESM or EMA researchers may consider adding scheduled checks to assess whether relevant events occurred without being reported. Alternatively, you can provide time-based reminders that repeat the event definition, so participants are periodically reminded which events they should report and are less likely to forget to initiate an assessment.
Sensing-triggered sampling
Sensing-triggered sampling automatically initiates a survey when changes in sensor data derived from
passive sensing meet a predefined trigger rule. This is also sometimes called context-aware ESM or EMA surveying.
Examples include entering a predefined geographic area, such as a bar or museum, coming into proximity with a Bluetooth beacon, for example to indicate social presence, or detecting a period of physical activity based on heart rate or movement data. More complex systems can combine multiple signals to estimate whether a relevant state is likely to occur, such as craving or heightened stress.
Examples include entering a predefined geographic area, such as a bar or museum, coming into proximity with a Bluetooth beacon, for example to indicate social presence, or detecting a period of physical activity based on heart rate or movement data. More complex systems can combine multiple signals to estimate whether a relevant state is likely to occur, such as craving or heightened stress.
Advantages
Sensing-triggered sampling can capture events or contexts more objectively without relying on participants to recognize them and initiate an assessment themselves. Assuming wearable devices are worn correctly, and mobile sensor coverage is adequate, this can make events visible that would otherwise go unreported in participant-initiated designs.
It can also bring self-report assessments closer in time to relevant events or contexts that time-based prompts may miss. In addition, the sensor stream can provide more objective information about what happened before, during and after the event, such as changes in location, movement, or physiology. These data do not replace self-report in ESM or EMA, but they can complement it by adding information about the event and its precursors that participants may not notice or report themselves.
It can also bring self-report assessments closer in time to relevant events or contexts that time-based prompts may miss. In addition, the sensor stream can provide more objective information about what happened before, during and after the event, such as changes in location, movement, or physiology. These data do not replace self-report in ESM or EMA, but they can complement it by adding information about the event and its precursors that participants may not notice or report themselves.
Disadvantages
The quality of sensing-triggered sampling depends heavily on the accuracy and reliability of the detection system. False negatives mean that relevant events are missed, whereas false positives generate assessments when the intended event did not actually occur. Technical issues such as missing sensor data, battery restrictions, synchronization delays, or operating-system limitations can further affect which events result in a prompt.
Moreover, a detected signal is not necessarily equivalent to the psychological state of interest. This is called the mind-body problem. An elevated heart rate, for example, could reflect exercise rather than stress. A short confirmation question can therefore be useful to establish whether the intended event or context was actually present.
Sensing-triggered designs in ESM or EMA also require more technical development, testing, and monitoring than conventional time-based schedules and may involve privacy-sensitive sensor information. Finally, because assessments are concentrated around detected events, they provide limited information about ordinary moments outside those contexts unless combined with another sampling scheme.
Moreover, a detected signal is not necessarily equivalent to the psychological state of interest. This is called the mind-body problem. An elevated heart rate, for example, could reflect exercise rather than stress. A short confirmation question can therefore be useful to establish whether the intended event or context was actually present.
Sensing-triggered designs in ESM or EMA also require more technical development, testing, and monitoring than conventional time-based schedules and may involve privacy-sensitive sensor information. Finally, because assessments are concentrated around detected events, they provide limited information about ordinary moments outside those contexts unless combined with another sampling scheme.
When to use
Sensing-triggered sampling is most appropriate when the target event or context can be detected with sufficient accuracy and timeliness, and researchers want ESM or EMA assessments close to those moments without relying on participants to initiate them.
It is especially useful when objective information about the occurrence, timing, or context of an event is valuable alongside participants’ subjective reports. Typical examples include location transitions, activity episodes, or other sensor-detectable behaviors.
It is especially useful when objective information about the occurrence, timing, or context of an event is valuable alongside participants’ subjective reports. Typical examples include location transitions, activity episodes, or other sensor-detectable behaviors.
Burst sampling designs
Burst designs in Experience Sampling Methods (ESM) or Ecological Momentary Assessment (EMA) concentrate repeated measurements into selected periods of intensive assessment.
A burst describes when intensive measurement periods occur, but not how assessments are scheduled within those periods. Each burst therefore still requires a sampling scheme, such as fixed, semi-random, participant-initiated, or sensing-triggered sampling.
Depending on the time scale, bursts can be delivered across months or years, or around an event or short-term trajectory.
A burst describes when intensive measurement periods occur, but not how assessments are scheduled within those periods. Each burst therefore still requires a sampling scheme, such as fixed, semi-random, participant-initiated, or sensing-triggered sampling.
Depending on the time scale, bursts can be delivered across months or years, or around an event or short-term trajectory.
Measurement bursts across months or years
A longitudinal ESM or EMA study may collect six semi-random surveys per day for seven days and then repeat the same assessment week three and six months later. This allows researchers to study both short-term variation within each burst and longer-term change across bursts.
Advantages
Repeated bursts make it possible to connect intensive short-term dynamics with longer-term developmental or clinical change. Because intensive assessment is concentrated into selected periods, researchers can study processes over months or years without requiring continuous ESM or EMA measurements throughout the entire period. This significantly reduces participant burden.
When bursts are designed comparably, they can also reveal whether characteristics of daily life, such as average levels, variability, or temporal associations, change over time or remain stable.
When bursts are designed comparably, they can also reveal whether characteristics of daily life, such as average levels, variability, or temporal associations, change over time or remain stable.
Disadvantages
The periods between bursts remain largely unobserved, so researchers may detect that a change occurred without knowing exactly when or how it developed.
Repeated bursts also require participants to remain engaged with the study over a relatively long period, sometimes for months or years. This increases the risk of attrition across successive bursts and should be anticipated when planning recruitment and statistical power. Selective dropout can also reduce comparability across ESM or EMA waves if participants who remain differ systematically from those who leave the study.
In addition, individual bursts may coincide with seasonal effects, holidays, examinations, illness, or other unusual periods, which can be mistaken for longer-term change. Analyses also need to reflect the additional hierarchical structure created by repeated observations within bursts and bursts within participants.
Repeated bursts also require participants to remain engaged with the study over a relatively long period, sometimes for months or years. This increases the risk of attrition across successive bursts and should be anticipated when planning recruitment and statistical power. Selective dropout can also reduce comparability across ESM or EMA waves if participants who remain differ systematically from those who leave the study.
In addition, individual bursts may coincide with seasonal effects, holidays, examinations, illness, or other unusual periods, which can be mistaken for longer-term change. Analyses also need to reflect the additional hierarchical structure created by repeated observations within bursts and bursts within participants.
When to use
Repeated measurement bursts across months or years are particularly suitable when the research question concerns how everyday patterns change over longer timescales, for example across development, treatment, relapse, or other transitions, while still requiring intensive measurement of short-term processes. Bursts should be timed and repeated often enough to capture the expected pace of longer-term change.
Short intensive bursts
Short intensive bursts consist of several closely spaced assessments collected over a relatively short period. They can begin at a predefined time or be triggered by an event or detected state.
For example, an ESM or EMA study may repeatedly assess participants during the first hour after waking or after a stressful interaction. The goal is not to represent the entire day, but to obtain dense coverage of a short-term trajectory.
For example, an ESM or EMA study may repeatedly assess participants during the first hour after waking or after a stressful interaction. The goal is not to represent the entire day, but to obtain dense coverage of a short-term trajectory.
Advantages
Closely spaced assessments allow researchers to capture how a process unfolds over minutes or hours, including its onset, escalation, persistence, or recovery. Unlike traditional time-based sampling, which often prioritizes an accurate representation of an entire day, short ESM or EMA bursts prioritize dense coverage of a limited time window. This provides much finer temporal resolution without requiring such intensive sampling throughout the entire day.
Unlike a single event-based assessment, short intensive bursts do not merely capture whether or how an event occurred, but how the participant’s state changes across multiple moments surrounding or following it.
Unlike a single event-based assessment, short intensive bursts do not merely capture whether or how an event occurred, but how the participant’s state changes across multiple moments surrounding or following it.
Disadvantages
Short bursts can be highly burdensome while they are active, particularly during demanding or emotionally intense periods. If bursts are event-triggered, repeated or overlapping events can launch competing sequences and require clear rules for prioritizing or suppressing assessments (a so-called “buffer window”).
Because ESM or EMA measurement is concentrated within selected windows, little is learned about periods outside the burst. Interpretation therefore depends strongly on how the burst is initiated and whether relevant measurements before or after the trajectory are available.
Compliance within the burst is particularly important. A participant may technically respond to the burst by completing its first assessment, but if later follow-ups are missed, the intended trajectory cannot be estimated. Researchers should therefore consider not only overall compliance, but also completion across the sequence of assessments within each burst.
Because ESM or EMA measurement is concentrated within selected windows, little is learned about periods outside the burst. Interpretation therefore depends strongly on how the burst is initiated and whether relevant measurements before or after the trajectory are available.
Compliance within the burst is particularly important. A participant may technically respond to the burst by completing its first assessment, but if later follow-ups are missed, the intended trajectory cannot be estimated. Researchers should therefore consider not only overall compliance, but also completion across the sequence of assessments within each burst.
When to use
Short intensive bursts are most appropriate when the aim is to characterize a short-term trajectory in detail (e.g., describe intense emotional episodes) rather than obtain representative coverage across the day. The spacing and duration of the ESM or EMA assessments should match the expected speed and duration of the process being studied.
Hybrid designs
Finally, a hybrid design combines two or more sampling schemes within the same Experience Sampling (ESM) or Ecological Momentary Assessment (EMA) study, with each component serving a different purpose.
For example, semi-random prompts can sample everyday mood across the working day, participant-initiated reports can capture stressful interactions, and selected meetings can trigger a short recovery burst. Hybrid designs in ESM or EMA are therefore useful when a single sampling scheme cannot provide all observations needed to answer the research question.
For example, semi-random prompts can sample everyday mood across the working day, participant-initiated reports can capture stressful interactions, and selected meetings can trigger a short recovery burst. Hybrid designs in ESM or EMA are therefore useful when a single sampling scheme cannot provide all observations needed to answer the research question.
Advantages
Hybrid designs can combine the strengths of different ESM or EMA sampling schemes. Time-based assessments can provide coverage of ordinary daily life and comparison moments, while event-based assessments can target specific or infrequent episodes. Short follow-up bursts can then capture how experiences change after those episodes. Together, these components can provide information about everyday states, the occurrence of specific events, and short-term trajectories surrounding them.
Time-based assessments can also provide observations before an event happens or reveal events that participants did not report themselves, helping to address some of the blind spots of purely event-based sampling.
Time-based assessments can also provide observations before an event happens or reveal events that participants did not report themselves, helping to address some of the blind spots of purely event-based sampling.
Disadvantages
Combining several sampling rules makes the ESM or EMA protocol more complex and burdensome. Prompts from different components may overlap or accumulate around events, so researchers need clear rules for prioritization, minimum time gaps, or buffer windows.
More importantly, the different components sample different types of moments. Event-triggered assessments deliberately overrepresent target events, whereas time-based assessments are intended to sample everyday experience more broadly. Pooling all observations without considering how they were selected can therefore produce misleading estimates.
Suppressing scheduled prompts after a target event can also introduce bias, because moments close to those events are then systematically excluded from the time-based sample. Simply including prompt type as a variable does not necessarily remove this bias.
More importantly, the different components sample different types of moments. Event-triggered assessments deliberately overrepresent target events, whereas time-based assessments are intended to sample everyday experience more broadly. Pooling all observations without considering how they were selected can therefore produce misleading estimates.
Suppressing scheduled prompts after a target event can also introduce bias, because moments close to those events are then systematically excluded from the time-based sample. Simply including prompt type as a variable does not necessarily remove this bias.
When to use
Hybrid sampling schemes in ESM or EMA are most appropriate when the research question requires both broad coverage of everyday experience and targeted information about specific events or trajectories. Each component should have a clear methodological purpose and contribute information that could not be obtained as effectively from the other components alone.
Researchers should record which sampling rule generated each assessment and define in advance how the different components will be analysed. Estimates of everyday averages may, for example, need to rely primarily on the time-based component, whereas event-related analyses should reflect the selective nature of event-triggered observations.
Researchers should record which sampling rule generated each assessment and define in advance how the different components will be analysed. Estimates of everyday averages may, for example, need to rely primarily on the time-based component, whereas event-related analyses should reflect the selective nature of event-triggered observations.
An interactive sampling scheme selector
Choosing a sampling scheme starts with the research question and the type of moments or processes you need to observe. To make this choice more concrete, the step-by-step decision aid below walks you through a small number of questions and points toward sampling schemes that may fit your study.
Note that this wizard should be treated as a starting point rather than prescriptive methodological guidance. ESM or EMA designs involve many interdependent decisions, and many design choices are not included in this flow chart. The suggested scheme should therefore always be evaluated in the context of the full ESM or EMA study design.
Settings that matter across all sampling schemes
Match the timing to the process
Choose the interval based on how quickly the process is expected to unfold. Hourly assessments might describe broad daily patterns while missing a change that rises and resolves within ten minutes. Longer-term change may require repeated observation periods.
Define the eligible assessment period
Define the time periods relevant to your research question, including waking hours, workdays, weekends, and any exclusions. For instance, if you are studying workplace experiences, there may be no need to sample evenings or weekends. Conversely, if the aim is to characterize everyday life more broadly, restricting assessments to working hours would leave important parts of the day unobserved.
Assessment periods may differ for different ESM or EMA participants, for example to match individual work schedules or waking hours, as long as those differences are consistent with the study aims and accounted for in the analysis.
Assessment periods may differ for different ESM or EMA participants, for example to match individual work schedules or waking hours, as long as those differences are consistent with the study aims and accounted for in the analysis.
Distinguish prompt from response time
A survey scheduled for 14:00 and completed at 14:25 may describe a different situation. Consider how long an ESM or EMA survey remains available, whether reminders occur, and what time (window) the questions refer to. Inspect planned, delivered, and actual response timestamps where available, plus event timestamps for event-based designs.
Specify burden controls and their consequences
For some ESM or EMA sampling schemes daily limits, minimum time gaps, and priorities for overlapping assessments are required. Describe what happens to prompts that cannot be sent. For example, always sampling the first three events of a day favours earlier events, while randomly selecting eligible events would create a different sample. Burden controls change the observation process as well as the participant experience.
Plan for missingness and unequal intervals
Distinguish unanswered prompts from periods that were intentionally unsampled. For event-based designs, also distinguish no event from undetected or unreported events whenever possible.
For temporal models, a 10-minute interval and a three-hour interval should not be treated as equivalent. Even fixed schedules develop unequal gaps through delayed responses, missed prompts, and overnight breaks. Ignoring unequal timing can bias estimated dynamics. Use an analysis that handles the timing adequately, which may include continuous-time modelling.
For temporal models, a 10-minute interval and a three-hour interval should not be treated as equivalent. Even fixed schedules develop unequal gaps through delayed responses, missed prompts, and overnight breaks. Ignoring unequal timing can bias estimated dynamics. Use an analysis that handles the timing adequately, which may include continuous-time modelling.
Pilot the complete ESM or EMA protocol
Test the participant experience and inspect the resulting data. Check completed assessments, response delays, time-of-day coverage, event yield, technical failures, and days with unusually high burden. An attractive schedule on paper may behave differently when all its rules operate together.
Dr. Egon Dejonckheere
Egon Dejonckheere is an Assistant Professor at Tilburg University, affiliated with KU Leuven, specializing in ambulatory assessment and intensive longitudinal research. His work combines methodological innovation to improve these methods with applications in mental health and emotion dynamics, and includes dozens of smartphone-based studies in daily life.
FAQ
01.
What is the difference between random and semi-random sampling in ESM or EMA?
Random sampling draws assessment moments across the whole eligible sampling day. Semi-random sampling first divides that day into blocks and then draws moments within each block. Both sampling strategies produce irregular intervals, but semi-random sampling controls coverage across the sampling day.
02.
Can I combine time-based and event-based sampling in the same ESM or EMA study?
Yes. Hybrid designs are often useful when researchers need both representative everyday observations and targeted observations around specific events.
03.
Does participant-initiated event-based sampling in ESM or EMA eliminate recall bias?
No. Even when participants report shortly after an event, they are still looking back on something that has already happened. Their report may also be influenced by knowing how the episode ended, which can change how they evaluate what they felt or expected during the event.
04.
How far apart should ESM or EMA prompts be?
There is no universal minimum. The interval should reflect the timescale of the process, participant burden, and the statistical model, while avoiding unnecessary clustering.
05.
Should a daily diary study adopt a fixed time-based sampling design?
Often, yes. Daily diaries are commonly scheduled at a fixed time, such as every evening, especially when the aim is to compare the same period across days. However, the sampling time and recall period are separate design choices: a diary completed at 21:00 can ask about the whole day or about the participant’s current state, and some studies may instead allow participant initiation at an individually chosen time such as bedtime.