03 · What You Need to Know
How Nonwear and Incomplete Monitoring Affect Sensor-Based Research
What Is Wearable Device Adherence?
Wearable device adherence concerns the extent to which participants use a device according to the study's prescribed monitoring protocol.
For example, a protocol may require participants to wear an accelerometer during waking hours for seven consecutive days. Another study may require continuous monitoring, including overnight periods.
Adherence may involve wearing the device at the correct location, maintaining appropriate contact, charging it, and following instructions concerning removal.
These requirements vary by device and research purpose.
Researchers should distinguish adherence to the protocol from the mere presence of data files. A device may produce recordings even when worn incorrectly, and a correctly worn device may occasionally fail to record.
Distinguish Nonwear, Missing Data, and Genuine Inactivity
Nonwear
A period during which the participant is not wearing the device as required by the monitoring protocol.
Technical missingness
A period when valid observations are unavailable because of device, recording, storage, or transfer problems.
Genuine inactivity
A period when the participant is wearing the device appropriately but the measured activity is absent or low.
These situations can produce similar-looking sensor records.
For example, a prolonged period of near-zero acceleration may reflect sedentary behavior, sleep, or a device left motionless on a table.
Researchers should not interpret low sensor activity as proof of inactivity without considering device wear status and measurement conditions.
Why Is Nonwear a Measurement Problem?
Wearable devices can only capture the intended body-related signals when they are used under suitable conditions.
If a participant removes an activity monitor during exercise, the recorded data may underestimate activity.
If another participant removes the device only while sleeping, the missing observations may affect sleep-related outcomes more strongly than daytime activity estimates.
The consequences therefore depend on which periods are missing and what those periods would have contributed to the outcome.
Nonwear can also change the interpretation of derived daily summaries, especially when algorithms assume sufficiently complete monitoring.
What Causes Inconsistent Wearable Use?
| Potential cause |
Illustrative situation |
Possible data consequence |
| Discomfort |
Participant removes a wrist device |
Missing monitoring intervals |
| Charging requirements |
Device is removed for charging |
Repeated gaps at particular times |
| Activity restrictions |
Device cannot be worn during some activities |
Systematic undercoverage of those activities |
| Forgetfulness |
Participant does not put the device back on |
Irregular monitoring duration |
| Device fit |
Sensor contact becomes unreliable |
Artifacts or invalid measurements |
| Battery depletion |
Recording stops unexpectedly |
Technical missingness |
| Synchronization failure |
Data are not transferred successfully |
Incomplete or delayed records |
| Privacy concerns |
Participant removes the device in certain settings |
Selective absence of observations |
Researchers should investigate likely causes during feasibility testing because different mechanisms require different prevention and analytical strategies.
How Can Inconsistent Wear Bias Activity Estimates?
Activity summaries may be distorted when the recorded periods are not representative of the intended monitoring interval.
For example, a participant who wears a device primarily while sitting at work may appear less active than one who wears it throughout the day.
Simply calculating the average of observed readings may not correct this problem if missing periods differ systematically from observed periods.
Likewise, comparing raw daily step totals without considering wear time may disadvantage participants who contribute fewer monitored hours.
The appropriate correction depends on the outcome and assumptions about the missing periods.
Can Researchers Standardize Activity by Wear Time?
Some outcomes can be expressed relative to valid monitoring duration, such as activity counts per valid wear hour.
This may improve comparability when participants contribute different amounts of observation time.
However, rate standardization does not automatically eliminate bias.
If participants remove devices during unusually active or inactive periods, the observed rate may not represent the unobserved time.
Researchers should distinguish a measure of activity during observed wear time from an estimate of activity across the full intended day.
Those are different estimands and may require different assumptions.
What Is Valid Wear Time?
Valid wear time refers to monitoring duration that meets prespecified criteria for acceptable device use and usable measurements.
For example, an accelerometer study may require a minimum number of monitored hours for a day to contribute to particular daily activity estimates.
However, the appropriate threshold depends on the device, population, outcome, and monitoring design.
Some research protocols use established conventions, but these should not be treated as universal rules.
Researchers should justify their criteria using relevant methodological literature, validation evidence, and the intended temporal coverage.
How Many Valid Days Are Necessary?
There is no universal number of valid monitoring days suitable for every study.
The number required depends on day-to-day variability, the intended outcome, participant population, and desired reliability.
For example, estimating typical weekly activity may require coverage of both weekdays and weekends when behavior differs across them.
A study investigating a specific short-duration event may instead require monitoring during the relevant event period.
Researchers should evaluate whether the available days adequately represent the intended observation window.
A participant with several valid days concentrated in one type of activity may not provide equivalent evidence to a participant with observations distributed across the full period.
How Can Nonwear Be Detected?
Nonwear detection may use device-specific signals, prolonged low-variation periods, skin contact information, temperature, participant logs, or combinations of indicators.
For example, accelerometer algorithms may identify extended intervals of minimal movement as possible nonwear.
However, prolonged stillness may also occur during sleep or sedentary behavior.
Detection rules should therefore be evaluated for the device, placement, population, and monitoring protocol.
Researchers should avoid applying an algorithm developed for one device or population without considering its suitability for another.
Why Can Nonwear Detection Algorithms Misclassify Data?
Nonwear algorithms infer whether a device was worn from patterns in the available signals.
They may incorrectly classify genuine inactivity as nonwear or fail to detect removal when a device continues generating movement signals.
For example, a device placed in a moving bag may record acceleration despite not being worn.
Likewise, a participant with limited mobility may produce long periods of minimal movement while wearing the device correctly.
Researchers should evaluate the sensitivity of results to reasonable nonwear definitions and seek suitable validation evidence.
How Do Device Placement and Fit Affect Data Quality?
Inconsistent placement can change the signals captured by a wearable device.
For example, wrist and hip placement may produce different acceleration patterns, even during similar activities.
Optical sensors may require adequate skin contact for reliable measurement.
If participants wear devices loosely or move them between locations, recorded differences may reflect measurement conditions rather than genuine behavioral changes.
Researchers should provide clear placement instructions, document deviations, and evaluate whether the resulting observations remain valid.
Can Inconsistent Wear Introduce Selection Bias?
Yes. Participants who provide sufficient valid monitoring may differ systematically from those who do not.
For example, individuals with greater discomfort, demanding work schedules, limited technological familiarity, or particular health conditions may be less likely to meet wear-time requirements.
If these characteristics are related to the outcome, restricting analysis to highly adherent participants may produce a selected sample.
Researchers should compare available characteristics of participants who meet and do not meet the validity criteria.
However, observed comparisons cannot establish that missingness is unrelated to unmeasured characteristics.
What Are MCAR, MAR, and MNAR in Wearable Research?
Missing-data terminology helps clarify assumptions about why observations are unavailable.
Missing completely at random (MCAR)
Missingness is independent of both observed and unobserved data under the relevant statistical definition.
Missing at random (MAR)
Missingness can be explained by observed information under the assumed model, without depending additionally on the missing values themselves.
Missing not at random (MNAR)
Missingness remains dependent on unobserved information after conditioning on the observed information included in the model.
These mechanisms are statistical assumptions, not labels that can usually be established conclusively from observed data alone.
For example, device removal during vigorous exercise could produce informative missingness if the unobserved activity differs from recorded periods.
Researchers should investigate plausible causes and use sensitivity analyses when conclusions depend on uncertain missing-data assumptions.
Can Missing Sensor Values Be Imputed?
Missing-data methods may be appropriate when their assumptions are defensible and the available information supports estimation.
For example, certain short gaps in a slowly varying environmental signal may be handled differently from extended gaps in rapidly changing physiological activity.
However, interpolation or imputation can create misleading values when important events occur during unobserved intervals.
Replacing missing activity with zero is particularly problematic when the absence of a signal reflects nonwear rather than genuine inactivity.
Researchers should document the imputation model, assumptions, and proportion of information affected.
Where the missingness mechanism is uncertain, sensitivity analyses may be necessary.
Should Participants With Insufficient Wear Time Be Excluded?
Exclusion may be appropriate when a participant's observations cannot support the intended outcome under prespecified validity criteria.
However, excluding participants reduces the analytical sample and may introduce bias if adherence is associated with participant characteristics or outcomes.
Researchers should establish criteria before examining substantive results where possible.
They should report how many participants or monitoring days were excluded and describe relevant differences between included and excluded observations.
Alternative defensible thresholds may be examined through sensitivity analysis.
Exclusion should not be treated as a purely technical cleaning step without methodological consequences.
How Can Researchers Improve Wearable Adherence?
Adherence may be improved through clear instructions, appropriate device selection, comfortable fit, feasible charging arrangements, and timely support for technical problems.
Participants should understand when the device must be worn, when removal is permitted, and how to report interruptions.
Where appropriate and ethically approved, reminders may help participants maintain the monitoring schedule.
However, excessive reminders or device feedback may increase burden or influence the behavior being measured.
Researchers should evaluate adherence strategies during feasibility testing rather than assume that participants will follow instructions perfectly.
Should Participants Keep Wear-Time Diaries?
Wear-time diaries can provide contextual information about device removal, charging, unusual activities, and technical difficulties.
They may help distinguish some missing observations from genuine inactivity.
However, diaries introduce additional participant burden and may themselves contain recall or reporting errors.
Their value depends on the monitoring protocol and the information available from the device.
Researchers may use brief, targeted logs when the expected benefits justify the additional effort.
Can Real-Time Monitoring Improve Data Completeness?
Some devices and research platforms allow investigators to monitor whether data are being recorded or synchronized.
This may help identify battery problems, connection failures, or prolonged gaps before the monitoring period ends.
However, real-time monitoring depends on connectivity, device capabilities, and appropriate permissions.
It may also increase privacy concerns if researchers can observe detailed participant behavior continuously.
Researchers should distinguish monitoring data completeness from monitoring participants' substantive activities and apply appropriate safeguards.
How Should Inconsistent Wear Be Handled in Longitudinal Analysis?
Longitudinal wearable studies may contain different monitoring durations across participants and occasions.
Researchers should distinguish changes in the underlying outcome from changes in wear-time coverage.
For example, an apparent decline in daily activity could reflect reduced wear time rather than reduced movement.
Analyses may require appropriate standardization, repeated-measures modeling, or missing-data procedures depending on the estimand.
However, statistical adjustment cannot automatically recover unobserved activity when missingness is systematically related to the outcome.
The analysis should account for the observation process as well as the recorded measurements.
How Should Wear-Time Completeness Be Reported?
Researchers should describe the intended monitoring schedule, valid wear criteria, nonwear detection procedures, and handling of incomplete observations.
Reports should indicate the number of participants contributing valid data, the distribution of monitoring duration, and relevant exclusions.
Where completeness differs across groups or time periods, those differences should be examined and reported.
Software versions and processing algorithms should be documented when they influence wear-time classification.
Transparent reporting allows readers to evaluate whether the observed data adequately represent the intended monitoring period.
When Does Inconsistent Wear Make a Study Inadequate?
Inconsistent wear may seriously compromise a study when essential periods are systematically unobserved, too few participants provide usable data, or the missingness cannot be addressed under defensible assumptions.
For example, a study investigating physical activity during exercise sessions may be inadequate if participants routinely remove their devices during those sessions.
In such circumstances, the remaining data may describe activity during device wear but not the full activity pattern the study intended to investigate.
The researcher may need a different device, a revised protocol, additional measurement sources, or a narrower question.
Watch Out
Do not treat missing wearable readings as zero activity or assume that participants with more recorded hours are necessarily more active. Wear-time differences can change the meaning of sensor summaries and may systematically distort comparisons.
