Should Researchers Retain Raw Sensor Data?

Should You Keep the Original Sensor Recordings After Producing Your Research Variables?

You are conducting a study using wearable devices, accelerometers, physiological monitors, environmental sensors, or other digital measurement instruments. The devices generate raw recordings that are subsequently filtered, classified, aggregated, or transformed into variables for analysis.

Once you have calculated daily step counts, average heart rates, activity intensity, sleep duration, or environmental exposure, the original recordings may seem unnecessary. Retaining them can require substantial storage capacity, additional security arrangements, and ongoing data management.

However, what happens if you later discover a processing error? What if a manufacturer changes its algorithm, a reviewer questions your analytical decisions, or another researcher develops a better method for interpreting the same signals?

Should you retain everything, preserve only the derived variables, or establish a selective retention strategy?

The decision requires balancing scientific reproducibility and future analytical value against privacy, security, storage, and legal obligations.

Retain Raw Sensor Data When Their Scientific Value Justifies the Associated Responsibilities

How Raw Sensor Data Retention Affects Reproducibility, Validity, and Research Governance

What Counts as Raw Sensor Data?

Raw sensor data generally refer to measurements captured at an early stage of the acquisition process, before substantial transformations intended to produce analytical outcomes.

Examples include accelerometer samples, electrocardiographic waveforms, optical sensor signals, temperature readings, and other recorded physical quantities.

However, the term raw requires qualification. Many devices apply amplification, calibration, filtering, digitization, compression, or other processing before researchers can access their recordings.

Consequently, a manufacturer's raw-data export may already represent a processed version of the physical signal.

Researchers should document what the accessible source data contain and which transformations occurred before export.

Distinguish Raw, Processed, and Derived Data

These representations serve different purposes. Derived variables may be sufficient for a prespecified statistical analysis, while source recordings may be necessary to investigate how those variables were produced.

Retaining both can support reproducibility when the relationship between them is documented.

Why Might Researchers Need Raw Sensor Data Later?

Original recordings can support verification of processing decisions, correction of errors, and alternative analyses.

For example, a researcher may discover that an activity-classification algorithm used an inappropriate threshold. If compatible source signals remain available, the data may be reprocessed using corrected rules.

Similarly, a reviewer may question whether a particular filtering procedure removed meaningful signal components.

Access to the source recordings may allow the research team to evaluate the concern rather than repeat data collection.

However, retaining raw data does not guarantee that every error can be corrected. Information lost during acquisition cannot necessarily be reconstructed afterward.

What Scientific Benefits Can Raw Data Retention Provide?

These benefits should be assessed against the actual study requirements. A large archive without a credible plan for interpretation, security, or future use may offer limited scientific value.

Does Retaining Raw Data Guarantee Reproducibility?

No. Reproducibility requires more than possession of the original recordings.

Researchers may also need information about device models, sampling frequencies, sensor placement, firmware versions, calibration procedures, timestamps, and data-processing algorithms.

For example, acceleration samples may be difficult to interpret if the measurement units, axis orientation, or device configuration are unknown.

Similarly, an analytical result may be impossible to reproduce if the software version or classification thresholds used to generate it were not documented.

Raw data retention is therefore one component of a broader reproducible research workflow.

What Metadata Should Accompany Retained Sensor Data?

Metadata describe the conditions under which recordings were generated and the procedures needed to interpret them.

Relevant metadata may include device manufacturer and model, hardware and firmware versions, sampling frequency, measurement units, sensor orientation, calibration, acquisition dates, and relevant operating conditions.

For wearable studies, device placement, participant wear instructions, and valid-wear criteria may also be important.

Processing metadata should identify filtering, interpolation, artifact handling, aggregation, and algorithm versions.

Researchers should maintain sufficient documentation to connect source recordings with the corresponding derived variables without unnecessarily exposing participant identities.

Why Is Data Provenance Important?

Data provenance describes the origins of information and the transformations through which it passes.

For example, a daily activity total may originate from a particular device recording, undergo artifact filtering, be processed by a step-detection algorithm, and then be aggregated into a participant-level dataset.

Each transformation can affect the resulting measurement.

A provenance record may identify the source file, processing script, software version, relevant parameters, and output dataset.

This documentation makes it possible to determine which source observations contributed to an analytical result.

Without provenance, retaining raw recordings may not permit meaningful reconstruction of the research workflow.

Should Researchers Preserve Original Files Without Modification?

Maintaining a protected source copy can help preserve the integrity of the original research record.

Data cleaning and signal processing can then be performed on working copies or through reproducible computational procedures.

For example, a researcher might preserve the original accelerometer export while generating separate files containing filtered signals and daily activity summaries.

Checksums or other integrity verification methods may help detect unintended file changes.

However, preservation arrangements should follow applicable governance requirements. A source file containing unnecessary personal information should not automatically be retained indefinitely merely because it is original.

How Can Raw Data Help When Algorithms Change?

Wearable manufacturers and research software developers may update their signal-processing algorithms.

When raw or sufficiently detailed recordings are available, researchers may be able to apply one documented algorithm to observations originally processed under different software versions.

This can improve comparability when the underlying acquisition characteristics are compatible.

However, if firmware changes the sampling process, filtering, or available channels, consistent reprocessing may not fully resolve measurement differences.

Researchers should evaluate whether the source signals remain comparable before assuming that one algorithm can harmonize the dataset.

Can Raw Data Reveal Problems Hidden by Summary Scores?

Yes. Aggregated outcomes can conceal signal artifacts, missing intervals, and unusual measurement patterns.

For example, a daily heart-rate summary may appear plausible even when portions of the underlying recording contain poor optical contact or motion artifacts.

Inspecting the source signal may reveal whether the summary relied on incomplete or unreliable measurements.

Similarly, a daily activity total may obscure prolonged periods when a wearable was not worn.

Raw-data inspection can therefore support quality assurance, although the ability to detect particular problems depends on the recorded signals and available reference information.

When Might Derived Data Be Sufficient?

Derived measurements may be sufficient when the research question concerns a well-defined output whose validity and processing have been adequately established.

For example, an environmental monitoring study may use documented hourly summaries produced by a validated system when finer temporal information is unnecessary.

However, researchers should consider whether the original recordings are needed for verification, regulatory requirements, or potential correction of processing errors.

The decision should not rely solely on the convenience of smaller files.

Retention of derived data without raw recordings may be defensible when the scientific and governance requirements can still be satisfied.

How Does Data Volume Affect Retention Decisions?

High-frequency, multichannel sensor systems can generate substantial volumes of data.

The expected volume depends on sampling rate, recording duration, number of channels, data format, compression, and participant numbers.

Researchers should estimate storage requirements before collection rather than wait until the dataset exceeds available capacity.

Storage planning should include secure backups, integrity checks, access management, and eventual disposal.

Data volume alone does not justify deleting scientifically essential recordings, but it may support selecting a more efficient acquisition or retention strategy before the study begins.

Can Compression Be Used for Raw Sensor Data?

Compression can reduce storage requirements, but researchers should distinguish lossless from lossy approaches.

Lossless compression preserves the encoded information so that it can be reconstructed exactly.

Lossy compression removes or approximates information and may affect particular analyses.

For example, a compression method that preserves broad activity trends may still alter short-duration signal features important for event detection.

Researchers should evaluate whether any information loss is acceptable for the intended measurements and future verification requirements.

The selected format and compression procedure should be documented.

Should Every Sensor Channel Be Retained?

Not necessarily. Some devices collect multiple signals, only some of which may be necessary for the approved research purpose.

For example, a wearable may collect acceleration, optical signals, location information, and additional device metadata.

Retaining every channel can increase storage and privacy risks without providing corresponding scientific value.

Researchers should identify which signals are necessary for the primary analysis, quality assurance, and reasonably anticipated approved reanalysis.

Data minimization may support excluding unnecessary channels, particularly when they reveal sensitive information.

How Do Privacy Risks Differ Between Raw and Aggregated Data?

Raw sensor recordings may reveal detailed patterns that are less apparent in aggregated summaries.

For example, high-resolution location or activity traces may reveal daily routines, visits to sensitive locations, or distinctive behavioral patterns.

Physiological recordings may also contain sensitive health-related information.

Removing direct identifiers does not necessarily eliminate re-identification risks, particularly when detailed temporal patterns can be linked with external information.

Researchers should evaluate privacy risk according to the actual content and granularity of the retained recordings.

Aggregating data may reduce some risks, but the degree of protection depends on the aggregation method and surrounding information.

Does Research Consent Automatically Permit Indefinite Retention?

No. Retention and future use depend on the applicable consent arrangements, legal basis, institutional policies, and data governance requirements.

Participants may have agreed to a particular research purpose or retention period.

Some studies may permit approved secondary analyses, while others impose more restrictive conditions.

Researchers should not assume that permission to collect sensor data necessarily authorizes unrestricted storage, sharing, or future use.

Requirements vary across jurisdictions and research contexts.

Retention plans should be reviewed with the appropriate institutional or data protection authorities when necessary.

How Long Should Raw Sensor Data Be Retained?

There is no universal retention period applicable to every sensor-based study.

Requirements may depend on institutional policies, funder agreements, applicable regulations, contractual obligations, publication requirements, and the type of data collected.

Researchers should establish a documented retention schedule identifying the information to preserve, the authorized storage period, and the conditions for secure disposal.

Where multiple requirements apply, the research team should determine which obligations govern the data.

Scientific usefulness alone does not automatically justify retaining identifiable information beyond an authorized period.

Should Raw Sensor Data Be Shared Publicly?

Not automatically. Open science and reproducibility are important, but public release may be inappropriate when recordings contain sensitive or potentially identifying information.

The FAIR principles emphasize making data findable, accessible, interoperable, and reusable under appropriate conditions. Accessibility does not necessarily mean unrestricted public availability.

Researchers may consider controlled access, de-identified derivatives, synthetic examples, metadata publication, or sharing processing code when these arrangements support verification without unacceptable disclosure risks.

The appropriate approach depends on participant protections, permissions, and the scientific purpose of sharing.

Can Researchers Retain Only a Validation Subset?

Selective retention may be appropriate when preserving every recording is unnecessary or disproportionate, provided the strategy satisfies relevant scientific and governance requirements.

For example, a research team may retain a justified subset of source signals for quality assurance while preserving validated derived variables for the complete dataset.

However, a subset cannot support full reprocessing of recordings that have been discarded.

Its adequacy depends on the purpose of retention, sampling strategy, and likely future verification needs.

Researchers should make these decisions prospectively where possible rather than selectively retain observations based on the substantive findings.

What Should a Sensor Data Retention Plan Include?

A retention plan should specify which data representations will be preserved, their scientific purpose, the necessary metadata, storage arrangements, access permissions, and disposal procedures.

It should also address file integrity, backup procedures, version control, and links between source and derived datasets.

Where external service providers process or store information, their role and relevant data-use conditions should be documented.

Responsibilities should be assigned to appropriate members of the research team or institution.

The plan should be consistent with approved research protocols and applicable governance requirements.

When Should Raw Sensor Data Not Be Retained?

Retention may be inappropriate when it exceeds the authorized purpose, creates disproportionate privacy risks, violates applicable agreements, or provides insufficient scientific value to justify the associated responsibilities.

For example, retaining detailed location traces indefinitely may be difficult to justify when the approved study requires only validated daily environmental exposure summaries.

Researchers should also consider whether suitable source records are already maintained securely by an authorized institution.

However, reliance on an external archive should be supported by clear access and preservation arrangements rather than assumptions about future availability.

Preserving Accelerometer Recordings for a University Physical Activity Study

If only daily activity summaries had been retained, the team might be unable to evaluate the processing error directly.

However, retaining raw recordings would not independently establish which algorithm is more accurate. Appropriate reference evidence would still be necessary.

The example illustrates how preservation can support methodological correction and transparency when accompanied by sufficient documentation and governance.

Common Misconceptions About Raw Sensor Data Retention

How to Decide Which Sensor Recordings to Preserve

Start by identifying the role of the source signals in the study's primary outcomes, validation procedures, and reproducibility requirements.

Then evaluate the scientific benefits of preservation against storage, privacy, security, and applicable governance obligations.

When planning preservation, first examine how wearable algorithms transform source signals into research measurements.

For studies spanning system changes, consider whether software updates affect the comparability of sensor outputs.

Where retained signals may support accuracy assessment, review how to validate sensor measurements against a reference.

When storage becomes difficult, assess whether the volume of sensor data exceeds the study's meaningful analytical capacity.

The appropriate retention strategy preserves information necessary for defensible research without accumulating data whose risks and costs exceed their justified value.

Before Establishing a Raw Sensor Data Retention Policy

Frequently Asked Questions About Raw Sensor Data Retention

Preserve the Sensor Evidence Needed for Reproducibility, Within Appropriate Limits

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.