03 · What You Need to Know
How to Manage Partial Data While Respecting Participant Withdrawal
What are partial research data?
Partial data are observations available for some, but not all, measurements, occasions, or study activities relevant to a research project.
In a longitudinal educational study, a participant might provide baseline achievement, motivation, and two monthly assessments but have no observations for later months.
Partial data may also arise when a participant declines a sensitive questionnaire item while completing the rest of the instrument, or stops an intervention while continuing outcome assessments.
These situations should not automatically be described as complete withdrawal from research.
The available observations may remain valuable for descriptive analyses, longitudinal estimation, or other purposes consistent with the protocol. Their usefulness depends on the research question, statistical method, and missingness assumptions.
What different forms of withdrawal should researchers distinguish?
The word withdrawal can refer to several decisions with different scientific and ethical consequences.
| Participant decision |
Meaning |
Potential implication |
| Intervention discontinuation |
The participant stops receiving or completing the assigned intervention. |
Outcome collection may continue when permitted and relevant to the estimand. |
| Withdrawal from a specific activity |
The participant declines a questionnaire, interview, assessment, or other component. |
Other approved activities may continue if the participant agrees. |
| Withdrawal from future data collection |
The participant no longer wishes to provide additional research information. |
Future collection generally stops according to the withdrawal request and applicable requirements. |
| Request concerning previously collected data |
The participant asks whether existing information can be removed or no longer used. |
Retention or deletion depends on applicable rights, lawful bases, consent terms, and research-related exceptions. |
| Loss to follow-up |
The research team cannot obtain further information despite permitted procedures. |
The observation is incomplete, but voluntary withdrawal should not be inferred without evidence. |
These distinctions should be explained during informed consent and reflected in the protocol.
Researchers should not assume that a participant who misses an appointment has withdrawn, or that someone who discontinues an intervention has declined all future contact.
Why can retaining partial data be scientifically valuable?
Previously collected observations may contain information relevant to the primary outcome, baseline characteristics, or changes over time.
For example, baseline measurements from participants who later miss follow-up can help characterize differences between those with and without observed outcomes.
Repeated measurements obtained before discontinuation may contribute to longitudinal analyses under appropriate statistical assumptions.
Discarding every observation from participants with incomplete follow-up can reduce precision and may create a selected analytical sample.
However, retaining partial data does not automatically eliminate bias. The analysis must still address why later information is missing and whether the observed measurements support the intended estimand.
Can deleting partial observations introduce selection bias?
Yes. If participants with incomplete data differ systematically from those who complete every measurement, excluding them may distort the study population and the estimated relationship.
Suppose students experiencing academic difficulties are more likely to discontinue a digital intervention. Removing all their baseline and intermediate observations could produce an analytical sample disproportionately composed of students with more favorable experiences.
If discontinuation is influenced by intervention assignment and outcome-related characteristics, conditioning on continued participation may also introduce collider-related selection bias.
These concerns are related to differences in missingness patterns between study groups.
Nevertheless, ethical restrictions on data use must be respected even when exclusion creates analytical difficulties. Scientific convenience does not establish permission to retain information.
Does withdrawal from an intervention require stopping outcome collection?
Not necessarily. Intervention discontinuation and withdrawal from research follow-up are different events.
In a randomized trial, a participant may stop receiving the assigned intervention but remain willing to provide outcome measurements. Those outcomes may be important when estimating the effect of assignment or a treatment-policy estimand.
For example, if students assigned to a tutoring program stop attending sessions, their final examination performance may remain relevant to evaluating the consequences of offering or assigning the program.
However, the appropriateness of continued follow-up depends on the research question, consent, approved procedures, and participant preferences.
Researchers should not pressure participants to continue an intervention or assessment after they have declined it.
The National Research Council's recommendations on missing data emphasize the importance of distinguishing treatment discontinuation from discontinuation of outcome collection in clinical trials.
What is the relationship between partial data and intention-to-treat analysis?
Intention-to-treat analysis generally preserves participants in their randomized assignment groups when estimating the effect of assignment, regardless of adherence or intervention discontinuation.
However, describing an analysis as intention-to-treat does not solve missing outcome data.
If a participant discontinues the intervention but provides the final outcome, that outcome can contribute to the assigned-group comparison under the relevant analysis.
If the outcome is missing, the researcher must still use an appropriate missing-data strategy and acknowledge its assumptions.
Removing participants because they did not adhere to treatment may undermine the protection of randomization and change the estimand.
At the same time, retaining participants in their original groups does not authorize continued data collection contrary to their wishes.
Can longitudinal models use incomplete participant records?
Some longitudinal statistical methods can use observations from participants who do not complete every measurement occasion.
For example, likelihood-based mixed models may use available repeated measurements under suitable model and missingness assumptions.
Multiple imputation may also incorporate observed baseline and intermediate information to support analyses of incomplete outcomes.
These approaches can be more informative than discarding every participant with a missing follow-up, but their validity is not automatic.
In particular, if later missingness depends on unobserved outcomes after conditioning on available information, conventional MAR-based analyses may remain biased.
Researchers should therefore distinguish the ability of a method to process incomplete records from its ability to identify the intended causal or descriptive quantity.
What happens when a participant requests deletion of previously collected data?
Researchers should follow the applicable consent terms, institutional policies, legal requirements, and ethics-approved procedures.
There is no universal rule that all previously collected research data must always be deleted following withdrawal, nor a universal rule that they may always be retained.
The applicable position may depend on the lawful basis for processing, whether information is identifiable or genuinely anonymized, whether results have already been aggregated, and whether a relevant research exception applies.
For example, the European Union's General Data Protection Regulation recognizes a right to erasure subject to specified conditions and exceptions, including certain scientific research circumstances under Article 17(3)(d). This is not an unrestricted permission to retain research data, and its applicability depends on the legal context.
In the Philippines, researchers must consider the Data Privacy Act of 2012, applicable National Privacy Commission guidance, institutional requirements, and the specific basis for processing personal information.
Researchers should seek guidance from their ethics committee or data-protection office when the retention status is uncertain.
Does anonymization change the retention decision?
It may. Genuinely anonymized information that cannot reasonably be linked to an identifiable individual may be treated differently from identifiable or pseudonymized personal data under applicable law.
However, removing names or replacing identifiers with study codes does not necessarily make data anonymous. Pseudonymized records may remain personal data when reidentification is possible using additional information.
Researchers should not promise that withdrawal will permit deletion of an individual's information if the data will become irreversibly anonymous and individual records can no longer be located.
These limitations should be communicated clearly during consent, before the participant decides whether to participate.
Any anonymization process should also be evaluated for reidentification risks, particularly in small or distinctive populations.
Should participants be allowed to withdraw from selected activities?
Where scientifically and operationally feasible, allowing participants to decline particular activities may support autonomy and preserve participation in other acceptable components.
For example, a participant may decline a sensitive interview but agree to complete a standardized outcome assessment.
However, not every study activity can be separated without affecting eligibility, safety, or the scientific design. Researchers should explain any genuine dependencies between activities.
Consent procedures should distinguish required eligibility conditions from optional activities and avoid presenting participation as an all-or-nothing choice when alternatives are available.
Partial participation may reduce unnecessary loss of information, but it should not be used to pressure participants into activities they would otherwise decline.
How should partial withdrawal be documented?
The research team should record the scope of the participant's decision, the date it takes effect, and the activities or data uses affected.
For example, documentation may distinguish stopping intervention sessions, declining future questionnaires, permitting administrative outcome collection, or requesting review of previously collected data.
Records should identify which observations were already collected, which future measurements remain permitted, and whether any data must be restricted or deleted under applicable requirements.
Researchers should avoid recording unnecessary personal explanations for withdrawal.
Where the reason is voluntarily provided and scientifically relevant, it may be documented with appropriate confidentiality safeguards.
This approach complements systematic documentation of reasons for missing data without assuming that every withdrawal reason establishes a statistical missingness mechanism.
How should researchers analyze retained partial data?
The analytical approach should follow the estimand, study design, available observations, and missingness assumptions.
Researchers should distinguish participants with complete outcomes, participants with partial repeated measurements, and participants whose data cannot be used because of applicable withdrawal or retention restrictions.
For longitudinal analyses, methods that use available repeated measurements may be appropriate. For missing primary outcomes, multiple imputation, weighting, likelihood-based estimation, or sensitivity analyses may be considered according to the circumstances.
Researchers should not automatically replace missing post-withdrawal outcomes with the last observed value. Last observation carried forward can impose unrealistic assumptions about outcome trajectories and produce misleading estimates.
Nor should missing outcomes automatically be coded as failures or zero scores unless that treatment follows a scientifically justified outcome definition.
Retaining valid partial data is a data-management decision; obtaining an unbiased estimate remains a separate analytical challenge.
How should partial withdrawal be reported?
Research reports should distinguish participants who discontinued interventions, missed particular assessments, withdrew from further data collection, and were lost to follow-up.
Where relevant, provide counts by study group and measurement occasion.
Explain which partial observations contributed to each analysis and why some records were excluded or unavailable.
In randomized studies, participant flow reporting should distinguish allocation, intervention receipt, follow-up, and analysis. In observational studies, reporting should explain how incomplete participation affects the analytical sample and interpretation.
Researchers should also identify relevant ethical or data-governance restrictions without disclosing unnecessary participant information.
Transparent reporting allows readers to evaluate whether the analysis reflects the intended population and how missing outcomes may affect conclusions.
