03 · What You Need to Know
How Participant Burden Leads to Missing Data and Affects Research Conclusions
What Is Attrition in Research?
Attrition refers to the loss of participants or observations during a study after initial enrollment. It is particularly relevant to longitudinal studies, repeated-measures designs, cohort studies, and intervention evaluations.
Attrition may involve permanent withdrawal, loss of contact, refusal to complete further assessments, or failure to obtain required outcome data.
However, these situations should not automatically be treated as identical. A participant may discontinue an intervention but remain willing to provide outcome data. Another may miss one assessment and return for subsequent follow-ups.
Researchers should distinguish intervention discontinuation, study withdrawal, intermittent missingness, and permanent loss to follow-up because their methodological and ethical implications differ.
How Can Participant Burden Contribute to Attrition?
Participant burden can influence retention when the time, effort, discomfort, expense, or disruption associated with participation becomes difficult to sustain.
A participant may initially agree to a study involving six visits but later find that transportation expenses or missed working hours are greater than anticipated. Someone completing daily questionnaires may experience repeated interruptions or fatigue.
These circumstances can increase the likelihood of missed assessments or withdrawal, although burden is only one possible explanation for attrition.
Health changes, relocation, loss of interest, technical problems, dissatisfaction with the research experience, and unrelated personal circumstances may also contribute.
Researchers should therefore investigate reasons for noncompletion where feasible rather than attributing all attrition to burden.
Which Types of Participant Burden May Affect Retention?
| Burden |
Potential pathway to attrition |
Possible response |
| Lengthy assessments |
Fatigue or insufficient time to complete procedures |
Review duration, breaks, and unnecessary measurements |
| Frequent follow-up |
Cumulative demands and repeated scheduling conflicts |
Justify frequency and consider suitable alternative modes |
| Transportation |
Travel expenses, distance, or unreliable transport |
Consider appropriate reimbursement or local assessments |
| Employment conflicts |
Missed wages or inability to attend fixed appointments |
Offer scientifically acceptable scheduling flexibility |
| Caregiving responsibilities |
Difficulty arranging care during research activities |
Evaluate alternative appointment arrangements |
| Digital requirements |
Connectivity problems, incompatible devices, or technical frustration |
Improve accessibility and consider alternative data collection modes |
| Emotional demands |
Distress associated with sensitive questions or procedures |
Assess risks, provide appropriate safeguards, and respect withdrawal decisions |
| Communication barriers |
Difficulty understanding instructions or obtaining assistance |
Provide clear materials and appropriate communication support |
These mechanisms may overlap. A participant with limited financial resources and inflexible employment may experience both transportation and scheduling barriers.
Why Does Attrition Matter for Scientific Validity?
Attrition can affect research through loss of precision, changes in the composition of the observed sample, and potential bias.
When fewer participants contribute outcome data, estimates may become less precise because less information is available. The consequences depend on the study design and analytical method.
Bias is a separate concern. It may arise when participants with observed outcomes differ systematically from those with missing outcomes in ways relevant to the intended estimate.
For example, a study evaluating academic stress may underestimate average stress if students experiencing the greatest difficulties are less likely to complete the final questionnaire.
However, the direction and magnitude of bias cannot be established solely from the number of missing observations.
Does Every Participant Who Drops Out Introduce Attrition Bias?
No. Attrition bias depends on the mechanism generating missing data and the estimand being investigated.
If missingness is unrelated to the relevant outcomes and variables under appropriate assumptions, the consequences may primarily involve reduced precision.
If missingness depends on outcome-related characteristics, analyses restricted to observed participants may be biased.
Researchers should not use the percentage of participants lost to follow-up as a complete measure of bias risk. Even relatively modest attrition can be consequential when it is strongly associated with outcomes, while higher attrition does not necessarily produce the same distortion under every missing-data mechanism.
The appropriate assessment requires consideration of why data are missing and how the analytical method handles them.
What Are MCAR, MAR, and MNAR?
Missing-data theory commonly distinguishes three mechanisms: missing completely at random, missing at random, and missing not at random. These terms have precise statistical meanings and should not be interpreted as everyday descriptions of whether missingness seems accidental.
Missing completely at random (MCAR)
The probability of missingness does not depend on observed or unobserved data values in the relevant complete-data model.
Missing at random (MAR)
Conditional on the observed information included in the model, missingness does not additionally depend on the unobserved values.
Missing not at random (MNAR)
Missingness continues to depend on unobserved values even after conditioning on the relevant observed information.
For example, suppose students with higher baseline stress are more likely to miss follow-up. If baseline stress is observed and adequately accounts for the missingness mechanism, an MAR assumption may be defensible for a suitable analysis.
If students are more likely to miss follow-up because their unobserved follow-up stress is especially high, even after accounting for observed information, the missingness may be MNAR.
These mechanisms generally cannot be conclusively identified from observed data alone. Researchers should use substantive knowledge, auxiliary information, and sensitivity analyses to evaluate plausible assumptions.
How Can Attrition Distort Descriptive Estimates?
In descriptive research, attrition may change the characteristics of respondents whose data contribute to a population estimate.
Suppose a longitudinal survey investigates satisfaction with a public service. Participants who become dissatisfied may be less likely to complete later questionnaires.
The average satisfaction among remaining respondents could then exceed the average among the originally enrolled population.
This is a plausible mechanism, not a universal pattern. Dissatisfied participants may also be more motivated to respond in some contexts.
Researchers should examine available characteristics of respondents and nonrespondents, evaluate changes in sample composition, and avoid assuming that complete cases adequately represent everyone enrolled.
How Does Differential Attrition Affect Randomized Trials?
Randomization supports comparability between treatment groups at assignment, but missing outcome data can complicate the interpretation of subsequent comparisons.
If the probability of obtaining outcome data differs between groups or depends on outcome-related factors, analyses of observed cases may no longer preserve the benefits of random assignment for the intended estimand.
For example, participants experiencing adverse effects may be more likely to discontinue a treatment or miss outcome assessments. Excluding their data without appropriate consideration may distort estimates of intervention effects.
Equal attrition rates between randomized groups do not guarantee the absence of bias. The reasons for missingness and their relationships with outcomes also matter.
The Cochrane Handbook's guidance on missing outcome data emphasizes evaluating the risk of bias associated with missingness rather than relying exclusively on arbitrary dropout thresholds.
Can Attrition Affect Qualitative Research?
Yes. Longitudinal qualitative studies may lose participants whose experiences differ from those who remain available for later interviews.
For example, participants who experience severe barriers to an educational program may be less willing or able to continue discussing their experiences.
Attrition may therefore narrow the range of perspectives available for interpreting change over time.
Qualitative researchers should consider whether missing later accounts affect the completeness and credibility of their interpretations. Statistical representativeness is not necessarily the appropriate standard, but the relationship between participant selection and the phenomenon remains important.
Researchers should document relevant changes in participation and avoid attributing missing experiences to particular causes without supporting evidence.
How Can Researchers Measure Attrition?
A basic retention calculation compares the number of participants with the required follow-up information against the relevant enrolled or eligible denominator.
The denominator should be defined carefully. Participants who withdraw consent, become ineligible under prespecified rules, or have different follow-up opportunities may require separate reporting, depending on the study design.
This percentage describes data availability under the stated definition. It does not establish whether the missing outcomes introduce bias.
Researchers should also report missingness at important assessment points and distinguish participant-level dropout from missing individual variables.
Why Should Researchers Examine Who Withdraws?
Comparing available baseline characteristics between participants with and without follow-up data may reveal patterns relevant to missingness.
For example, participants who miss the final assessment may differ in baseline outcomes, geographic location, employment circumstances, or intervention allocation.
These comparisons can inform the plausibility of missing-data assumptions and identify participation barriers requiring attention.
However, similar observed baseline characteristics do not prove that missingness is unrelated to unobserved outcomes.
Researchers should also avoid interpreting a nonsignificant baseline comparison as evidence that attrition is harmless. Such tests may have limited power and cannot evaluate all relevant missing-data mechanisms.
Can Reducing Participant Burden Improve Retention?
It may. Reducing unnecessary appointments, simplifying instructions, improving accessibility, or offering suitable scheduling alternatives can make continued participation more feasible.
However, retention interventions vary in effectiveness across populations and research settings. A strategy that helps one group may have little effect elsewhere.
Researchers should examine whether nonessential measurements can be removed without sacrificing important information.
Reducing burden should not involve eliminating essential safety assessments or changing validated outcome measures without appropriate evaluation.
Should Researchers Offer Flexible Follow-Up Arrangements?
Flexible scheduling, reasonable assessment windows, and scientifically appropriate alternative data collection modes may help participants remain involved when their circumstances change.
For example, a participant unable to attend an in-person questionnaire session may be able to complete an equivalent approved assessment by telephone.
However, different administration modes may influence measurement. Researchers should evaluate whether the alternatives preserve the intended outcome and whether mode differences require analytical consideration.
Scheduling decisions should also reflect participant availability and convenience without compromising necessary assessment timing.
Can More Frequent Contact Improve Retention?
Appropriate reminders and communication may help participants understand upcoming requirements and resolve practical difficulties.
However, frequent contact can itself become burdensome or intrusive. Researchers should respect participant preferences, privacy, and approved communication arrangements.
Retention contact should be distinguished from repeated outcome measurement. A reminder may support participation without requiring an additional questionnaire.
The number and timing of follow-up visits and assessments should remain linked to scientific and safety objectives rather than increasing automatically in response to concerns about dropout.
What Happens When Participants Discontinue an Intervention but Remain in the Study?
Intervention discontinuation does not necessarily mean that a participant has withdrawn consent for further outcome collection.
In randomized trials, outcome data after treatment discontinuation may be important for estimating the effect defined by the trial's estimand.
For example, a treatment-policy estimand may concern outcomes regardless of whether participants continue the assigned intervention. Obtaining appropriate follow-up data after discontinuation can therefore be scientifically important.
Researchers must nevertheless respect participants' withdrawal decisions and applicable consent requirements.
The International Council for Harmonisation's E9(R1) addendum provides a framework for defining estimands and addressing intercurrent events, including treatment discontinuation, in clinical trials.
Can Statistical Methods Correct Bias From Attrition?
Statistical methods can address missing data under specified assumptions, but no method automatically eliminates bias from an inadequately understood missingness mechanism.
Common approaches include multiple imputation, likelihood-based longitudinal models, inverse probability weighting, and sensitivity analyses.
Multiple imputation may be appropriate when a defensible imputation model and relevant assumptions support its use. Inverse probability weighting can address certain observed-data selection mechanisms when response probabilities are appropriately modeled.
Likelihood-based models may use incomplete repeated observations under assumptions concerning the missing-data process.
These methods require careful specification. They should not be selected merely because software makes them available.
Researchers should consider the estimand, outcome structure, observed auxiliary information, and plausibility of missing-data assumptions.
Why Are Sensitivity Analyses Important?
Because missing outcomes are unobserved, the assumptions required for primary analyses often cannot be verified conclusively.
Sensitivity analyses examine how conclusions change under alternative plausible assumptions about missing data.
For example, researchers may investigate whether an estimated intervention effect remains similar when missing outcomes are assumed to be systematically worse than predicted under the primary model.
Pattern-mixture models, selection models, tipping-point analyses, and other approaches may be useful depending on the design and estimand.
A sensitivity analysis should be scientifically interpretable. Simply applying several methods without explaining their assumptions does not establish robustness.
Can Increasing the Initial Sample Size Solve Attrition Problems?
Increasing enrollment can help preserve statistical precision when some loss to follow-up is anticipated.
For example, researchers may inflate a planned sample size to account for an expected proportion of participants without final outcome data.
However, this does not automatically prevent attrition bias. If missingness is systematically related to outcomes, recruiting more participants who are subject to the same selection mechanism may leave the bias unresolved.
Sample size planning should therefore be accompanied by retention strategies and an appropriate missing-data analysis plan.
How Should Attrition Be Reported?
Researchers should report participant flow, the number contributing data at relevant time points, reasons for missingness where known, and how incomplete observations were handled.
For randomized trials, CONSORT 2025 provides recommendations concerning participant flow, outcome availability, and statistical methods. For observational studies, STROBE addresses participant selection, missing data, and relevant limitations.
Researchers should distinguish the number enrolled from the number analyzed and explain why those numbers differ.
Transparent reporting helps readers evaluate the risk of bias, but it does not independently establish that the missing-data assumptions are correct.
How Should Researchers Balance Retention Efforts With Participant Autonomy?
Improving retention is scientifically valuable, but participants retain the right to withdraw according to the applicable consent and research ethics framework.
Researchers should not pressure participants to continue, imply that withdrawal is unacceptable, or use excessive contact to obtain data.
Appropriate retention efforts may include clear communication, reasonable accommodations, and support for resolving practical difficulties.
The broader ethical question concerns whether the demands of participation remain proportionate to the scientific value of the study.
Retention should be pursued through respectful procedures rather than through pressure or unnecessary incentives.
Watch Out
Do not assume that a high retention rate guarantees unbiased findings or that statistical imputation can repair every attrition problem. Evaluate who is missing, why observations may be unavailable, and whether the assumptions supporting the analysis are defensible.
