03 · What You Need to Know
How Attrition Affects Longitudinal Research and What Researchers Can Do About It
What Is Participant Attrition?
Participant attrition occurs when individuals who initially enter a longitudinal study cease participating in some or all subsequent assessments.
Attrition may occur because participants withdraw consent, become unreachable, lose eligibility under the protocol, experience competing obligations, or are unable or unwilling to continue.
Not every missed observation constitutes permanent attrition.
A participant may miss the second wave but return for the third. This produces intermittent missingness rather than permanent dropout.
Another participant may complete every questionnaire but skip a particular item. That produces item-level missingness.
Researchers should distinguish these patterns because they affect data structure and may require different analytical assumptions.
How Is Attrition Different From Other Missing Data?
| Missing-data pattern |
Description |
Illustrative example |
| Permanent attrition |
Participant stops contributing intended follow-up observations |
A student withdraws after the second wave and never returns |
| Intermittent missingness |
Participant misses one or more occasions but later resumes |
A teacher misses the midyear survey but completes the final assessment |
| Item nonresponse |
Participant completes an assessment but omits selected items |
A respondent skips the workload questions |
| Planned missingness |
Measurements are deliberately omitted under a documented design |
A supplementary module is assigned to a random subset |
| Administrative missingness |
Data are unavailable because of collection or processing problems |
Platform activity logs are not retained during a system migration |
These categories describe observable patterns or reasons for missing information. They do not, by themselves, establish the statistical missing-data mechanism.
A participant who permanently leaves may have done so for reasons unrelated to the outcome, or because of the very outcome the study investigates.
Why Does Attrition Threaten Longitudinal Validity?
Attrition can reduce the amount of information available for estimating change and may alter the composition of the observed sample.
Suppose a study examines academic motivation. Students with low motivation may be more likely to stop completing follow-up surveys.
If the final sample disproportionately contains highly motivated students, the observed average may be higher than the corresponding average in the original target population.
However, the direction of bias is not universal. If highly engaged participants become too busy to respond, the observed average could move in another direction.
The problem depends on the relationship between missingness, measured variables, unmeasured variables, and the intended estimand.
Attrition can therefore affect conclusions even when the remaining sample is numerically large.
How Do You Calculate an Attrition Rate?
A simple attrition rate compares the number of participants who no longer contribute the relevant follow-up information with the number originally eligible for that assessment.
The result describes the proportion without the specified final-wave assessment. It does not indicate why those observations are missing or establish that the resulting estimates are biased by 20%.
Researchers should specify whether retention means completing the final wave, completing every wave, or contributing sufficient data for a particular analysis.
These definitions can produce different percentages.
Is There an Acceptable Attrition Rate?
There is no universal percentage below which attrition is harmless or above which a longitudinal study becomes invalid.
A study with relatively little attrition may still be biased if the missing participants differ systematically in consequential ways.
Conversely, a study with more substantial missingness may support useful estimation when the missing-data mechanism is sufficiently understood and appropriate information remains available.
Researchers should examine the amount, timing, reasons, and potential consequences of missingness rather than rely solely on a numerical threshold.
Retention targets can be useful for operational planning, but they should be justified by the study's population, duration, outcomes, and previous evidence.
How Should Expected Attrition Affect Recruitment Targets?
Researchers often recruit more participants than the number they expect to retain at the final assessment.
A simple planning calculation can estimate the initial sample needed under an assumed retention proportion.
This calculation helps estimate recruitment needs but does not correct selection bias caused by differential dropout.
Furthermore, longitudinal models may use incomplete observations from participants who do not complete the final wave, so the number of final-wave completers is not always the appropriate basis for sample-size planning.
More detailed simulations may be necessary when dropout varies over time or differs across study groups.
Why Should Attrition Be Considered Before Recruitment?
Participant retention is partly determined by decisions made during study design.
Long questionnaires, inconvenient assessment schedules, unclear communication, and excessive follow-up requirements can make continued participation difficult.
Researchers should therefore examine whether each assessment is necessary and whether the study imposes avoidable demands.
For example, a twelve-wave study may offer excellent temporal detail but experience substantial missingness if participants cannot realistically complete monthly questionnaires.
The number and timing of observations should balance scientific value with participant feasibility.
These decisions connect directly to the number of time points required and the spacing between assessments.
What Retention Strategies Can Be Planned?
Retention strategies should make participation feasible, predictable, and respectful of participants' circumstances.
Depending on the setting and ethical approval, researchers may use reminders, flexible assessment windows, concise instruments, accessible survey platforms, and reasonable compensation for participants' time or expenses.
Participants should understand the expected duration of follow-up and how they can communicate changes in availability.
Researchers may also seek permission to maintain updated contact details or use alternative contact methods.
However, retention procedures must respect informed consent, privacy, and the right to withdraw.
Compensation should not become coercive, and contact attempts should not become intrusive.
The National Research Council emphasizes reducing missing data through study design and conduct rather than assuming statistical analysis can repair substantial losses afterward.
Should Participants Be Contacted After Missing a Wave?
Often, yes, when follow-up contact is permitted by the consent process and research protocol.
A missed assessment does not necessarily indicate permanent withdrawal.
Researchers may offer a reminder, an alternative completion window, or assistance resolving a technical problem.
However, repeated contact should follow an ethically approved procedure and respect participants' communication preferences.
Researchers should distinguish nonresponse from an explicit decision to withdraw consent.
If a participant has withdrawn permission for further contact, the study must respect that decision.
Can Participants Return After Missing an Assessment?
Yes, when the protocol permits continued participation.
Automatically excluding participants after one missed wave may unnecessarily increase permanent attrition.
For example, a student who misses the midyear survey may still contribute valuable end-of-year information.
Many longitudinal analytical methods can use unbalanced observations under appropriate assumptions.
Researchers should therefore consider allowing participants to resume follow-up rather than requiring complete participation at every wave.
Any such procedure should be defined consistently and documented in the protocol.
Should Researchers Continue Collecting Outcomes After Intervention Discontinuation?
In intervention studies, discontinuing an assigned intervention and withdrawing from outcome follow-up are different events.
A participant may stop using an educational platform but remain willing to complete subsequent assessments.
Continuing to collect outcomes can be important when the estimand concerns the consequences of assignment or an intervention policy rather than outcomes only among adherent participants.
However, continued collection must remain consistent with informed consent and the protocol.
The National Research Council's missing-data recommendations explicitly distinguish discontinuation of treatment from loss of outcome information in clinical trials. The underlying methodological distinction is also relevant to many nonclinical intervention studies.
What Are MCAR, MAR, and MNAR?
Missing-data mechanisms describe how the probability of missingness relates to observed and unobserved information.
| Mechanism |
Meaning |
Illustrative situation |
| Missing completely at random (MCAR) |
Missingness does not depend on the relevant observed or unobserved data under the specified framework |
An independent technical failure removes randomly selected survey responses |
| Missing at random (MAR) |
Missingness can depend on observed information but not on the missing values after conditioning on the relevant observed data |
Follow-up response depends on previously recorded workload and motivation, with no remaining dependence on the unobserved outcome under the assumed model |
| Missing not at random (MNAR) |
Missingness still depends on unobserved values after accounting for the relevant observed information |
Participants with unusually low unreported motivation are less likely to complete follow-up even after conditioning on available measurements |
These categories depend on the variables included in the statistical framework and cannot generally be established from observed data alone.
For example, a difference between retained and missing participants on baseline motivation may demonstrate selective response, but it does not by itself establish whether the full missing-data mechanism is MAR or MNAR.
Researchers should use substantive knowledge and observed participation patterns to develop plausible assumptions and examine their consequences.
Can Mixed-Effects Models Handle Attrition?
Mixed-effects models can often use available repeated observations from participants who have incomplete follow-up.
Under an appropriately specified likelihood model and suitable missing-data assumptions, including MAR for the relevant outcomes conditional on modeled observed information, such analyses may provide valid estimates without restricting the dataset to complete cases.
However, the model does not automatically correct all attrition bias.
If missingness depends on unobserved outcomes in ways not captured by the model, estimates may remain biased.
Researchers should distinguish the software's ability to fit an unbalanced dataset from the scientific validity of the missing-data assumptions.
When Is Multiple Imputation Appropriate?
Multiple imputation generates several completed datasets using models for the missing values, analyzes each dataset, and combines the estimates while accounting for imputation uncertainty.
It can be useful when the imputation model includes relevant observed variables associated with missingness and the missing outcomes.
However, its validity depends on the imputation model, missing-data assumptions, and compatibility with the substantive analysis.
Longitudinal data may require imputation procedures that respect repeated-measure correlations, time trends, variable types, and hierarchical structure.
Multiple imputation is not a universal remedy for missing information.
When important MNAR mechanisms are plausible, sensitivity analyses are needed to evaluate how conclusions change under alternative assumptions.
What Is Inverse Probability Weighting?
Inverse probability weighting assigns greater analytical weight to observed participants who represent others with similar measured histories but lower probabilities of remaining observed.
For example, if retention probabilities can be modeled using observed baseline and follow-up information, inverse probability weights may help address selective observation under appropriate assumptions.
However, the method requires correctly specified or sufficiently robust models, relevant predictors of retention, and adequate positivity.
Very low estimated retention probabilities can produce large weights and unstable estimates.
Researchers should evaluate weight distributions and the plausibility of the underlying assumptions rather than treat weighting as an automatic correction.
Why Is Complete-Case Analysis Often Problematic?
Complete-case analysis includes only participants with all measurements required by the analysis.
This may substantially reduce the effective sample size and discard useful partial information.
It can also introduce selection bias when the probability of having complete data is related to relevant variables.
Complete-case analysis is not always biased for every estimand under every missing-data mechanism, but its validity requires specific assumptions that researchers should justify.
It should not be chosen merely because it is the simplest option in statistical software.
Why Is Last Observation Carried Forward Usually Inadequate?
Last observation carried forward replaces missing follow-up values with a participant's most recent observed value.
The method effectively assumes that the outcome remains unchanged after the last observation for the purpose of the imputation.
That assumption is often implausible in studies designed specifically to investigate change.
It can distort estimated trajectories and underrepresent uncertainty because a single imputed value is treated as though it were observed.
More defensible methods should be selected according to the data structure and missingness assumptions.
Methodological recommendations from the National Research Council and related scholarly analyses caution against routine reliance on such simplistic approaches.
How Can Researchers Investigate Attrition Bias?
Researchers should examine who leaves, when they leave, and which observed characteristics are associated with nonresponse.
For example, baseline motivation, age, program type, academic performance, or prior survey completion may predict later participation.
Comparisons between retained and nonretained participants can reveal observable differences in sample composition.
However, such comparisons cannot establish that attrition is ignorable or identify all relevant unobserved differences.
Researchers should report missingness patterns, examine plausible mechanisms, and evaluate whether conclusions are sensitive to alternative assumptions.
What Is a Sensitivity Analysis for Attrition?
A sensitivity analysis examines whether substantive conclusions change under alternative plausible assumptions about missing outcomes.
For example, researchers may investigate what happens if participants who dropped out had systematically lower motivation than predicted by an MAR-based model.
Pattern-mixture models, selection models, delta adjustments, and other methods can support such investigations in appropriate settings.
The objective is not to identify one universally correct missing-data mechanism from incomplete observations.
Instead, sensitivity analysis shows how strongly the findings depend on assumptions that cannot be fully verified.
The National Research Council recommends evaluating robustness to plausible missing-data assumptions rather than relying on a single analysis as definitive.
How Should Attrition Be Reported?
Reports should state how many participants entered the study, how many contributed observations at each wave, and how many were included in each principal analysis.
Researchers should distinguish intermittent missingness, permanent dropout, and other reasons for unavailable observations when this information is known.
They should also describe retention procedures, analytical assumptions, and sensitivity analyses.
For observational longitudinal research, the STROBE reporting framework provides relevant guidance on participant flow, missing data, and study limitations.
Transparent reporting allows readers to evaluate the possible influence of attrition rather than infer data completeness from the final sample size alone.
How Does Attrition Affect the Generalizability of Findings?
When the retained sample differs from the original cohort, the study may describe a narrower population than intended.
For example, a university study may disproportionately retain students who remain enrolled, have reliable internet access, or maintain regular contact with the institution.
Those characteristics may also be related to academic motivation or technology use.
Even when a model appropriately estimates change among observed participants, generalizing that estimate to the original population may require additional assumptions.
Researchers should distinguish the target population, original enrolled cohort, retained sample, and analytical population.
The problem becomes particularly important when the research question concerns groups most likely to leave follow-up.
Can Attrition Be Completely Eliminated?
Usually not in studies extending over substantial periods.
Participants' circumstances change, and continued participation must remain voluntary.
The realistic objective is to reduce avoidable loss, preserve useful partial observations, document missingness, and apply analyses grounded in defensible assumptions.
Planning longitudinal data collection therefore includes planning for participants who cannot complete every intended assessment.
A study that anticipates attrition and reports its consequences transparently may be more credible than one that reports a high retention percentage without examining who was lost.
