Attrition refers to the loss of participants or required observations after recruitment or enrollment, depending on how the study defines and measures participation over time.
For example, students may enroll in an AI tutoring experiment but fail to complete the final mathematics assessment.
In a longitudinal study, participants may complete the baseline survey but miss later measurement occasions.
Attrition can reduce the amount of usable information and, depending on its causes, introduce bias into estimates or comparisons.
Researchers should define what counts as attrition in the particular study rather than assume that every missing response or missed activity represents the same event.
These concepts overlap but are not interchangeable.
For example, a student who declines the initial invitation is a recruitment nonrespondent, while a student who completes the baseline assessment but not the follow-up may contribute to attrition.
A participant who remains in the study but skips one questionnaire item has missing data without necessarily having dropped out.
These distinctions matter because different problems may require different recruitment and analytical responses.
When a study requires a specified number of completed observations and assumes a constant anticipated attrition proportion, a common recruitment adjustment is:
For example, if 200 completed observations are required and 20% attrition is anticipated:
Worked Example
Nrecruit = 200 / (1 − 0.20)
Nrecruit = 200 / 0.80 = 250
The recruitment target is therefore 250 participants under the stated assumptions.
The calculation is appropriate as a simple planning adjustment, but more complex study designs may require additional considerations.
Why Is Adding 20% Not the Same as Adjusting for 20% Attrition?
A common error is to multiply the required completed sample by 1.20 when anticipating 20% attrition.
Suppose 200 completed observations are needed.
Approach
Calculation
Expected completed observations at 20% attrition
Add 20%
200 × 1.20 = 240
240 × 0.80 = 192
Divide by retention
200 / 0.80 = 250
250 × 0.80 = 200
The first approach produces fewer expected completed observations than required because the 20% loss applies to the recruited sample, not to the original required sample.
The second approach correctly accounts for the anticipated retention proportion under the simple model.
AI may perform either calculation depending on how the prompt is interpreted. Researchers should therefore verify the result by multiplying the proposed recruitment target by the expected retention rate.
Why Must the Attrition Rate Be Expressed as a Proportion?
The formula uses attrition as a decimal proportion rather than a whole-number percentage.
For example, 15% attrition is represented as 0.15, and 30% attrition is represented as 0.30.
Substituting 20 instead of 0.20 would produce an invalid result.
Researchers should also confirm that the attrition proportion is between zero and one, with a value below one required for a finite recruitment target under the simple formula.
AI-generated calculations should display the conversion explicitly when percentages are provided.
How Does the Recruitment Target Change With Different Attrition Rates?
Consider a study requiring 200 completed observations.
Anticipated attrition
Expected retention
Minimum recruitment target
5%
95%
211
10%
90%
223
15%
85%
236
20%
80%
250
25%
75%
267
30%
70%
286
Each result is rounded upward to the next whole participant.
These are illustrative calculations, not recommended attrition rates for every study.
The appropriate planning value depends on the research setting, follow-up duration, participant characteristics, and relevant prior evidence.
Where Should the Expected Attrition Rate Come From?
Researchers should justify anticipated attrition using information relevant to the proposed study.
Possible sources include previous studies involving comparable populations and procedures, institutional participation records, pilot research, or documented experience with similar follow-up arrangements.
For example, attrition observed in a one-week online survey may not be an appropriate estimate for a six-month intervention requiring repeated assessments.
Similarly, retention patterns among working adults may differ from those among university students participating during an academic semester.
AI should not recommend a universal 10%, 15%, or 20% attrition assumption without acknowledging that it is a planning scenario requiring justification.
Can AI Help Estimate Attrition When No Previous Data Exist?
AI can help develop sensitivity scenarios using several plausible attrition rates.
For example, a researcher might compare recruitment requirements under 10%, 20%, and 30% attrition to understand how retention uncertainty affects the planned sample.
However, these scenarios should not be mistaken for empirical predictions.
When no reliable estimate is available, the researcher may need to examine the study's retention risks, consult relevant methodological literature, or conduct preliminary feasibility work.
The uncertainty should be acknowledged in the sample size justification rather than concealed behind a precise numerical recommendation.
Does Attrition Always Occur at the Same Rate Across Study Groups?
No. Attrition may differ between intervention and comparison conditions.
For example, students assigned to AI tutoring may discontinue participation because of technical difficulties, while students in the comparison condition may have fewer intervention-related demands.
If the study requires a particular number of completed observations in each group, different anticipated retention rates may require different recruitment targets.
AI should therefore not automatically apply one overall attrition rate when the study has reason to expect differential loss.
However, unequal recruitment targets must be considered alongside the planned allocation procedure and statistical analysis.
How Do You Adjust Sample Size When Groups Have Different Attrition Rates?
Suppose a two-group study requires 100 completed observations in each condition.
The researcher anticipates 20% attrition in the intervention group and 10% attrition in the comparison group.
Illustrative Group-Specific Adjustment
Intervention group: 100 / 0.80 = 125 participants.
Comparison group: 100 / 0.90 = 111.11, rounded upward to 112 participants.
Total recruitment target: 237 participants.
This calculation assumes that the required completed sample is 100 per group and that the group-specific retention estimates are appropriate.
However, the resulting recruitment allocation is unequal.
If the research protocol requires equal randomization, the researcher must consider whether the allocation plan should be revised or whether a different recruitment strategy is necessary.
Group-specific adjustments should therefore be integrated with the original sample size and allocation plan rather than applied mechanically.
Can Attrition Adjustments Be Used in Longitudinal Research?
Yes, but longitudinal studies may require more detailed assumptions about retention across measurement occasions.
For example, a study measuring academic engagement at baseline, three months, and six months may experience different patterns of missingness at each stage.
Some participants may miss one assessment and return later, while others may permanently withdraw.
These patterns are not equivalent to a single constant dropout rate.
Researchers should identify which observations are required for the intended analysis and whether the statistical method can appropriately use incomplete longitudinal records.
For complex longitudinal designs, simulation-based planning may be more appropriate than a single attrition multiplier.
Does Attrition Adjustment Guarantee Statistical Power?
No. Increasing the recruitment target improves the expected number of completed observations under the assumed retention rate.
However, actual attrition may exceed expectations, and the loss of observations may affect precision or power differently depending on the study design.
Furthermore, some analytical methods can use partially observed data under specified assumptions, while others may require particular complete measurements.
The simple adjustment does not automatically capture these differences.
Researchers should distinguish the anticipated completed sample from the effective information available for the intended analysis.
Does Recruiting More Participants Eliminate Attrition Bias?
No. Attrition can introduce bias when the probability of missingness is related to outcomes, treatment conditions, or other relevant characteristics in ways not adequately addressed by the analysis.
For example, students struggling academically may be more likely to discontinue an AI tutoring intervention.
If only students who complete the intervention are analyzed, the estimated outcomes may disproportionately reflect participants who experienced fewer difficulties.
Recruiting additional participants may increase the number of completed observations without correcting this selection process.
Little and Rubin (2019) provide a foundational treatment of missing-data mechanisms and statistical approaches for incomplete data.
Researchers should therefore plan both retention strategies and appropriate missing-data analyses.
What Are MCAR, MAR, and MNAR?
Missing-data methodology commonly distinguishes three mechanisms.
Mechanism
Meaning
Important consideration
Missing completely at random (MCAR)
Missingness is independent of observed and unobserved data values under the relevant model.
A strong assumption that may not be plausible in many longitudinal studies.
Missing at random (MAR)
Conditional on observed information, missingness does not depend on the missing values.
Can support certain likelihood-based or multiple-imputation approaches under additional assumptions.
Missing not at random (MNAR)
Missingness remains related to unobserved values after conditioning on observed information.
May require explicit missingness models or sensitivity analyses.
These mechanisms are statistical assumptions, not labels that AI can reliably assign from a brief description.
Researchers should examine the data collection process, available information, and plausible reasons for missingness.
In particular, MAR and MNAR generally cannot be conclusively distinguished from observed data alone without additional assumptions.
Can AI Recommend Retention Strategies?
AI can suggest practical measures to reduce participant loss, including clear communication, manageable assessment schedules, appropriate reminders, technical support, and reducing unnecessary participant burden.
However, retention procedures must be ethically appropriate and consistent with the study protocol.
For example, researchers should avoid pressuring students to remain in a study or creating the impression that withdrawal will affect their academic standing.
Retention strategies should preserve voluntary participation and comply with institutional requirements.
The goal is to reduce avoidable loss without compromising participant rights.
Can AI Confuse Attrition With Initial Recruitment Nonresponse?
Yes. A chatbot may use the terms dropout, nonresponse, and attrition interchangeably.
However, recruiting participants who never respond to the initial invitation differs from losing enrolled participants during follow-up.
For example, if the study requires 200 completed participants, anticipates 20% attrition after enrollment, and expects only half of invited eligible individuals to enroll, the planning process involves separate stages.
Illustrative Multistage Recruitment
Required completed observations: 200.
Required enrolled participants at 80% retention: 200 / 0.80 = 250.
Approximate invitations at 50% enrollment: 250 / 0.50 = 500.
This calculation assumes the stated rates apply to the corresponding stages and that invited individuals are eligible.
It should not be interpreted as a universal recruitment formula.
AI should distinguish the number invited, number enrolled, and number expected to provide usable outcomes.
Can AI Calculate Attrition Adjustments for Cluster-Randomized Studies?
AI can help identify the relevant considerations, but cluster-randomized studies may require adjustments at more than one level.
Individual participants may withdraw, and entire clusters such as classrooms or schools may also be lost.
These forms of attrition can have different effects on statistical power and the number of independent allocation units.
For example, losing an entire school may be more consequential than losing the same number of students distributed across several schools.
A simple participant-level attrition adjustment may therefore be insufficient.
Researchers should use a design-specific calculation that accounts for the cluster structure and anticipated loss mechanisms.
What Should Be Reported in the Sample Size Justification?
A transparent justification should identify the sample size required for the intended analysis, the anticipated attrition rate, the basis for that assumption, the adjustment method, and the resulting recruitment target.
For multi-group or longitudinal studies, additional details may include group-specific retention assumptions and the number of usable observations expected at each relevant measurement occasion.
Researchers should also describe important limitations of the assumptions.
For example, a 20% attrition rate obtained from a comparable previous study may provide a reasonable planning estimate, but it does not guarantee that the same rate will occur.
AI can help draft the explanation, but the numerical and methodological details must be independently verified.
How Can You Audit an AI-Generated Attrition Adjustment?
1. Confirm the required analytical sample. Identify how many usable observations are needed before accounting for attrition.
2. Define attrition operationally. Specify which missing observations or withdrawals count as loss for the planned analysis.
3. Justify the expected attrition rate. Use relevant evidence or clearly labeled planning scenarios.
4. Apply the appropriate adjustment. Divide the required completed sample by the expected retention proportion under the simple model.
5. Verify the arithmetic and rounding. Confirm that the expected completed sample meets the analytical requirement.
6. Examine design complexity. Consider differential attrition, repeated measures, clustering, and allocation constraints.
7. Plan missing-data handling. Identify analytical assumptions and procedures for incomplete observations.
8. Document the recruitment rationale. Report the inputs, adjustment method, assumptions, and final target.
The adjustment should be treated as one component of sample size planning rather than a complete solution to participant loss.
Watch Out
Do not assume that increasing the recruitment target eliminates attrition-related bias. A study may retain enough participants for statistical analysis while still producing distorted estimates if missingness is systematically related to the outcomes or intervention conditions.
04 · A Practical Example
Correcting an AI Attrition Adjustment for a Semester-Long AI Tutoring Study
Hypothetical Example
Recruitment Planning for an AI Tutoring Experiment
A researcher plans a semester-long randomized study comparing AI tutoring with usual mathematics instruction.
The verified sample size calculation indicates that 100 participants with usable final achievement scores are required in each group.
The researcher anticipates 20% attrition and asks AI to adjust the recruitment target.
AI recommends adding 20 participants to each group, producing 120 participants per condition.
The researcher verifies the calculation.
Planning element
AI's initial calculation
Corrected calculation
Required completed observations per group
100
100
Anticipated attrition
20%
20%
Recruitment adjustment
100 × 1.20
100 / 0.80
Recruitment target per group
120
125
Expected completed observations per group
96
100
Total recruitment target
240
250
The researcher adopts 125 participants per group as the simple attrition-adjusted target, provided the 20% assumption is defensible.
However, the researcher also considers whether students experiencing technical problems or lower achievement may be more likely to withdraw.
If attrition is related to the outcomes, the resulting data may require appropriate missing-data analysis and sensitivity assessment.
The decision: The researcher corrects the arithmetic while recognizing that the recruitment adjustment does not resolve all methodological consequences of attrition.
The example demonstrates why attrition planning requires both numerical verification and an understanding of the processes that may produce missing observations.
05 · What Researchers Often Get Wrong
Common Mistakes in AI-Generated Attrition Adjustments
Misconception
Adding 20% Is the Same as Adjusting for 20% Attrition
Adding a percentage to the required completed sample generally produces too few expected completions. The simple adjustment divides by the anticipated retention proportion.
Misconception
Every Study Should Assume 10% or 20% Attrition
Attrition assumptions should reflect the population, procedures, follow-up duration, and relevant evidence rather than a universal default.
Misconception
Attrition and Initial Nonresponse Are the Same
Failure to enroll after an invitation differs from loss after enrollment. Recruitment planning may require separate assumptions for each stage.
Misconception
Recruiting More Participants Eliminates Missing-Data Bias
Additional recruitment may increase the number of usable observations but does not automatically correct systematic missingness related to outcomes or exposures.
Misconception
Attrition Is Always Equal Across Experimental Groups
Intervention demands, participant experiences, and implementation conditions may produce differential loss. Group-specific assumptions may be necessary.
Misconception
One Attrition Formula Works for Every Longitudinal or Clustered Design
Complex studies may involve intermittent missingness, cluster loss, or different retention patterns. Design-specific calculations may be required.
06 · What This Means for You
Ask AI to Calculate Expected Retention, Not Merely Add a Dropout Percentage
A useful attrition adjustment begins with the number of usable observations required by the original sample size calculation.
AI should then identify the anticipated retention proportion, calculate the recruitment target, and explain which assumptions support the adjustment.
A simple decision framework
If a single attrition rate applies under a simple design
Divide the required completed sample by the anticipated retention proportion.
If attrition may differ across groups
Evaluate group-specific recruitment requirements and implications for allocation.
If repeated measurements are required
Examine the timing and pattern of missing observations rather than assuming one constant loss rate.
If participants are recruited in clusters
Consider both individual-level attrition and potential loss of entire clusters.
If missingness may be related to outcomes
Plan appropriate missing-data analyses and sensitivity assessments in addition to recruitment adjustments.
A Prompt for Calculating Attrition-Adjusted Sample Size
Example Prompt
"My randomized study requires 100 participants with usable final outcome data in each of two groups. I anticipate 20% attrition over one semester. Calculate the recruitment target using the appropriate retention-based adjustment, show each step, and round upward where necessary. Verify the result by calculating the expected number of completed observations. Explain the assumptions underlying the attrition rate and distinguish recruitment adjustment from missing-data bias. Identify whether differential attrition, repeated measurements, or clustering would require a more complex calculation. Do not simply add 20% to the original sample size."
The resulting adjustment should be checked against the original sample size assumptions and independently verified before recruitment begins.
Researchers should also document how retention will be monitored and how incomplete observations will be addressed during analysis.
The purpose is to recruit enough participants for the intended study while recognizing that the quantity and quality of the final data are separate concerns.
07 · A Quick Checklist
Before Accepting an AI Attrition-Adjusted Sample Size
Before finalizing the recruitment target, check:
Is the original required analytical sample independently justified?
Have I defined what counts as attrition for the intended analysis?
Is the anticipated attrition rate supported by relevant evidence or clearly labeled as a scenario?
Has the adjustment divided by the retention proportion rather than simply adding the attrition percentage?
Have I verified the arithmetic and rounded recruitment targets upward?
Does the expected completed sample meet the analytical requirement?
Have I considered differential attrition between study groups?
Are repeated measurements or clustered observations handled appropriately?
Have I distinguished initial recruitment nonresponse from post-enrollment attrition?
Is there an appropriate plan for missing-data analysis and retention monitoring?
08 · Frequently Asked Questions
Frequently Asked Questions About AI and Attrition Adjustments
Can ChatGPT calculate sample size adjustments for attrition?
Yes. It can perform the arithmetic and explain the method, but the expected attrition rate, design assumptions, and resulting recruitment target require verification.
What is the formula for adjusting sample size for attrition?
Under a simple constant-retention assumption, divide the required completed sample by one minus the anticipated attrition proportion.
If I expect 20% attrition, should I add 20% to my sample?
No. If 200 completed observations are required, dividing by 0.80 gives a recruitment target of 250. Simply adding 20% gives 240, which would yield only 192 expected completions.
What attrition rate should I assume?
Use relevant prior evidence, institutional experience, or justified planning scenarios. No single attrition rate is appropriate for every study.
Should attrition be calculated separately for experimental groups?
Potentially, especially when retention is expected to differ between conditions. The resulting targets must also be considered alongside the allocation and analysis plan.
Does increasing sample size eliminate attrition bias?
No. Recruiting additional participants may help preserve sample size, but missingness related to outcomes or exposures may still introduce bias.
Can the same formula be used for longitudinal studies?
A simple adjustment may be useful for preliminary planning, but complex missingness patterns and repeated measurements may require design-specific methods or simulation.
What if the actual attrition rate exceeds the estimate?
The final usable sample may be smaller than planned. Researchers should evaluate the implications for precision, power, missing-data assumptions, and interpretation rather than assume that the original calculation remains satisfied.
09 · The Bottom Line
Recruit for Expected Retention, but Plan for the Consequences of Missing Data
The Bottom Line
Generative AI can help calculate attrition-adjusted recruitment targets, but the adjustment must use the correct retention-based method and a defensible estimate of participant loss. Increasing recruitment does not automatically resolve the statistical or methodological consequences of missing data.
Use AI to verify the arithmetic and compare plausible retention scenarios. The final recruitment plan should account for the study's design, expected completion rates, and the possibility that missing observations may affect the validity of the findings.
10 · Sources and Further Reading
Sources and Further Reading
Little, R. J. A., & Rubin, D. B. (2019)
Statistical Analysis with Missing Data
White, I. R., Horton, N. J., Carpenter, J., & Pocock, S. J. (2011)
Strategy for Intention to Treat Analysis in Randomised Trials with Missing Outcome Data
National Research Council (2010)
The Prevention and Treatment of Missing Data in Clinical Trials
Lakens, D. (2022)
Sample Size Justification
Hopewell, S., et al. (2025)
CONSORT 2025 Statement: Updated Guideline for Reporting Randomised Trials
11 · Cite this Guide
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.
Recommended (Field Guide)
APA
MLA
Chicago
Copy Citation
Has the Field Guide helped your research?
If a guide helped clarify a question, inform a research decision, or move your work forward, I would love to hear about your experience. Your story may also help other researchers discover the Field Guide.
Share Your Experience
Takes only a few minutes
