What Should You Do When the Calculated Sample Size Is Impossible to Recruit?

What If Your Power Analysis Requires 600 Participants but You Can Recruit Only 150?

You are planning an experiment to determine whether AI-generated feedback improves university students' academic writing performance. After specifying the primary outcome, meaningful effect, significance level, and desired statistical power, your calculation indicates that approximately 600 usable observations are required.

However, your institution has only 180 eligible students in the relevant courses. Even with strong participation, obtaining 600 observations from this population is impossible.

A colleague suggests recruiting as many students as possible and proceeding with the original hypotheses. Another recommends increasing the assumed effect size until the calculated requirement falls below 180. Someone else proposes reporting that the sample was determined by availability.

Which approach is methodologically defensible?

An unattainable sample size does not necessarily mean the research idea has no value. It does mean that the original design, assumptions, resources, and inferential objectives need to be examined together. The important distinction is between conducting a smaller study that can answer a meaningful question and conducting a study that cannot adequately support the conclusion it promises to establish.

Reassess the Calculation, Improve the Design, or Revise the Study's Inferential Objective

How to Respond When Recruitment Capacity Falls Below the Required Sample

Does an Unattainable Sample Size Mean the Study Cannot Be Conducted?

Not necessarily. A sample size requirement is calculated for a particular statistical objective under specified assumptions. It does not determine whether every possible investigation of the topic is worthwhile.

For example, a definitive experiment intended to detect a small educationally meaningful effect may require several hundred participants. A preliminary investigation of intervention feasibility may require fewer participants because it addresses a different objective.

However, the distinction matters. A smaller feasibility study cannot automatically support the same effectiveness claim as the original confirmatory experiment.

The researcher should determine whether the planned sample is insufficient for the original question or whether the original calculation was based on inappropriate assumptions.

What Should You Check Before Accepting the Calculated Sample Size?

Start by reviewing the statistical model, primary outcome, effect size, significance level, desired power, and variability assumptions.

For example, a calculation for two independent groups may be inappropriate if the actual study uses repeated measurements and plans to adjust for baseline performance.

Likewise, a calculation may be inflated because the standard deviation was entered as a variance, the allocation ratio was specified incorrectly, or clustering was counted twice.

These are technical problems that should be corrected.

However, researchers should distinguish correcting an error from changing a scientifically justified assumption merely to produce a smaller number.

The objective is to obtain an accurate calculation, not the most convenient calculation.

How Do You Distinguish Statistical Requirements from Recruitment Capacity?

The required sample describes the information needed to achieve a specified statistical objective under the planning model.

Recruitment capacity describes the number of participants the researcher can realistically enroll and retain.

These quantities should be evaluated separately before making a design decision.

A sample of 150 participants does not become statistically adequate simply because 150 is the maximum available.

Conversely, an unavailable target of 600 should not automatically prevent a researcher from pursuing a more limited but coherent objective.

What If the Entire Eligible Population Is Smaller Than the Calculated Sample?

Sometimes the target population is genuinely finite and small.

For example, a university may have only 180 students enrolled in the specialized program under investigation.

If the objective is to describe that particular population, inviting every eligible student may be reasonable. For design-based estimation of a finite population, a finite population correction may also be relevant when probability sampling occurs without replacement.

However, studying every available student does not automatically provide adequate power for a broader treatment-effect claim.

The distinction between a finite-population descriptive estimand and an effect intended to generalize across future students or institutions should be made explicit.

Researchers should not apply a finite population correction indiscriminately to solve an underpowered experiment.

Can You Improve Statistical Efficiency Without Increasing Recruitment?

Potentially. Some design and analytical improvements can increase precision without changing the substantive research question.

For example, a randomized experiment measuring writing performance may benefit from adjusting the follow-up outcome for a reliable baseline writing score.

Appropriate blocking, reliable outcome measurement, and efficient allocation may also improve precision under suitable conditions.

However, the expected efficiency gain should be supported by credible assumptions.

Researchers should not claim that baseline adjustment or repeated measurement will halve the sample requirement without a calculation reflecting the planned analysis.

Can Repeated Measures Solve an Infeasible Sample Size?

Repeated measurements may improve precision for certain estimands, particularly when within-person correlations provide useful information about change.

For example, assessing writing performance before and after an intervention may improve efficiency compared with an analysis that ignores baseline performance.

However, additional measurements from the same students do not create additional independent participants.

The gain depends on the estimand, measurement schedule, covariance structure, and statistical model.

Repeated assessments can also increase burden and missing-data risk.

See how repeated measures affect sample size requirements.

Can Better Measurement Reduce the Required Sample?

Sometimes. Measurement error can increase uncertainty or weaken observed associations.

For example, inconsistently scored essays may introduce avoidable variability into an educational intervention study.

Using a well-developed rubric, suitable rater training, and defensible scoring procedures may improve measurement quality.

However, improved reliability does not guarantee a particular reduction in sample size.

Researchers should estimate the expected statistical benefit using relevant evidence.

Measurement improvements are desirable because they strengthen the study, not simply because they may reduce recruitment demands.

Should You Consider Recruiting from Additional Institutions?

Multi-institutional recruitment may expand the eligible population and improve the feasibility of achieving the required sample.

For example, researchers evaluating AI-generated feedback might collaborate with several universities offering comparable writing courses.

However, recruitment across institutions introduces additional considerations.

Differences in curricula, student characteristics, assessment procedures, and implementation conditions may affect outcomes.

Institutional clustering and treatment-effect heterogeneity may also need to be considered in sample size planning.

Expanding recruitment is useful when the additional settings remain appropriate for the target population and research question.

Can You Extend the Recruitment Period?

Extending recruitment may be appropriate when additional eligible participants will become available over time.

For example, a study may recruit across several academic terms rather than one semester.

However, changes in curriculum, AI platform capabilities, institutional policy, or assessment procedures across terms may introduce temporal differences.

The researcher should consider whether the intervention and outcome remain sufficiently comparable.

Extending recruitment may improve feasibility, but it does not automatically preserve the original design assumptions.

Can You Use Unequal Allocation to Reduce Recruitment Requirements?

Not necessarily. For a conventional two-group mean comparison with equal variances and similar observation costs, equal allocation generally minimizes the total sample needed for a specified precision.

Unequal allocation may be justified by different intervention costs, group variances, or recruitment conditions.

However, it commonly increases the total sample requirement under equal-variance assumptions.

Researchers should therefore not assume that assigning more participants to the easier-to-recruit group solves an inadequate sample.

See how unequal allocation affects sample size.

Should You Lower the Desired Statistical Power?

Reducing the target power lowers the calculated sample requirement under otherwise fixed assumptions.

However, it also increases the probability of failing to reject the null hypothesis when the specified alternative is true.

For example, reducing the target from 80% to 60% means accepting a substantially greater probability of missing the effect used in planning.

Such a tradeoff may be considered in some exploratory contexts, but it should not be presented as equivalent to the original confirmatory objective.

Researchers should report the chosen power and its rationale transparently.

Lowering power solely because recruitment is difficult does not make the scientific consequences disappear.

Should You Increase Alpha to Obtain a Smaller Sample?

A less stringent significance level can increase power for a fixed sample or reduce the required sample for a fixed power target.

However, it also changes the nominal Type I error behavior of the test.

For example, increasing alpha from 0.05 to 0.10 is not merely an administrative adjustment. It changes the threshold for rejecting the null hypothesis.

Any change should follow the study's inferential objectives and consequences of erroneous conclusions.

Researchers should not increase alpha after examining the data or merely to describe an otherwise inadequate sample as acceptable.

See how significance level affects sample size.

Can You Assume a Larger Effect Size?

A larger assumed effect generally reduces the sample required for conventional hypothesis testing.

However, the planning effect should reflect credible evidence or a substantively justified effect of interest.

Suppose a study originally aims to detect a standardized effect of 0.25 but the researcher changes the assumption to 0.50 because fewer participants are available.

Under a conventional independent-means approximation, the required sample may decrease to approximately one quarter.

Yet the revised design is now powered for a different effect magnitude. It may have limited power for the smaller effect that originally mattered.

See why expected effect size affects sample size.

What Is a Sensitivity Power Analysis?

A sensitivity power analysis examines the statistical power achievable with a fixed sample across plausible effect sizes or other planning assumptions.

For example, a researcher limited to 150 participants may evaluate the probability of detecting standardized effects of 0.20, 0.35, and 0.50.

This helps clarify what the feasible design can and cannot reliably detect.

However, a sensitivity calculation does not establish that the detectable effect is likely to occur or substantively meaningful.

The results should inform a transparent design decision rather than be used to reverse-engineer a convenient effect assumption.

Can You Report the Minimum Detectable Effect?

Yes. A minimum detectable effect, often abbreviated MDE, describes the effect magnitude for which the planned design achieves a specified power under stated assumptions.

For example, a study with a fixed recruitment capacity may have 80% power to detect a standardized mean difference of approximately 0.50.

However, the MDE is not necessarily the smallest effect that matters scientifically.

If the smallest meaningful effect is 0.20, a design capable of reliably detecting only 0.50 may not adequately address the original question.

Researchers should distinguish the design's detection capability from substantive importance.

Can an Underpowered Study Still Produce Useful Evidence?

Yes, but its value depends on the objectives, design quality, and interpretation.

A smaller study may provide effect estimates, confidence intervals, information about outcome variability, or evidence about recruitment and implementation feasibility.

However, low power may produce imprecise estimates and a high probability of nonsignificance even when meaningful effects exist.

Significant estimates from small studies may also be exaggerated under some conditions, particularly when results are selectively reported.

Researchers should therefore avoid treating either significance or nonsignificance as a definitive conclusion unsupported by the design.

See what statistical power means for study design.

When Is a Pilot or Feasibility Study More Appropriate?

A pilot or feasibility study may be preferable when the available sample cannot support the original confirmatory objective but can answer useful preliminary questions.

For example, researchers may investigate whether students can complete the intervention, whether writing assessments are reliable, and whether recruitment and retention procedures are workable.

These objectives can inform a subsequent definitive study.

However, a pilot study should not be represented as a smaller version of a definitive effectiveness trial with equivalent inferential claims.

Its sample size should be justified according to the feasibility parameters or preliminary estimates it is intended to obtain.

What If the Study Is Primarily Descriptive?

Descriptive studies may be better planned around estimation precision than statistical power.

For example, a university survey may aim to estimate the proportion of students using generative AI weekly.

If the required sample for a narrow confidence interval is unattainable, the researcher may consider whether a wider interval would still provide useful information.

However, the precision target should be chosen according to the intended interpretation and decisions.

Reducing sample size necessarily changes the uncertainty of the estimate under the same sampling assumptions.

The revised precision should be reported rather than concealed.

Can You Narrow the Target Population?

Sometimes, but narrowing the population changes the research question.

For example, a study originally intended to represent university students nationally might be restricted to students in one institution.

This may improve recruitment access and implementation consistency.

However, the resulting findings cannot automatically be generalized to the original national population.

For hypothesis-testing designs, a narrower population does not necessarily reduce the sample size required to detect the same standardized effect.

Researchers should distinguish greater access to a defined population from improved statistical efficiency.

Does Recruiting the Entire Available Population Solve the Problem?

Recruiting all eligible members of a finite population may eliminate sampling variability associated with selecting only some members for a descriptive census of that population, assuming complete measurement.

However, measurement error, nonresponse, confounding, and uncertainty about broader populations may remain.

For an intervention study, enrolling every eligible student does not guarantee adequate power to detect the specified effect.

The interpretation depends on whether the research targets a finite-population description or a broader scientific effect.

A census is not a universal substitute for a statistically appropriate design.

Should You Continue Recruiting Until the Results Become Significant?

No. Unplanned repeated significance testing while adding participants can undermine the nominal Type I error control of a conventional fixed-sample test.

For example, repeatedly checking whether p has fallen below 0.05 and stopping recruitment when it does can produce misleading inference.

Sequential or adaptive designs may permit interim analyses under prespecified statistical procedures.

However, they require suitable planning and error-control methods.

Researchers should not treat informal significance-driven recruitment as an acceptable solution to an inadequate sample.

What If the Study Is Already Underway?

If recruitment difficulties emerge after data collection begins, researchers should document the problem and reassess feasibility without making outcome-driven decisions.

Potential responses include approved recruitment extensions, additional sites, or transparent changes to the study objective.

Any protocol amendment should distinguish decisions made before outcome inspection from those influenced by emerging results.

Researchers should also evaluate whether the planned analysis remains appropriate for the achieved sample.

Changes in recruitment should not be concealed or retrospectively described as original design decisions.

How Should an Infeasible Sample Size Be Reported?

The methodology should explain the original sample size calculation, including the effect or precision target, significance level, power, and relevant design assumptions.

If recruitment limitations required a revision, describe the revised objective and the reasons for the change.

Where the study proceeds with a smaller sample, report the achieved sample and its implications for estimation uncertainty and detection capability.

Avoid relying on observed post hoc power calculated from the estimated effect as an explanation of completed results.

Effect estimates, confidence intervals, and transparent limitations generally provide more useful information.

What Can a Researcher Do with 150 Participants When the Original Design Requires Approximately 500?

Using the conventional normal approximation for independent group means:

The researcher evaluates what different effect assumptions would imply under the same design.

The values use a normal approximation and exclude attrition adjustments. Exact noncentral t calculations may produce slightly different requirements.

The university's capacity of 150 students is sufficient for the approximate requirement associated with d = 0.50 but not for d = 0.25.

However, the researcher cannot conclude that the true effect is 0.50 merely because this assumption produces a feasible sample.

With 75 participants per group, the approximate standardized effect detectable with 80% power is:

The feasible study therefore has limited power for the originally meaningful effect of 0.25 under these assumptions.

The researcher might seek additional institutions, investigate a defensible baseline-adjusted analysis, or revise the study as a preliminary investigation emphasizing effect estimation and feasibility.

If the original small-effect confirmatory objective remains essential, the available sample is inadequate for that purpose.

The appropriate decision is to align the study's claims with its achievable information rather than revise the effect assumption solely to justify recruitment.

Common Mistakes When the Required Sample Cannot Be Recruited

How to Make a Defensible Decision When Recruitment Is Limited

Begin by determining whether the recruitment shortfall reflects an incorrect calculation, a design that can be improved, or a genuine mismatch between the research objective and available resources.

The most important decision is whether the feasible sample can answer a question worth asking. If the original question requires information that cannot be obtained, changing the question may be more scientifically responsible than pretending the requirement has been met.

Before Proceeding with an Unattainable Sample Size, Verify the Alternatives

Frequently Asked Questions About Infeasible Sample Sizes

An Impossible Sample Size Calls for a Better Design Decision, Not a Convenient Calculation

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.