Can Sample Size Be Re-Estimated During a Research Study?

What If the Original Sample Size Calculation Turns Out to Be Wrong?

A researcher calculates the required sample size before recruitment, but the calculation depends on assumptions that may be uncertain. Perhaps the outcome is more variable than expected, the event rate differs from earlier estimates, or the anticipated treatment effect is less certain than the literature suggested.

Once data collection begins, can the researcher revise the sample size without invalidating the study? Or must the original enrollment target remain unchanged regardless of what the accumulating evidence reveals?

The answer depends on the information used to justify the revision, whether the decision was planned, and how the statistical analysis accounts for the resulting design.

Can Researchers Re-Estimate Sample Size During an Ongoing Study?

How Sample Size Re-Estimation Works and When It Is Defensible

What Is Sample Size Re-Estimation?

Sample size re-estimation, sometimes called sample size reassessment, is a procedure for revising the planned number of participants or observations using information accumulated after a study begins.

It is a form of adaptive design when the possibility of revision and the relevant decision process are prospectively established.

The purpose is usually to address uncertainty in assumptions underlying the original sample size calculation or, in some designs, to modify enrollment in response to interim evidence about the treatment effect.

The distinction matters because the statistical consequences of reassessing outcome variability differ from those of changing sample size after examining an observed treatment difference.

Sample size re-estimation should not be confused with simply recruiting additional participants because the original sample size was inconvenient or because the preliminary results were disappointing.

Why Might the Original Sample Size Be Inaccurate?

Sample size calculations depend on statistical assumptions that are often estimated from earlier studies, pilot data, or substantive expectations.

For a conventional comparison of two independent means, important assumptions include the target effect size, outcome variability, significance level, desired statistical power, and allocation ratio.

Other study designs may depend on event rates, intracluster correlation, repeated-measure correlations, anticipated attrition, or additional parameters.

When these assumptions are inaccurate, the actual statistical power may differ from the value originally planned.

For example, if the outcome is more variable than anticipated, the same treatment difference becomes more difficult to detect. The study may therefore require more participants to maintain the intended power for the originally specified effect.

Conversely, lower-than-expected variability may indicate that the original enrollment target is larger than necessary, although reducing sample size requires its own justification and suitable procedure.

What Is the Difference Between Blinded and Unblinded Re-Estimation?

Blinding does not automatically guarantee statistical validity. A blinded procedure still requires evaluation of its assumptions, estimation method, and final analysis.

Similarly, unblinded re-estimation is not inherently invalid. Established methods can accommodate interim treatment-effect information, provided the design and analysis appropriately account for the adaptation.

How Does Blinded Sample Size Re-Estimation Work?

Blinded re-estimation often focuses on nuisance parameters. These are quantities needed for statistical planning but are not themselves the principal treatment effect of interest.

For example, the variance of a continuous outcome influences the number of participants required to detect a specified difference between groups.

Suppose the original calculation assumes a standard deviation of 10 points. After a designated number of participants have completed outcome assessment, a blinded analysis produces an updated standard deviation estimate of 12 points.

Because greater variability generally requires more observations to detect the same target difference, the planned sample size may need to increase.

However, pooled variance calculated without treatment labels can contain contributions from genuine between-group differences. Its properties therefore depend on the design and the method used.

Researchers should not assume that every blinded estimate is an unbiased estimate of the within-group variance needed for the original calculation.

How Is Sample Size Related to Outcome Variability?

For a simplified two-arm study comparing independent means with equal allocation, a common approximate sample size formula is:

Under these assumptions, the revised variance estimate increases the approximate requirement from 126 to 182 participants overall.

The calculation illustrates the relationship between variability and sample size. It does not establish that a particular adaptive re-estimation procedure is valid or that the resulting study will achieve exactly 80% power.

The formula is a normal-approximation calculation for a simplified independent-groups setting. Real studies may require exact or alternative methods, adjustments for clustering or attrition, and evaluation of the adaptive procedure itself.

What Is Unblinded Sample Size Re-Estimation?

Unblinded re-estimation uses interim information that may reveal the observed treatment effect.

For example, an interim analysis might indicate that the treatment difference is smaller than the target effect used in the original sample size calculation.

Researchers may consider whether increasing enrollment would improve the probability of achieving the prespecified study objective.

However, an interim treatment-effect estimate is subject to sampling variability. Decisions based on that estimate can affect the distribution of the final test statistic and the interpretation of the resulting evidence.

Consequently, unblinded re-estimation generally requires methods specifically designed to accommodate data-dependent changes.

Possible approaches include conditional error methods, combination-test procedures, and other adaptive inferential frameworks.

The choice depends on the study's design, endpoint, adaptation rule, and inferential objectives.

What Is Conditional Power?

Conditional power is the probability of rejecting the null hypothesis at the planned final analysis, conditional on the interim information already observed and under a specified assumption about the future data-generating process.

It is frequently used in adaptive trial planning and interim decision-making.

Conditional power differs from the original design power because it incorporates information accumulated during the study.

It also differs from the probability that the alternative hypothesis is true. Conditional power is calculated under an assumed treatment effect or predictive model, rather than directly measuring the truth of a hypothesis.

Researchers may calculate conditional power under the originally targeted effect, the observed interim effect, or another scientifically justified assumption.

These choices can produce substantially different results.

For example, conditional power based on an unusually favorable interim effect may be overly optimistic if the observed difference is largely due to random variation.

A sample size re-estimation rule based on conditional power therefore needs clearly specified assumptions and appropriate evaluation.

What Is the Promising-Zone Approach?

Some adaptive sample size procedures use a promising-zone strategy, in which sample size increases are considered only when interim evidence falls within a prespecified range.

The general rationale is that a study showing very weak interim evidence may be unlikely to benefit sufficiently from additional enrollment, while a study already showing strong evidence may not require an increase.

An intermediate range may represent circumstances in which additional information could meaningfully improve the chance of meeting the study objective.

However, the boundaries of such a zone are not universal. They depend on the statistical procedure, study objectives, assumed effects, and constraints on enrollment.

Not every promising-zone implementation automatically controls Type I error. The full design must be validated, including the sample size decision rule and final hypothesis test.

Can Sample Size Be Reduced After Re-Estimation?

Potentially. Some prospectively planned procedures permit reductions when updated information indicates that fewer participants may be needed.

However, reducing enrollment may affect precision, safety assessment, secondary outcomes, and the ability to investigate heterogeneity.

A study may also have scientific or regulatory requirements that are not fully captured by the power calculation for its primary endpoint.

For example, a clinical trial might require sufficient exposure to characterize important adverse events even if the primary efficacy endpoint could be evaluated with fewer participants.

Any reduction should therefore be evaluated against the complete study objectives, not only the revised primary-outcome calculation.

Does Re-Estimation Automatically Preserve Type I Error?

No. The probability of falsely rejecting a true null hypothesis depends on the full statistical procedure.

Some blinded nuisance-parameter re-estimation methods have favorable Type I error properties under specified conditions. Other procedures, particularly those using unblinded treatment-effect information, may require explicit adjustments.

The relevant concern is not simply that the sample size changes. It is that the change may depend on information statistically related to the final hypothesis test.

Repeatedly increasing enrollment after inspecting interim significance levels can create additional opportunities to obtain a statistically significant result.

Researchers should therefore evaluate the false-positive consequences of adaptive decisions rather than assuming that a conventional final test remains valid after every possible modification.

When Should Re-Estimation Be Planned?

Sample size re-estimation should ordinarily be considered during study design, especially when important planning parameters are uncertain and interim data can provide useful information.

The protocol should specify the relevant timing, data requirements, estimation method, decision algorithm, permissible sample size range, and final analytical approach.

The interim assessment must occur early enough for additional recruitment to be feasible, but late enough to provide sufficiently informative estimates.

Researchers should also establish who can access interim results and how decisions will be communicated without compromising blinding or study conduct.

These arrangements are part of the broader prespecification requirements for adaptive designs.

What If Sample Size Must Change for an Unplanned Reason?

Unexpected circumstances may make the original enrollment target impractical or scientifically inappropriate.

For example, external evidence may change the assumptions underlying the study, or an unforeseen operational problem may affect recruitment.

Such circumstances do not automatically justify an unplanned adaptive analysis.

Researchers should distinguish amendments made without access to relevant comparative results from changes influenced by emerging treatment effects.

They should document the reason for the amendment, its timing, the information available to decision-makers, and its implications for the final analysis.

Depending on the circumstances, independent statistical review, ethical approval, protocol amendment, or regulatory consultation may be necessary.

If the revised design cannot support the original confirmatory interpretation, researchers should acknowledge that limitation rather than treating the amended study as equivalent to a prospectively planned adaptive trial.

Re-Estimating Sample Size in an Educational Intervention Study

This example illustrates the arithmetic and decision sequence rather than validating a particular re-estimation procedure. A real trial would need to account for its actual variance estimator, outcome distribution, allocation structure, attrition, and statistical operating characteristics.

If students were randomized by classroom or school, for instance, the independent-observation formula would generally be inadequate because intracluster correlation affects the information provided by the sample.

The scientifically relevant distinction is that the enrollment change responds to a predefined uncertainty about outcome variability, rather than to an observed treatment difference that the researchers hope to make statistically significant.

Common Mistakes in Sample Size Re-Estimation

How to Decide Whether Your Study Needs Sample Size Re-Estimation

Start by identifying which assumption in the original sample size calculation is genuinely uncertain. A re-estimation procedure is most defensible when accumulating data can improve that assumption in a way that meaningfully informs enrollment.

Then determine whether blinded information is sufficient. If the principal uncertainty concerns outcome variability or an aggregate event rate, an appropriately designed blinded procedure may be suitable.

If the proposed decision depends on interim treatment differences, the statistical and operational requirements become more demanding.

Researchers should evaluate the planned procedure using appropriate analytical methods or simulation. Important quantities include Type I error, power, expected sample size, maximum sample size, and the precision of final estimates.

For complex designs, specialized statistical expertise is often necessary. The study protocol should also describe how the interim analysis will be conducted and who may access its results.

Sample size re-estimation is most useful when it resolves meaningful planning uncertainty. It should not become a mechanism for repeatedly renegotiating the evidence required to support a preferred conclusion.

What to Verify Before Re-Estimating Sample Size

Frequently Asked Questions About Sample Size Re-Estimation

Sample Size Can Change, but the Rules for Changing It Matter

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.