What Is Pseudoreplication, and Why Does It Matter?

Are You Replicating the Experiment or Merely Measuring the Same Experimental Units More Often?

Suppose a researcher evaluates a new instructional approach using two classrooms. One classroom receives the intervention, while the other follows conventional instruction. Each classroom contains 40 students, producing 80 achievement scores.

The researcher conducts an independent-samples t-test using the 80 scores and reports a statistically significant intervention effect. Yet the intervention was applied to only two classrooms, not independently assigned to 80 students.

A similar problem arises when researchers measure ten plants from one treated plot, collect several tissue samples from one animal, or analyze multiple lesions from patients who received the same systemic treatment.

Although these measurements provide information, they do not necessarily represent independent experimental replication. Treating them as though they do can make the evidence appear stronger than the design warrants.

This problem is called pseudoreplication. It concerns a fundamental distinction between obtaining additional measurements and conducting additional independent experimental replications.

Pseudoreplication Is False Independent Replication

Understanding Pseudoreplication, Independent Replication, and Statistical Inference

Where Does the Concept of Pseudoreplication Come From?

Pseudoreplication became especially prominent through Stuart H. Hurlbert's 1984 paper, Pseudoreplication and the Design of Ecological Field Experiments, published in Ecological Monographs.

Hurlbert examined experimental designs in which researchers treated subsamples, repeated measurements, or observations within insufficiently replicated treatment units as independent evidence of treatment effects.

Although the original discussion focused on ecological field experiments, the underlying problem extends to education, medicine, psychology, laboratory research, agriculture, and organizational studies.

The central concern remains the same: the statistical analysis must reflect the level at which independent experimental replication actually occurs.

Pseudoreplication is therefore not merely a technical problem involving a particular statistical test. It can arise from a mismatch between the scientific question, experimental design, and interpretation of the collected observations.

What Is an Experimental Unit?

The experimental unit is the smallest entity independently assigned to a treatment condition under the experimental design.

Depending on the study, an experimental unit might be a student, classroom, school, patient, laboratory animal, culture dish, agricultural plot, or another independently assigned entity.

For example, if classrooms are randomly assigned to different teaching strategies, the classroom is the experimental unit. Students within each classroom provide outcome measurements, but they do not constitute independently randomized classrooms.

If individual students are independently randomized within classrooms, the student may instead be the experimental unit, although classroom-level dependence or treatment interference may still require attention.

The experimental unit is determined by how treatment is assigned, not by the number of rows in the dataset.

This distinction is closely related to the difference between randomization and measurement units.

What Is the Difference Between Replication and Repeated Measurement?

Replication involves obtaining independent experimental units or independently implemented experimental comparisons relevant to the treatment effect.

Repeated measurement involves recording multiple observations from an existing unit.

For example, assigning a treatment independently to ten laboratory animals provides a different experimental structure from assigning it to one animal and measuring that animal ten times.

The repeated measurements may reveal temporal changes or improve measurement precision. However, they do not create ten independently treated animals.

Similarly, assessing students repeatedly within one classroom does not create additional independently assigned classrooms.

Researchers should distinguish repeated observations from independent measurements and determine what each contributes to the intended estimand.

What Are Common Forms of Pseudoreplication?

Hurlbert's original discussion distinguished several forms of pseudoreplication. The following categories illustrate how they can arise in contemporary research.

These categories are useful for diagnosis, although the exact terminology and classification can vary between methodological disciplines.

The practical requirement is to identify which observations represent independent treatment replication and which represent measurements within those units.

How Does Simple Pseudoreplication Occur?

Simple pseudoreplication occurs when a treatment comparison lacks adequate independent experimental replication but multiple observations within treatment units are analyzed as though they provide it.

Suppose one hospital implements a new patient safety protocol and another hospital retains its existing procedures.

Researchers compare outcomes from 500 patients in each hospital and treat the 1,000 patients as independent treatment replications.

Because treatment condition is aligned with hospital identity, the observed difference may reflect the intervention, hospital resources, patient case mix, or other institutional characteristics.

The patient observations provide information about the two hospitals, but they do not establish independently replicated hospital-level treatment effects.

Additional patient measurements cannot eliminate this fundamental limitation.

What Is Temporal Pseudoreplication?

Temporal pseudoreplication involves treating repeated observations over time as independent experimental replications when they arise from the same treatment-assigned unit.

For example, a researcher may apply one intervention to a classroom and record student engagement every week for ten weeks.

The repeated observations can reveal temporal patterns, but the ten weeks do not constitute ten independently randomized classrooms.

Furthermore, observations close together in time may be correlated because of persistent conditions and temporal processes.

Longitudinal models can account for repeated-measurement dependence, but they do not automatically resolve a lack of independent treatment replication.

Researchers should distinguish the value of repeated observation for estimating change from its limitations for establishing treatment effects.

What Is Sacrificial Pseudoreplication?

Sacrificial pseudoreplication occurs when a design contains independently replicated treatment units, but the analysis fails to preserve the relevant experimental structure.

For example, suppose ten classrooms are independently randomized, with five assigned to each treatment condition. Thirty students are assessed in every classroom.

The study has ten independent classroom-level assignments and 300 student observations.

If researchers analyze all 300 students as independently randomized without accounting for classroom assignment, they improperly use student-level variation to support a treatment comparison whose independent replication occurs at the classroom level.

Unlike a design with only one classroom per condition, this problem may be addressable through reanalysis because independent classroom replication exists.

However, the appropriate method depends on the number of clusters, the estimand, and the correlation structure.

What Is the Difference Between Biological and Technical Replication?

Biological replication involves distinct biological units relevant to the scientific question, while technical replication involves repeated measurements or processing of the same underlying biological material.

For example, independently sampled patients may constitute biological replicates. Measuring the same patient's specimen several times provides technical measurements.

Technical replication can help evaluate instrument precision, assay variability, and measurement reliability.

It does not automatically increase the number of independently sampled patients or independently assigned treatment units.

The distinction is especially important when the study contains several organs, tissues, lesions, or specimens from one participant.

Researchers should consider whether multiple anatomical observations from one participant are independent before defining their analytical sample.

Does Pseudoreplication Always Mean Observations Are Correlated?

Statistical dependence is an important mechanism underlying pseudoreplication, but the broader problem concerns independent experimental assignment and replication.

Observations from the same experimental unit may be correlated because they share treatment, environment, or unmeasured characteristics.

However, even when the estimated within-unit correlation is small, the level of treatment assignment remains unchanged.

For example, a classroom-randomized trial still contains classroom-level treatment assignments regardless of whether student outcomes exhibit a small observed ICC.

Researchers should therefore avoid defining experimental replication solely by the estimated correlation between measurements.

The assignment mechanism and the research question must also be considered.

How Does Pseudoreplication Affect Standard Errors?

When observations within experimental units are positively correlated, treating them as independent can underestimate uncertainty for certain treatment comparisons.

This can produce confidence intervals that are too narrow and p-values that are too small.

The magnitude depends on the number and size of experimental units, within-unit correlation, and the statistical model.

For a simple equal-cluster-size setting, an illustrative design effect is:

Such calculations illustrate the consequences of dependence but do not replace design-specific power calculations or appropriate treatment-effect analysis.

Does Pseudoreplication Always Change the Estimated Effect?

No. Some pseudoreplication errors primarily affect estimated uncertainty rather than the numerical difference between group means.

For example, the observed difference in mean achievement between intervention and comparison students may remain unchanged after clustering is recognized.

However, its confidence interval and significance test may change substantially.

Other corrections, such as different cluster weighting or covariate-adjusted modeling, can change the point estimate itself.

Researchers should distinguish a mathematically correct descriptive contrast from a statistically or causally defensible interpretation of that contrast.

Is Pseudoreplication the Same as an Inflated Sample Size?

The concepts overlap but are not identical.

Sample size inflation occurs when related observations are counted as more independent information than the design supports.

Pseudoreplication specifically concerns false independent replication at the relevant experimental or analytical level.

For example, reporting 300 student measurements from ten classrooms is accurate. Claiming that the study contains 300 independently randomized classrooms would be incorrect.

The issue is not the number of observations collected but how those observations are counted and used for inference.

Researchers should distinguish the nominal number of observations from the independent information available.

Is Every Clustered Study Pseudoreplicated?

No. Clustering is a property of the data structure, while pseudoreplication is an error in how replication is represented or analyzed.

Cluster-randomized trials can legitimately measure individual outcomes while assigning treatment to groups.

Similarly, longitudinal studies can legitimately collect many measurements from each participant.

These designs are not inherently pseudoreplicated when the analysis respects the assignment and dependence structures.

For example, a classroom-randomized trial may analyze student-level outcomes using an appropriate cluster-aware model.

The critical distinction is whether the analysis recognizes the independently assigned units rather than treating every measurement as independent treatment replication.

Can Pseudoreplication Occur in Observational Research?

Related problems can arise in observational studies when clustered or repeated observations are incorrectly treated as independent.

For example, a researcher may analyze multiple hospital admissions from the same patient as though they came from unrelated individuals.

Such an analysis can misrepresent uncertainty or give disproportionate influence to patients with more admissions.

Strictly speaking, pseudoreplication is most directly associated with experimental replication. In observational research, the more precise description may be unmodeled clustering, nonindependent observations, or an incorrectly identified analytical unit.

Using terminology carefully helps researchers distinguish errors in treatment replication from other forms of dependence.

Can Multilevel Models Correct Pseudoreplication?

Multilevel models can represent observations nested within experimental units and may address certain dependence structures.

For example, a model may include student outcomes and classroom-level random effects when classrooms were independently randomized.

However, a multilevel model cannot create independent experimental units that were never included.

If one classroom received the intervention and another served as the comparison, the model cannot independently distinguish treatment from classroom identity using that simple design alone.

Likewise, models with very few clusters may produce unstable or assumption-sensitive estimates.

The usefulness of multilevel modeling depends on the actual replication, estimand, and statistical assumptions.

Can Cluster-Robust Standard Errors Correct the Problem?

Cluster-robust standard errors can adjust uncertainty for certain forms of within-cluster dependence when sufficient independent clusters are available.

They may be appropriate for a correctly specified treatment-effect estimator whose observations are clustered within independently assigned groups.

However, conventional cluster-robust inference may be unreliable with few clusters.

More importantly, a variance correction cannot resolve complete confounding between treatment condition and a single group per condition.

Researchers should distinguish analytical problems that can be addressed through appropriate uncertainty estimation from fundamental deficiencies in experimental design.

How Can You Prevent Pseudoreplication Before Data Collection?

Begin by identifying the smallest entity independently assigned to each experimental condition.

Then determine how many such units are needed to support the intended comparison.

Specify whether additional observations will represent independent experimental units, subsamples, repeated measurements, or technical replicates.

Preserve identifiers connecting every observation to its experimental unit.

Sample size calculations should account for clustering, expected effect size, outcome variation, and the number of independent assignment units.

These considerations are closely related to planning sample size using participants and clusters.

Can Pseudoreplication Be Corrected After Data Collection?

Sometimes. If sufficient independent experimental units were included and their identifiers are available, reanalysis may be possible.

For example, student-level outcomes from a trial with many randomized classrooms can be analyzed using an appropriate cluster-aware method.

However, a design with only one independently treated group per condition may lack the replication necessary for conventional treatment-effect inference.

In that situation, researchers may need to narrow the conclusions, treat the comparison as exploratory, or conduct a better-replicated study.

Statistical reanalysis cannot fully repair a fundamental absence of independent experimental replication.

How Should Experimental Replication Be Reported?

Methods sections should identify the experimental unit, assignment procedure, number of independent units per condition, and number of observations within each unit.

For example, report that 24 classrooms were randomized and 720 students provided outcome measurements.

Describe whether subsamples were aggregated or analyzed using appropriate correlated-data methods.

When relevant, report ICC assumptions, group sizes, repeated measurements, missing outcomes, and small-cluster limitations.

Clear reporting allows readers to distinguish genuine replication from the total volume of recorded observations.

How a Classroom Experiment Can Mistake Student Measurements for Experimental Replication

The difference of 84 − 77 = 7 points is mathematically correct. What remains uncertain is its causal interpretation.

Suppose a naive analysis reports p = 0.01. This hypothetical p-value does not establish that the treatment effect was independently replicated.

Even if the two classrooms were randomly allocated to conditions, the study would have extremely limited assignment-level information.

A more defensible design would independently assign the instructional strategy across a sufficient number of classrooms and analyze student outcomes using methods that respect classroom-level assignment.

The additional classrooms would provide genuine experimental replication, while additional students within each classroom would improve measurement of classroom outcomes.

Common Misconceptions About Pseudoreplication

How to Recognize and Prevent Pseudoreplication

Begin by identifying what was independently assigned or manipulated. Then determine whether the observations used in the analysis represent independent experimental units or measurements within those units.

The objective is not to discard useful measurements. It is to distinguish what those measurements reveal about existing experimental units from what independently replicated units reveal about treatment effects.

What to Verify Before Claiming Independent Experimental Replication

Frequently Asked Questions About Pseudoreplication

Independent Replication Is a Design Property, Not a Row Count

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.