Avoiding Pseudoreplication: A Guide to True Replication in Experiments

Pseudoreplication occurs when researchers treat non-independent samples as if they were independent replicates, leading to inflated confidence in results and conclusions that may not hold up under scrutiny. This guide explains what pseudoreplication is, why it undermines experimental validity, and how to design studies with genuine replication. The content is written for students, researchers, and life-science professionals who design experiments, analyze data, or interpret published findings. The practical outcome is a checklist you can apply to identify and avoid pseudoreplication in your own experimental design, with examples of correct and incorrect replication drawn from ecology and biology.

At a Glance

The table above summarizes the core distinction. The remainder of this article explains each element in detail, provides practical steps for designing experiments with proper replication, and describes common failure patterns to avoid.

What Pseudoreplication Means in Practice

Pseudoreplication is the use of statistical analysis that assumes independence when the data points are not actually independent. The term was introduced to the scientific community in the mid-1980s and has since become a central concern in ecological and behavioral research. The core problem is that when samples share a common source, such as the same animal, the same cage, or the same field plot, they are more similar to each other than they are to samples from other sources. Treating them as independent inflates the effective sample size and makes statistical tests appear more powerful than they truly are.

A review of host-microbiota studies published in 2024 examined 115 manuscripts and found that 22 percent contained pseudoreplication, primarily due to co-housed organisms, while 52 percent lacked sufficient methodological details to determine whether pseudoreplication had occurred. Only 26 percent adequately addressed the issue through proper experimental design or statistical analysis. This finding underscores how common the problem remains even in modern research.

The consequences are not trivial. When pseudoreplication goes undetected, researchers may report statistically significant effects that are actually driven by the shared environment or shared genetics of the samples instead of by the treatment itself. This contributes to poor reproducibility across the life sciences and can lead to wasted resources and misguided follow-up studies.

The Distinction Between Experimental Units and Subsamples

The experimental unit is the smallest unit to which a treatment is independently assigned. It is the level at which replication occurs. Subsamples are multiple measurements taken from the same experimental unit. Confusing these two levels is the most common source of pseudoreplication.

Consider a study measuring cellular physiology in animals. If a researcher takes blood from five animals per treatment group and measures ten cells from each blood sample, the experimental unit is the animal, not the cell. The ten cells are subsamples that describe the same animal. Analyzing all fifty cells as if they were independent observations inflates the sample size fivefold and can produce erroneous positive results. A review in the Journal of General Physiology used simulations to demonstrate this exact problem, showing that simple t tests or ANOVA applied to cells from each animal can give misleading results because the cells are not independent of each other. The recommended solution is a hierarchical, nested statistical approach that accounts for the structure of the data.

The same logic applies across biological scales. In a field experiment, individual plants within a plot are subsamples of that plot. In an animal study, individual animals within a cage are subsamples of that cage. In a genomics study, loci on the same chromosome do not assort independently, creating a form of pseudoreplication that can reduce the precision of genetic analyses and widen confidence intervals.

Why Independence Matters for Statistical Validity

Standard statistical tests such as t tests and ANOVA rest on the assumption that observations are independent. When this assumption is violated, the tests produce confidence intervals that are too narrow and p values that are too small. The result is an inflated type I error rate, meaning researchers conclude an effect exists when it does not.

A study in Pest Management Science demonstrated this problem using data from agricultural experiments. The researchers compared results from suboptimal approaches, such as two-sample t tests and ordinary ANOVA assuming independent observations, with results from appropriate linear mixed models. The different approaches produced both quantitatively and qualitatively different conclusions. Simulations confirmed that the suboptimal approaches led to differences in confidence interval coverage and type I error rates. The authors recommended that statistical analysis should always reflect the experimental design and that author guidelines should require researchers to indicate how their analysis accounts for the design structure.

The practical implication is that pseudoreplication does beyond make results slightly less accurate. It can change the direction of conclusions, leading researchers to recommend interventions that have no real effect or to dismiss treatments that actually work.

Common Sources of Pseudoreplication in Biological Research

Co-housing and Shared Environments

When animals are housed together in the same cage, pen, or tank, they share environmental conditions, microbiota, and social interactions. These shared factors make individuals within the same housing unit more similar to each other than to individuals in other units. If the housing unit is the experimental unit but individual animals are treated as independent replicates, pseudoreplication occurs.

The host-microbiota review identified co-housing as the primary cause of pseudoreplication in the studies it examined. This is particularly relevant for animal research because co-housing is often necessary for welfare reasons, as social animals may suffer when housed alone. The solution is not to isolate animals but to treat the cage or pen as the experimental unit and assign multiple cages to each treatment.

Repeated Measurements From the Same Subject

Longitudinal studies that measure the same animal or plot multiple times over a study period create temporal autocorrelation. Each measurement is influenced by the previous measurements and by the shared characteristics of the subject. Analyzing all time points as independent observations inflates the sample size and can produce misleading results.

This issue is common in animal behavior research. A review of neophobia experimental design identified pseudoreplication as one of seven recurring problems, alongside lack of habituation, problems with stimulus selection, non-standardized motivation, insufficient controls, fixed treatment order, and arbitrary thresholds for data analysis. The authors noted that many animal behavior studies suffer from one or more of these issues and recommended more consistent design practices to facilitate comparisons across populations and species.

Spatial Autocorrelation in Field Studies

Field studies in ecology face a related challenge from spatial autocorrelation. Samples collected close together in space tend to be more similar than samples collected far apart because they share soil conditions, microclimate, and neighboring organisms. If a researcher samples multiple points within the same field plot and treats them as independent, the analysis ignores this spatial structure.

Fire ecology researchers have specifically addressed this issue, noting that spatial autocorrelation violates the traditional statistical assumption of observational independence. The same concern applies to resource selection functions in animal tracking studies, where consecutive locations from the same animal are highly autocorrelated. Advances in tracking technology have increased the level of autocorrelation in such datasets, and researchers have developed weighting methods to mitigate the negative consequences.

Genetic Relatedness

Samples from related individuals are not independent because they share genetic material. This is particularly relevant in genomics and quantitative genetics, where loci on the same chromosome do not assort independently. A study in Entropy explored the use of entropy statistics to quantify pseudoreplication in genomic data, noting that pseudoreplication can substantially reduce the precision of genetic analyses and make confidence intervals wider.

In animal breeding and livestock research, this issue arises when multiple offspring from the same sire or dam are treated as independent observations. The offspring share genetic and maternal environmental effects that make them more similar to each other than to unrelated animals.

Designing Experiments With True Replication

Step 1: Identify the Experimental Unit Before You Start

The experimental unit must be identified at the design stage, not after data collection. Ask yourself: what is the smallest unit to which I can independently assign a treatment? The answer depends on your research question and the logistics of your study.

In a pen trial with pigs, if treatments are assigned to pens, the pen is the experimental unit. Individual pigs within a pen are subsamples. In a cell culture study, if treatments are applied to different flasks, the flask is the experimental unit. Individual cells within a flask are subsamples. In a field trial, if treatments are assigned to plots, the plot is the experimental unit. Individual plants within a plot are subsamples.

Step 2: Replicate at the Level of the Experimental Unit

True replication means having multiple independent experimental units per treatment. The number of units needed depends on the variability between units and the size of the effect you want to detect. More replication increases statistical power but also increases cost and animal use.

In animal research, the ethical imperative is to use neither more nor fewer animals than necessary. A review in BMJ Open Science emphasized that experiments should be robust, use appropriate numbers of animals, and contribute meaningfully to scientific knowledge. The review discussed how individual experiments can be designed in several different ways and used the NC3Rs Experimental Design Assistant to visualize design differences and consider strengths and weaknesses.

Step 3: Randomize Treatment Assignment

Randomization ensures that treatments are assigned to experimental units without systematic bias. Fisher's three essential design principles, randomization, replication, and blocking, are presented as nonoptional practices for controlling bias, managing variation, and ensuring valid statistical inferences. Probability-based random allocation using validated computer-generated randomization plans is recommended over informal methods.

Randomization does not guarantee that groups will be perfectly balanced on all characteristics, but it ensures that any imbalances are due to chance instead of to conscious or unconscious bias. This is essential for the validity of statistical inference.

Step 4: Use Blocking to Control Nuisance Variation

Blocking groups experimental units that are similar in some important way and assigns treatments within each block. This reduces the impact of nuisance variation and increases the precision of treatment comparisons. For example, in an animal study, you might block by litter, by body weight, or by the day on which the experiment is conducted.

A review in eNeuro emphasized the role of blocking in reducing nuisance variation and noted that blocking is one of the three essential design principles established by Fisher. The review also highlighted that many preclinical investigators are unfamiliar with formal design of experiments methods, which contributes to poor reproducibility and unnecessary animal use.

Step 5: Plan the Statistical Analysis to Match the Design

The statistical analysis must reflect the experimental design. If the experimental unit is the pen, the analysis must account for the fact that pigs within a pen are correlated. This typically requires a mixed-effects model with pen as a random effect, or an analysis that aggregates subsamples to the level of the experimental unit.

A review in the Journal of Medical Entomology explained that modern statistical techniques such as linear mixed-effects models can separate the different scales of variability in a dataset, such as variability between individual insects and variability between the containers where they were observed. The authors noted that the perception of pseudoreplication as a problem without solution is outdated, as these models explicitly consider the different scales of variability.

Step 6: Document Your Design Decisions

Transparent reporting of experimental design is essential for readers to assess whether pseudoreplication has occurred. The host-microbiota review found that 52 percent of studies lacked sufficient methodological details to determine whether pseudoreplication had occurred. This is a reporting failure as much as a design failure.

Your methods section should state the number of experimental units per treatment, how treatments were assigned, how subsamples were collected, and how the statistical analysis accounted for the design structure. The EQUATOR Network provides reporting guidelines for health research, and the NC3Rs Experimental Design Assistant provides a graphical tool for planning and documenting experimental designs.

Statistical Approaches for Handling Non-Independent Data

Mixed-Effects Models

Mixed-effects models include both fixed effects, which are the treatment effects of interest, and random effects, which account for the grouping structure of the data. For example, a model might include treatment as a fixed effect and pen as a random effect. The random effect captures the correlation between pigs within the same pen.

The Journal of Medical Entomology review described how linear mixed-effects models can separate the different sources of variability in a dataset, such as variability between individual mosquitoes and variability between the pans where they were observed. This allows correct inferences from data gathered in studies with constraints on randomization.

Nested or Hierarchical Models

Nested models are a specific type of mixed model where subsamples are nested within experimental units. For example, cells are nested within animals, and animals are nested within treatment groups. The model estimates variance at each level, allowing the analysis to account for the fact that cells from the same animal are more similar than cells from different animals.

The Journal of General Physiology review recommended a hierarchical, nested statistics approach for experiments comparing cellular physiology between groups of animals or people. The simulations in that review showed that this approach avoids the erroneous positive results that occur when cells from each animal are treated as independent.

Aggregation to the Experimental Unit Level

A simpler approach is to aggregate subsamples to the level of the experimental unit before analysis. For example, you might calculate the mean response for each animal and then analyze the animal-level means. This approach avoids the complexity of mixed models but discards information about within-unit variability.

Aggregation is appropriate when the research question is about differences between experimental units instead of about variation within units. It is a conservative approach that cannot inflate the sample size.

Considerations for Few Random Effect Levels

A common guideline in ecology is that random effects should have at least five levels. However, a simulation study in bioRxiv challenged this view, showing that having too few random effect terms does not influence the parameter estimates or uncertainty around those estimates for fixed effects terms. The study suggested that it should be acceptable to use fewer levels of random effects if the researcher is not interested in making inference about the random effects themselves.

This finding is relevant for researchers working with small numbers of experimental units, which is common in animal studies due to cost and welfare constraints. The key is to be transparent about the limitations of the design and to interpret results with appropriate caution.

Common Failure Patterns and How to Recognize Them

The Single Cage or Single Pen Fallacy

A researcher assigns one cage to the treatment group and one cage to the control group, then measures multiple animals within each cage. The analysis treats each animal as an independent observation. This design has no true replication because there is only one experimental unit per treatment. Any difference between the cages could be due to the treatment or to any other difference between the two cages.

This pattern is surprisingly common in pilot studies and preliminary experiments. The solution is to use multiple cages per treatment, even if each cage contains fewer animals.

The Repeated Measures Oversight

A researcher measures the same animals at multiple time points and analyzes each time point as if it were an independent observation. This inflates the sample size by the number of time points and ignores the correlation between measurements from the same animal.

The solution is to use a repeated measures analysis or a mixed model with animal as a random effect. Alternatively, the researcher can analyze a summary measure for each animal, such as the area under the curve or the change from baseline.

The Subsample Confusion

A researcher takes multiple samples from the same experimental unit and treats each sample as an independent replicate. This is the cell culture version of the problem, where multiple wells from the same plate are treated as independent when the plate is the experimental unit.

The solution is to treat the plate as the experimental unit and either aggregate the well-level data or use a mixed model with plate as a random effect.

The Spatial Autocorrelation Blind Spot

A researcher samples multiple points within the same field plot and treats each point as an independent observation. The analysis ignores the spatial correlation between nearby points.

The solution is to treat the plot as the experimental unit or to use spatial statistical methods that account for autocorrelation. Fire ecology researchers have specifically addressed this issue, noting that spatial autocorrelation violates the assumption of observational independence and that understanding available methods can assist researchers in reducing its effects.

The Genomic Dependency Problem

A researcher analyzes thousands of genetic markers from a small number of individuals and treats each marker as an independent observation. Loci on the same chromosome do not assort independently, creating pseudoreplication that reduces the precision of genetic analyses.

The Entropy study explored entropy-based metrics as a tool for estimating the statistical information of complex genetic datasets. The authors noted computational limitations as the number of loci grows, making their approach limited to smaller datasets.

Practical Assessment Steps for Your Own Experiments

Step 1: Draw a Diagram of Your Experimental Design

Create a visual representation of your design showing the levels of nesting and the assignment of treatments. The NC3Rs Experimental Design Assistant provides a graphical tool for this purpose. The tool allows you to visualize the design and identify potential sources of pseudoreplication before you collect data.

Step 2: Answer the Experimental Unit Question

For each treatment in your study, ask: how many independent experimental units receive this treatment? If the answer is one, you have no replication. If the answer is two, you have minimal replication that will limit your ability to detect effects.

Step 3: Trace the Path From Treatment to Measurement

Follow the path from treatment assignment to each individual measurement. Identify every level of grouping between the treatment and the measurement. Each level of grouping is a potential source of non-independence that must be accounted for in the analysis.

Step 4: Check Whether Your Analysis Matches Your Design

Review your planned statistical analysis and ask whether it accounts for the grouping structure you identified in Step 3. If you plan to use a simple t test or ANOVA on individual measurements, you are likely assuming independence that does not exist.

Step 5: Consult Available Tools and Guidelines

The NC3Rs Experimental Design Assistant provides guidance on experimental design and helps identify potential problems. The EQUATOR Network provides reporting guidelines for health research. The Research Data Framework from the National Institute of Standards and Technology provides guidance on data management practices that support reproducibility.

Step 6: Seek Peer Review of Your Design

Before collecting data, ask a colleague or statistician to review your experimental design. A fresh perspective can identify pseudoreplication issues that you have overlooked. This is particularly important for early-career researchers who may not have encountered these issues before.

Records and Measurements to Document

Experimental Unit Identification

Document what constitutes an experimental unit in your study and justify this choice. This should be stated explicitly in your methods section.

Treatment Assignment Records

Record how treatments were assigned to experimental units, including the randomization method used. If blocking was used, document the blocking factors and the rationale for their selection.

Subsample Documentation

Document how many subsamples were collected from each experimental unit and how they were handled. This includes the timing of sample collection, the methods used, and any quality control measures.

Housing and Environmental Records

For animal studies, document housing conditions including group size, pen or cage dimensions, environmental enrichment, and any changes during the study period. This information is essential for readers to assess whether co-housing could have created non-independence.

Statistical Analysis Documentation

Document the statistical models used, including how random effects were specified and how the analysis accounted for the design structure. The Pest Management Science study recommended that author guidelines should explicitly require authors to indicate how the statistical analysis reflects the experimental design.

Limitations and Professional Escalation Criteria

When Pseudoreplication Cannot Be Fully Avoided

In some field studies, true replication is logistically impossible. A researcher studying a rare species may have access to only one population. A researcher studying a large-scale landscape phenomenon may not be able to replicate at the landscape level. In these cases, the study should be described as a case study or a preliminary investigation, and the limitations should be stated explicitly.

The Journal of Medical Entomology review noted that lack of independence in samples from ecological studies of insects reflects the complexity of working with living organisms, including the finite input of individuals, their relatedness, and the need to group organisms into functional experimental units. Modern statistical techniques can separate the different scales of variability, but the perception of pseudoreplication as a problem without solution remains.

When to Escalate to a Statistician

You should consult a statistician when you are uncertain about the appropriate experimental unit, when your design involves multiple levels of nesting, when you are planning to use mixed-effects models, or when reviewers have raised concerns about your statistical approach. A statistician can help you design the study properly before data collection and analyze the data correctly after collection.

When to Reject a Manuscript or Proposal

As a reviewer or editor, you should consider whether pseudoreplication undermines the conclusions of a study. If the experimental unit is misidentified and the analysis treats subsamples as independent, the results may not support the claims made. The severity of the problem depends on the degree of non-independence and the extent to which conclusions depend on the inflated sample size.

The debate about pseudoreplication includes perspectives that question whether it is always a fatal flaw. A 2009 article in the Journal of Comparative Psychology argued that the core ideas behind pseudoreplication are based on a misunderstanding of statistical independence and the nature of control groups, and that there are no universal criteria for accepting or rejecting experimental research. A companion article responded that pseudoreplication is still a problem. This debate highlights the importance of judging each study on its own merits while maintaining rigorous standards for experimental design.

Welfare and Safety Context

Ethical Animal Use

In animal research, the ethical imperative is to design experiments that use the minimum number of animals necessary to answer the research question. Pseudoreplication undermines this imperative because it creates the illusion of adequate sample size when the true sample size is much smaller. A study with pseudoreplication may use more animals than necessary while still failing to provide reliable answers.

The BMJ Open Science review emphasized that experiments should be robust, not use more or fewer animals than necessary, and truly add to the knowledge base of science. The review also noted the need to conduct a harm-benefit analysis to ensure that animal use is justified for the scientific gain.

The Standardization Fallacy

A related issue is the standardization fallacy, where researchers attempt to make experimental conditions as uniform as possible to reduce variability. While standardization can reduce noise, it can also limit the generalizability of results. The BMJ Open Science review discussed the standardization fallacy alongside pseudoreplication, blocking, covariates, sex bias, and inference space.

The practical implication is that researchers should balance the need for control with the need for generalizability. Overly standardized conditions may produce results that do not apply to the broader population of interest.

Reporting for Reproducibility

Transparent reporting of experimental design is essential for reproducibility. The host-microbiota review found that more than half of the studies it examined lacked sufficient methodological details to determine whether pseudoreplication had occurred. This reporting failure makes it impossible for readers to assess the validity of the conclusions.

The Research Data Framework from the National Institute of Standards and Technology provides guidance on data management practices that support reproducibility. The EQUATOR Network provides reporting guidelines for health research. The NC3Rs Experimental Design Assistant provides a graphical tool for planning and documenting experimental designs.

Frequently Asked Questions

What is the difference between a replicate and a subsample?

A replicate is an independent experimental unit that receives a treatment. A subsample is a measurement taken from an experimental unit. Multiple subsamples from the same experimental unit are not independent replicates because they share the characteristics of that unit. For example, if you have five pigs in a pen and the pen is the experimental unit, each pig is a subsample. If you have five pens per treatment, each pen is a replicate.

How many replicates do I need for a valid experiment?

The number of replicates depends on the variability between experimental units and the size of the effect you want to detect. More replicates increase statistical power but also increase cost and animal use. A statistician can help you determine the appropriate sample size based on your specific research question and expected variability. The NC3Rs Experimental Design Assistant can help you plan the design and estimate sample sizes.

Can I use mixed-effects models to fix pseudoreplication after data collection?

Mixed-effects models can account for non-independence in the data, but they cannot fix a design that lacks true replication. If you have only one experimental unit per treatment, no statistical model can create replication that does not exist. Mixed-effects models are most useful when you have multiple experimental units per treatment and multiple subsamples within each unit.

What should I do if I discover pseudoreplication in my published work?

You should consult with a statistician to assess the severity of the problem and determine whether the conclusions are affected. If the conclusions change, you should consider publishing a correction. If the conclusions are robust to the pseudoreplication, you should acknowledge the limitation in future communications about the work. Transparency is essential for maintaining scientific integrity.

Is pseudoreplication always a fatal flaw in a study?

The debate in the literature includes perspectives that pseudoreplication is not always fatal and that each study must be judged on its own merits. However, the weight of evidence suggests that pseudoreplication can produce misleading results and should be avoided whenever possible. The severity of the problem depends on the degree of non-independence and the extent to which conclusions depend on the inflated sample size.

How does pseudoreplication affect meta-analyses and systematic reviews?

Pseudoreplication in primary studies can distort the results of meta-analyses because the inflated sample sizes give those studies more weight than they deserve. Reviewers should assess the risk of pseudoreplication in individual studies and consider sensitivity analyses that exclude or downweight studies with design flaws. The EQUATOR Network provides reporting guidelines that can help reviewers assess study quality.

What is the relationship between pseudoreplication and spatial autocorrelation?

Spatial autocorrelation is a specific form of non-independence where samples collected close together in space are more similar than samples collected far apart. It is a common cause of pseudoreplication in field studies. Fire ecology researchers have specifically addressed this issue, noting that spatial autocorrelation violates the traditional statistical assumption of observational independence. Methods for reducing the effect of spatial autocorrelation include spatial statistical models and autocorrelation-informed weighting.

How can I explain pseudoreplication to colleagues who are not statisticians?

Use a concrete example from your own field. Explain that samples from the same animal, cage, or plot are more similar to each other than to samples from other sources, and that treating them as independent makes the study appear more powerful than it actually is. Emphasize that the goal is not to make research more difficult but to make it more reliable and reproducible.

Related Articles

• Journal Of Experimental Biology
• Journal Of Experimental Biology
• Journal Of Experimental Biology
• What Is a Buffer in Biology? Definition and Examples
• What Is a Stimulus in Biology? Definition and Examples

References and Further Reading

• Research Data Framework. National Institute of Standards and Technology.
• EQUATOR Network. EQUATOR Network.
• Experimental Design Assistant. NC3Rs.
• NCBI Literature Resources. National Center for Biotechnology Information.
• PubMed. National Library of Medicine.
• A host of issues: pseudoreplication in host-microbiota studies.. Applied and environmental microbiology, 2024.
• The "Seven Deadly Sins" of Neophobia Experimental Design.. Integrative and comparative biology, 2024.
• What is the optimum design for my animal experiment?. BMJ open science, 2021.
• Pseudoreplication in physiology: More means less.. The Journal of general physiology, 2021.
• Experimental design matters for statistical analysis: how to handle blocking.. Pest management science, 2018.
• Pseudoreplication is a pseudoproblem.. Journal of comparative psychology (Washington, D.C. : 1983), 2009.
• An entomologist guide to demystify pseudoreplication: data analysis of field studies with design constraints.. Journal of medical entomology, 2010.
• Experimental Designs for Preclinical Neuroscience Experiments: Part I-Design Basics.. eNeuro, 2026.
• Paradoxical G-quadruplex distribution in coronavirus genomes reveals functional constraints and antiviral therapeutic opportunities.. 2026.
• Brokering peace in the ape (culture) wars.. 2026.
• How thoughtful experimental design can empower biologists in the omics era.. 2025.
• Towards a systematic framework to assess restoration success of interventions in coral reef ecosystems.. 2026.
• Critical evaluation of drug response prediction models with DrEval.. 2026.
• Sub-millimeter quantification of alveolar bone loss using automated 40 MHz high-frequency ultrasound: A proof-of-concept ex vivo validation study.. 2026.
• Impact of phage therapy in post-weaning piglets challenged with ETEC strain in a controlled minitrial.. 2026.
• Potential Benefits and Challenges of Quantifying Pseudoreplication in Genomic Data with Entropy Statistics. Entropy, 2024.
• A Review of Overlapping Landscapes: Pseudoreplication or a Red Herring in Landscape Ecology?. Current Landscape Ecology Reports, 2020.
• Including random effects in statistical models in ecology: fewer than five levels?. bioRxiv, 2021.
• Spatial autocorrelation and pseudoreplication in fire ecology. 2006.
• Mitigating pseudoreplication and bias in resource selection functions with autocorrelation-informed weighting. bioRxiv, 2022.
• Practices and Applications in Fire Ecology SPATIAL AUTOCORRELATION AND PSEUDOREPLICATION IN FIRE ECOLOGY. 2007.
• Using Biological Insight and Pragmatism When Thinking about Pseudoreplication. Trends in Ecology and Evolution, 2018.
• Pseudoreplication Is (Still) a Problem. Journal of Comparative Psychology, 2009.

This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.