Experimental Unit
An experimental unit is the smallest physical entity that receives a treatment independently and is measured or observed. This guide explains how to identify, apply, and analyze experimental units in life sciences research. Use this guide if you are designing a study, writing a grant, analyzing data in bioinformatics, or reviewing a manuscript. A clear grasp of the experimental unit prevents pseudoreplication and ensures your statistical inferences are valid. The NCBI Bookshelf provides authoritative background on experimental design principles.
In a typical biomedical study, the experimental unit may be a single mouse, a cell culture dish, a patient, or a field plot. If you apply a treatment to three mice but then measure ten cells from each mouse, the experimental unit is the mouse, not the cell. Using cells as units inflates your sample size and risks false findings. The EMBL-EBI Training offers practical courses on experimental design that clarify this core concept.
At a Glance
| Concept | Definition | Example |
|---|---|---|
| Experimental unit | Smallest independently treated entity | A single rat, a Petri dish, an arrayed well |
| Replicate | Multiple independent experimental units receiving the same treatment | Four rats per drug dose |
| Pseudoreplication | Treating non independent measurements as replicates | Measuring three biopsies from one rat and calling them n equals 3 |
| Observational unit | The entity on which a measurement is taken | A cell, a tissue section, a sequencing read |
| Independence | No influence between treatments applied to different units | One cage per treatment group, no cross contamination |
Core Concepts
The experimental unit is the basis for calculating sample size and performing hypothesis tests. Replication must occur at the level of the experimental unit, not at the level of the measurement. The Galaxy Training Network includes tutorials that describe how to set up biological replicates versus technical replicates in sequencing experiments.
A replicate is a unit that receives the same treatment as other units but is handled independently. For example, if you treat five separate mice with the same compound, you have five replicates at the mouse level. If you take five blood samples from one mouse, those are technical replicates and do not count as independent biological replicates. Confusing these two types of replication is the most frequent error in experimental design.
Independence means that the outcome from one experimental unit does not depend on the outcome from another. In animal studies, this requirement often means housing animals in separate cages or using a randomized block design. The NCBI Sequence Read Archive stores metadata that documents whether samples come from independent individuals or are technical replicates.
Decision Criteria
You need to decide what the experimental unit is before you start collecting data. Consider the following criteria.
First, identify the level at which you physically apply the treatment. If you inject a drug into each mouse individually, the mouse is the unit. If you add drug to a culture flask that contains many cells, the flask is the unit.
Second, determine whether the units can be randomly assigned. Randomization must happen at the level of the experimental unit. For example, in a plant study you randomize plots, not individual plants within a plot. The Bioconductor documentation for the limma package includes guidance on defining experimental units in microarray and RNA sequencing studies.
Third, check for clustering or repeated measures. If you measure the same animal multiple times over weeks, the animal remains the experimental unit but the repeated measures require special modeling. If you use technical replicates to improve precision, they do not increase the sample size for your primary hypothesis.
Practical Workflow
Follow this sequence to establish and work with experimental units in any life sciences project.
Define your research question and primary outcome. Write down the exact comparison you want to test. For example, does drug X reduce tumor volume compared to vehicle?
Identify the experimental unit. Ask: What object receives the treatment independently? For a drug study in mice, it is each mouse. For a cell culture experiment, it is each well or dish.
Determine the number of replicates. Use a power analysis or published guidelines. As a rule of thumb, fewer than three replicates per treatment makes it nearly impossible to detect meaningful effects. Review methods from EMBL-EBI Training for sample size estimation.
Randomize treatments to experimental units. Use a random number generator to assign each unit to a group. Document the randomization in your lab notebook.
Apply treatments and collect measurements. Record which units received which treatments. Store raw data with unique identifiers for each experimental unit.
Analyze data using the correct error term. Use statistical models that treat the experimental unit as the random factor. In software like R or Galaxy, specify the unit explicitly. The Galaxy Training Network offers step by step workflows for differential expression analysis that incorporate biological replicates.
Check assumptions and validate. Plot the data to see if variability is similar across groups. If you used technical replicates, average them before analysis.
Common Mistakes
Mistaking technical replicates for biological replicates
This error inflates the apparent sample size. For example, a study on cartilage degeneration used three rats per group but measured multiple joint sections per rat. If the authors treated each section as an independent unit, they committed pseudoreplication. The PubMed article on intra articular VEGF inhibition in rats correctly treats the rat as the experimental unit.
Ignoring the hierarchy in nested designs
In a study of bariatric techniques in pediatric obesity, patients are the experimental units. But if each patient provides multiple measurements over time, those repeated measures are not independent. The systematic review emphasizes that clustering by patient must be accounted for.
Using the wrong error term in analysis
If you treat observations as independent when they are not, your p values will be too small. For example, in a retinal pigment epithelial cell study, cells from the same donor are not fully independent. The paper on soft matrix and ciliogenesis explains how donor identity is the experimental unit.
Equating sample size with number of measurements
A common mistake is to count every sequencing read or every cell as a replicate. For single cell RNA sequencing, the experimental unit is often the biological sample (the individual or the animal), not the thousands of individual cells. The study on spinal muscular atrophy motoneurons uses multiple patient derived cell lines, each line serving as a unit.
Limits and Interpretation
Even with careful definition, the experimental unit can be ambiguous. In some in vitro experiments using cell lines, you might treat a single flask and then measure triplicates from that flask. In that case you have only one experimental unit and three technical replicates. You cannot generalize beyond that single flask.
In high throughput sequencing, the experimental unit may be the library preparation batch, not the individual sample, if samples are processed in batches. The NCBI Sequence Read Archive metadata often includes batch identifiers, which you must consider.
Some designs involve multiple levels of experimental units. For example, a split plot design in an agricultural trial has whole plots and subplots. Each level has its own experimental unit and error term. The Bioconductor package nlme can handle such hierarchical models.
The limits of interpretation are especially important when using small numbers of units. With only two replicates per group, you cannot reliably estimate variance, and any difference may be driven by an outlier. Always be transparent about the number of independent units.
The study on D5 and D6 blastocysts illustrates how the embryo can be the experimental unit, but if multiple embryos come from the same donor, donor becomes a blocking factor. Similarly, the outcomes study for triple class resistant HIV treats each patient as a unit, but if some patients contribute multiple time points, those require careful modeling.
In summary, the experimental unit defines the limits of your conclusions. You can only claim that your treatment affects the type of entity you independently assigned. Do not extend the findings to larger populations unless your units were sampled from that population.
Frequently Asked Questions
What is the difference between an experimental unit and an observational unit?
The experimental unit receives the treatment independently. The observational unit is the entity from which you take a measurement. For example, in a rat study, the rat is the experimental unit, and a single tissue section is the observational unit. You should average or model the observational units to the level of the experimental unit.
Can a single cell be an experimental unit?
Yes, if you treat each cell independently. For example, in a microfluidic device where you flow drugs into individual chambers, each cell in its own chamber can be a unit. In standard cell culture, the well or dish is usually the unit, not the thousands of cells inside it.
How many replicates do I need for a valid experiment?
There is no fixed number, but three replicates per treatment is the absolute minimum for any statistical analysis. More replicates increase power and reliability. Use a power analysis based on expected effect size and variability. The EMBL-EBI Training offers tools for sample size calculation.
What is pseudoreplication and why is it bad?
Pseudoreplication occurs when you treat non independent measurements as if they are independent replicates. It artificially inflates the sample size, making results appear significant when they are not. It is a leading cause of irreproducible research.
References and Further Reading
- NCBI Bookshelf: Experimental Design , Fundamental concepts for clinical and basic research.
- EMBL-EBI Training: Design of Experiments , Courses on experimental planning in biology.
- Galaxy Training Network: Experimental Design , Workflow based tutorials for sequencing experiments.
- Bioconductor: Linear Models and Experimental Design , Software documentation for analyzing designed experiments.
- NCBI Sequence Read Archive: Metadata , How to interpret sample and replicate information in public data.
- Efficacy of bariatric techniques in pediatric and adolescent obesity , Example of patient level experimental units in a systematic review.
- Intra articular VEGF inhibition in a rat model of osteoarthritis , Example of animal level units in a histopathological study.
- Soft matrix and ciliogenesis in retinal epithelial cells , Example of donor as experimental unit in cell biology.
- Antisense oligonucleotides in spinal muscular atrophy motoneurons , Example of patient derived cell lines as units.
- Timing and molecular signatures in D5 and D6 blastocysts , Example of embryo as unit with donor blocking.