A within-groups design is an experimental setup where every participant experiences all of the conditions being tested. Instead of splitting people into separate groups and giving each group a different treatment, researchers run the same people through each treatment and compare their responses across conditions. This makes each person their own control, which is one of the design’s biggest strengths.
You’ll also see it called a “within-subjects design,” a “repeated measures design,” or a “dependent groups design.” These all refer to the same core idea: the comparison happens within the same individuals rather than between different groups of people.
How It Works
Imagine a researcher wants to know whether background music helps or hurts test performance. In a within-groups design, every participant would take a test with music playing and take a comparable test in silence. The researcher then compares each person’s scores across both conditions. Because the same person generated both scores, any differences in personality, intelligence, motivation, or mood apply equally to both conditions. Those individual quirks don’t muddy the comparison.
This stands in contrast to a between-groups design, where one group of people would take the test with music and a completely separate group would take it in silence. In that setup, you’d need to hope that both groups happened to contain a similar mix of fast learners, anxious test-takers, and people running on four hours of sleep. A within-groups design sidesteps that problem entirely.
Why Researchers Use It
The biggest advantage is statistical power, meaning the ability to detect a real effect when one exists. Because individual differences are held constant across conditions, there’s far less random noise in the data. Research on this topic has found that within-subjects designs can require roughly half the sample size of between-subjects designs to detect effects of the same magnitude. That translates directly into lower costs, less recruitment effort, and faster studies.
This matters especially in fields where participants are hard to find. If you’re studying a rare medical condition or recruiting from a small, specific population, needing 30 participants instead of 60 can be the difference between a feasible study and an impossible one.
Order Effects: The Main Drawback
The central vulnerability of a within-groups design is that the order in which participants experience conditions can shape their responses. These are called order effects, and they come in several forms.
- Practice effects: Participants get better at a task simply because they’ve done it before, not because the second condition is actually superior.
- Fatigue effects: Participants perform worse in later conditions because they’re tired or bored, not because the later treatment is less effective.
- Carryover effects: Exposure to one condition changes how a participant responds to the next. A drug tested first might still be active in the body when the second drug is introduced, for instance.
- Context effects: Experiencing one condition changes how participants perceive or interpret the next. Tasting a very sweet drink first might make a moderately sweet drink seem bland by comparison.
There’s also the risk that participants figure out what the study is testing. When people go through multiple conditions, it’s easier for them to guess the hypothesis and consciously or unconsciously adjust their behavior.
How Researchers Control for Order Effects
The standard solution is counterbalancing: varying the order of conditions across participants so that any order effects spread evenly rather than consistently favoring one condition. In the simplest version with two conditions (A and B), half the participants do A first and B second, while the other half do B first and A second.
When a study has three or more conditions, full counterbalancing (testing every possible order) becomes impractical because the number of possible sequences grows rapidly. Researchers often use a Latin square design instead, a grid-based approach where each condition appears in each position (first, second, third) exactly once. This was originally proposed by the statistician Ronald Fisher as a way to control for extraneous variables without needing every possible ordering.
Counterbalancing doesn’t eliminate order effects. It distributes them so they don’t systematically bias one condition over another. One important caveat: counterbalancing only works if the order effects don’t interact with the treatment effects themselves. If condition A genuinely changes how people respond to condition B in a way that doesn’t happen in reverse, no amount of reordering fully solves the problem.
Crossover Trials in Medicine
In clinical research, the within-groups approach appears as the crossover trial. Patients are randomly assigned to sequences of treatments: one group receives treatment A followed by treatment B, while the other receives B followed by A. Each patient’s response to one treatment is compared directly against their own response to the other.
Crossover trials work best for chronic, stable conditions where the treatment relieves symptoms without permanently altering the disease. They aren’t suitable when a treatment could cure the condition or cause irreversible changes, because there would be nothing left to measure in the second phase.
To handle carryover effects, researchers insert a washout period between treatments, a stretch of time with no active treatment that lets the first treatment’s effects fade. In one well-known crossover study comparing the effects of butter versus margarine on cholesterol, patients spent six weeks on each diet with a five-week washout period in between, during which they returned to their normal eating habits. The assumption is that five weeks is enough for the dietary effects to dissipate before starting the next phase. Getting the washout period right is critical: too short, and the first treatment contaminates the second; too long, and the study becomes impractical.
Participant Dropout
Because every participant must complete all conditions, losing even one person means losing data from every condition, not just one. In a between-groups design, a dropout affects only the group they were assigned to. In a within-groups design, that same dropout removes an entire set of matched data points, which can disproportionately reduce statistical power.
This is especially relevant in studies that stretch over weeks or months, like crossover trials with washout periods. The longer participants need to stay enrolled, the more likely some will miss sessions, move away, or simply lose interest. Researchers planning within-groups studies typically recruit extra participants to account for anticipated attrition, though the exact number depends on how many conditions are involved and how long the study runs.
How the Data Gets Analyzed
The statistical tools for within-groups designs are specifically built to account for the fact that the same people appear in every condition. For two conditions, the paired samples t-test compares each person’s score in condition A against their own score in condition B. For three or more conditions, the repeated measures ANOVA extends this logic, accounting for correlations within each person’s data across time points.
Repeated measures ANOVA carries an extra statistical assumption called sphericity, which essentially means the variability in differences between conditions should be roughly consistent across all pairs of conditions. When this assumption is violated, results can be misleading, so researchers test for it and apply corrections when needed. For data that doesn’t meet the standard assumptions at all, nonparametric alternatives like Friedman’s test can be used instead.
When to Choose a Within-Groups Design
A within-groups design is the stronger choice when individual differences between people are likely to be large relative to the effect you’re trying to detect. It’s also preferable when participants are scarce or expensive to recruit. Studies measuring perceptual judgments, reaction times, or preferences are natural fits because the same person can easily respond to multiple stimuli in a single session.
It’s a poor choice when exposure to one condition would permanently change how someone responds to the next, when the treatments take so long that dropout becomes a serious risk, or when participants would easily guess the study’s purpose by seeing all conditions. In those situations, a between-groups design, despite requiring more participants, produces cleaner results.