Definition · Plain-language
Quasi-experimental design
A quasi-experimental design tests a cause-and-effect relationship by manipulating or comparing conditions, but without random assignment of participants to groups.
The step most authors miss
Doing CRediT right? Don’t stop at the statement.
A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.
Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.
An experiment without random assignment
A quasi-experiment shares the goal of a true experiment — establishing whether an independent variable affects an outcome — and often involves manipulation, but it does not randomly assign participants to conditions. Instead it works with groups that already exist or that form by some non-random process: different schools, hospitals, regions, or those who happened to receive a treatment versus those who did not. This makes the design practical in real-world and field settings where randomisation is impossible, unethical or politically unworkable, such as evaluating an education reform or a public-health intervention already rolled out to part of a population.
Common quasi-experimental designs
Several recognised structures fall under the quasi-experimental umbrella. The non-equivalent groups design compares a treatment group with a similar but non-randomised comparison group, usually with before-and-after measures. The interrupted time-series design takes many measurements before and after an intervention to see whether the trend shifts at the point of change. The regression-discontinuity design assigns treatment by a cut-off score on a continuous variable and compares units just above and below it. Natural experiments exploit an externally imposed change — a law, a disaster, a lottery — that allocates exposure in a quasi-random way.
Threats to internal validity
Because participants are not randomly assigned, the groups may differ systematically from the start, so the central worry is selection bias — the chance that a pre-existing difference, not the treatment, explains the result. Other threats include maturation (natural change over time), history (outside events coinciding with the intervention), and regression to the mean. Researchers strengthen quasi-experiments by adding comparison groups, multiple pre-tests, and statistical controls such as matching or propensity scoring. These help, but cannot fully substitute for randomisation, so causal claims from quasi-experiments are held more cautiously than those from true experiments.
Key facts
At a glance
- Definition: tests cause and effect without random assignment to conditions
- Key absence: no randomisation; groups are pre-existing or non-randomly formed
- Used when: randomisation is impractical, unethical or impossible
- Common types: non-equivalent groups, interrupted time-series, regression-discontinuity
- Main threat: selection bias and other threats to internal validity
- Strengthened by: comparison groups, multiple pre-tests, matching, statistical control
Common misconceptions
What people often get wrong
Often heard: A quasi-experiment is just a poorly run true experiment.
Actually: It is a distinct design chosen deliberately when randomisation is impossible or unethical. It trades some internal validity for real-world feasibility, not through carelessness but by necessity.
Often heard: Quasi-experiments cannot say anything about cause and effect.
Actually: They can support causal inference, especially strong designs like regression-discontinuity and interrupted time-series, but conclusions are held more cautiously because confounding from non-equivalent groups is harder to rule out.
Often heard: The only difference from a true experiment is the lack of a control group.
Actually: The defining difference is the lack of random assignment, not the lack of a control group. Quasi-experiments often have comparison groups; what they lack is randomisation to those groups.
Going deeper
