Quasi-Experimental Design: Definition, Types, and Examples

What Is a Quasi-Experimental Design?

A quasi-experimental design tests whether a treatment or intervention causes a change in an outcome — but without random assignment. The prefix “quasi” means “almost”: a quasi-experiment looks like a true experiment and asks the same questions, but it is missing the one ingredient that defines a true experiment — random allocation of participants to groups.

Why would a researcher accept that limitation? Because in real life, random assignment is often impossible or unethical — you cannot randomly assign some students to a failing school and others to a good one. Whenever the researcher cannot control who gets the treatment, but still wants to compare groups and draw conclusions about cause and effect, a quasi-experimental design is the honest, practical choice.

In plain English: a quasi-experiment is what you do when you want to evaluate an intervention in the real world, where the world refuses to be randomized.

Quasi-Experimental vs True Experimental Design

The difference comes down to one question: who decides which group a participant belongs to?

FeatureTrue experimentQuasi-experiment
Random assignmentYes — participants are randomly allocated to treatment and control groupsNo — groups are pre-existing or formed by circumstances
Researcher controls the treatmentYesUsually yes, but sometimes the treatment happens naturally (e.g., a policy change)
Internal validityHigh — randomization balances out other differences between groupsModerate — differences between groups may explain the results instead of the treatment
Where it is usedLaboratories, controlled trialsSchools, hospitals, workplaces, communities, policy evaluation
Ethical and practical flexibilityLower — randomization is not always possibleHigher — works with real-world groups as they are

When Should You Use a Quasi-Experimental Design?

They are the right choice in four common situations:

  • Random assignment is unethical. You cannot randomly assign children to receive or not receive a reading programme if withholding it would harm them.
  • Random assignment is impossible. Groups already exist — two schools, two hospital wards — and you cannot reshuffle their members.
  • A policy or event has already happened. A new law or a curriculum change has affected one group but not another, and the researcher arrives after the fact.
  • You are evaluating a programme in the field. Real programmes run in real settings, and the evaluation has to fit around them.

If random assignment is possible, a true experiment is almost always the stronger design. Quasi-experiments are the design you choose when a true experiment cannot be done — and you then work harder to rule out alternative explanations.

Types of Quasi-Experimental Design

The four designs below cover most of what you will meet in student work.

1. Nonequivalent Groups Design

This is the most common quasi-experimental design. Two pre-existing groups are compared: one receives the treatment and the other does not. The groups are called “nonequivalent” because, without random assignment, you cannot assume they were similar to begin with.

Example: A researcher wants to test whether a new study-skills workshop improves exam scores. Two intact classes are available. Class A attends the workshop; Class B does not. At the end of the semester, exam scores are compared. The catch: Class A might have been stronger students before the workshop even started — so the researcher should ideally measure scores before the workshop too.

2. Pretest–Posttest Nonequivalent Groups Design

This adds a pretest to the design above. Both groups are measured before the treatment and again after it. Comparing the change in each group, rather than just the final scores, helps rule out the possibility that the groups were simply different from the start.

Example: Both classes take a test before the workshop and the same test afterwards. Class A improves by 15 points and Class B by 3 points. The difference in improvement is much more convincing evidence than a single end-of-term comparison.

3. Interrupted Time Series Design

Here there is only one group, but it is measured many times before and after the treatment. If the outcome was following a steady trend and then changes sharply at exactly the point the treatment began, that pattern is strong evidence the treatment caused the change.

Example: A hospital records medication errors every month for two years. A new double-check procedure is introduced in January. If errors were stable at around 20 per month and then drop to around 8 per month immediately after January and stay low, the timing supports the conclusion that the procedure worked.

4. Regression Discontinuity Design

Here, participants are assigned to treatment or control based on a cut-off score on some measure. Those just above the cut-off get the treatment; those just below it do not. Because people just above and just below the cut-off are nearly identical on average, comparing them gives a surprisingly clean estimate of the treatment effect.

Example: Students who score below 50 on a diagnostic test are offered remedial tutoring; students who score 50 or above are not. Comparing students who scored 48 or 49 with those who scored 50 or 51 is a fair comparison, because a one-point difference on a test is mostly chance.

A Worked Example: Evaluating a New Teaching Method

Imagine a university lecturer has developed a new way of teaching statistics that uses real datasets instead of textbook examples, and wants to know whether it improves understanding.

The problem: She teaches two sections of the same course. Students enrolled themselves in whichever section suited their timetable, so she cannot randomly assign them.

The design: A pretest–posttest nonequivalent groups design.

  1. Pretest: In week 1, both sections take the same 20-question statistics quiz. Section A averages 52%, Section B 54% — close, but not identical.
  2. Treatment: Section A is taught with the new real-dataset method for the rest of the semester. Section B is taught in the usual way, with the same syllabus, textbook, and exam.
  3. Posttest: In the final week, both sections take the same quiz again. Section A averages 74%, Section B 61%.
  4. Analysis: Section A improved by 22 points, Section B by 7. The difference — 15 points — is the estimated effect of the new method.

What to watch out for: The lecturer should check whether the sections differed in other ways — more students in Section A may have taken statistics before, or Section B may have met at 8 a.m. Good quasi-experimental research is largely about hunting down these alternative explanations and showing, one by one, that they do not explain the result.

Threats to Validity in Quasi-Experimental Designs

Because the groups are not equivalent at the start, quasi-experiments are more vulnerable to threats to internal validity:

  • Selection bias: The groups differed before the study began — the biggest threat, and the reason a pretest matters so much.
  • History: Something else happened at the same time as the treatment — for example, a new statistics app became popular halfway through the semester.
  • Maturation: Participants naturally change over time — students improve simply by being in a course — so a single-group design cannot separate the treatment effect from normal learning.
  • Testing effects: Taking the pretest itself can change the posttest score, because students may remember the questions.
  • Instrumentation: The measurement changed between pretest and posttest — for example, a different marker graded more generously.
  • Regression to the mean: If a group was chosen because its scores were unusually low, its scores will tend to rise on the next measurement even without treatment.
  • Attrition: If weaker students drop out of one group more than the other, the final groups differ from the ones that started.

You do not need to eliminate every threat — that is usually impossible outside a laboratory. What examiners want to see is that you recognized the threats, designed against the biggest ones, and discussed the rest honestly in your limitations section.

How to Strengthen a Quasi-Experimental Design

A quasi-experiment does not have to be a weak study. Several practical steps make it more convincing:

  • Use a pretest. Measuring both groups before the treatment lets you compare change rather than final scores.
  • Add a comparison group whenever possible. A single group measured before and after a treatment is the weakest design.
  • Measure many times. A time series with many observations before and after the treatment makes it harder for a one-off event to masquerade as a treatment effect.
  • Match groups on key variables. If you cannot randomize, choose a comparison group that resembles the treatment group in age, prior ability, and background — or adjust for differences statistically.
  • Collect data on possible confounders. Measure the things that might explain your result (prior grades, attendance, motivation) so you can check whether they differ between groups.
  • Use more than one outcome measure. If the treatment works, the effect should show up on several related measures.

Common Mistakes to Avoid

  • Calling it an experiment. Without random assignment, it is a quasi-experiment. Calling it a true experiment in your thesis is an error examiners notice immediately.
  • Ignoring baseline differences. Comparing only posttest scores when the groups started at different levels is the most common analysis mistake in student quasi-experiments.
  • Using a single-group design when a comparison group exists. If a comparison group is available, use it — a single-group design cannot rule out history or maturation.
  • Overclaiming causality. A quasi-experiment supports a causal argument; it does not prove one the way a randomized experiment does. Write “the results suggest the programme improved scores” rather than “the programme caused the improvement.”
  • Forgetting to discuss limitations. Naming the specific threats to your design and explaining what you did about them makes your study stronger, not weaker. Hiding the weaknesses damages credibility.

Frequently Asked Questions

Is a quasi-experimental design qualitative or quantitative?

It is quantitative. Quasi-experimental designs test the effect of a treatment on a measured outcome, usually with numerical data and statistical analysis. Interviews can sit alongside one, but the design itself is quantitative.

What is the difference between a quasi-experiment and a correlational study?

A quasi-experiment involves a treatment the researcher introduces (or that occurs naturally) and a comparison between groups or over time. A correlational study simply measures two variables as they naturally occur and looks for a relationship, without any intervention. Quasi-experiments are closer to experiments; correlational studies manipulate nothing.

Can quasi-experimental designs establish cause and effect?

They can provide good evidence for cause and effect, but weaker evidence than a randomized experiment. A regression discontinuity design or a well-run interrupted time series comes close to experimental evidence, while a single-group pretest–posttest design provides only weak evidence. The honest answer for your thesis is “the design supports a causal interpretation, with the following limitations.”

What is an example of a quasi-experimental design in education?

Comparing two intact classes where one receives a new teaching method and the other does not is the classic example. Another is evaluating a new curriculum introduced in one school but not a neighbouring school.

Why is there no control group in some quasi-experiments?

Sometimes a control group genuinely does not exist — for example, when a new national policy affects everyone at once. In those cases, the interrupted time series design replaces the control group with many measurements over time: the group’s own past performance becomes the comparison. Whenever a comparison group does exist, though, you should use one.

Key Takeaways

  • A quasi-experimental design tests the effect of a treatment without random assignment — the groups are pre-existing or formed by circumstances.
  • It is the right design when randomization is unethical, impossible, or has already been overtaken by events.
  • The main types are the nonequivalent groups design, the pretest–posttest nonequivalent groups design, the interrupted time series design, and the regression discontinuity design.
  • Its main weakness is selection bias: the groups may have differed before the study began. Pretests, comparison groups, and repeated measurements are the standard defences.
  • Write about it honestly: name the threats to validity, show what you did about them, and avoid overclaiming causality.

Leave a Comment