What is Quasi-Experimental Design? Definition, Types, Examples, and Applications

Quasi-experimental design is a research methodology used to investigate cause-and-effect relationships between variables when random assignment of participants to groups is not feasible or ethical. It occupies the methodological middle ground between a true experiment and a purely observational study, retaining some but not all features of a controlled experiment.

The prefix “quasi” means “resembling to a certain degree.” A quasi-experiment resembles a true experiment in that the researcher manipulates or studies an independent variable and measures its effect on a dependent variable. However, it differs from a true experiment in one or both of the following ways:

Participants are not randomly assigned to groups.
There may be no formal control group (though comparison groups are common).

Because random assignment is absent, quasi-experimental research cannot fully rule out the influence of confounding variables, making causal claims somewhat less certain than in a true experiment. Nevertheless, quasi-experimental designs produce substantially stronger evidence than purely correlational or observational studies because the researcher exercises at least partial control over the conditions of the study.

Table of Contents

How Does Quasi-Experimental Design Fit in the Research Hierarchy?

Quasi-experimental designs sit at a recognized level of the research evidence hierarchy. Below is a simplified map of where they fall relative to other study designs.

Design Type	Random Assignment	Control Group	Causal Strength
True experiment (RCT)	Yes	Yes	Highest
Quasi-experimental design	No	Often yes (comparison)	Moderate to high
Cohort/longitudinal study	No	Sometimes	Moderate
Cross-sectional study	No	No	Lower
Case study	No	No	Lowest

When Should You Use Quasi-Experimental Design?

Quasi-experimental design is appropriate whenever ethical, practical, or resource constraints prevent random assignment. The following situations commonly call for this approach:

Ethical constraints:

It is unethical to randomly assign participants to a harmful condition (for example, assigning people to smoke cigarettes) or to withhold a clearly beneficial treatment from a control group (for example, denying an effective vaccine). In these cases, quasi-experimental design allows investigation of causal questions without compromising participant welfare.

Group-level interventions:

When an intervention is delivered to an entire community, school, or workplace, it is logistically impossible to randomize individuals within that group. Comparing outcomes across two similar communities, one of which received the intervention, is a common quasi-experimental solution.

Small sample sizes:

Randomized controlled trials require sufficiently large samples to distribute confounders evenly across groups. In studies with small populations (for example, a rare disease cohort), quasi-experimental designs allow causal inquiry that an underpowered RCT cannot support.

Policy and program evaluation:

Governments and organizations frequently implement policies without a randomization mechanism. Quasi-experimental methods allow evaluators to estimate the causal impact of such policies after the fact using existing administrative data.

Cost and feasibility:

True experiments are expensive and logistically demanding. Quasi-experimental designs, which often use retrospective data already collected by governments or institutions, provide a more affordable route to causal evidence.

Real-world ecological validity:

When the research question concerns how an intervention works in natural conditions (rather than an artificial lab), quasi-experimental designs in field settings provide findings with greater generalizability.

Main Types of Quasi-Experimental Design

Nonequivalent Groups Design

This is the most common form of quasi-experimental design. The researcher selects two or more existing groups that appear similar to one another. One group receives the intervention (the treatment group) while the other does not (the comparison group). Because participants were not randomly assigned, the groups may differ on background characteristics: they are nonequivalent groups.

Researchers attempt to minimize these baseline differences by:

Selecting groups that are as similar as possible on key demographic and baseline variables.
Statistically controlling for known differences in the analysis.
Using propensity score matching to pair similar individuals across groups.

Pretest and Posttest Designs

Pretest-posttest designs measure the outcome variable both before (pretest) and after (posttest) the intervention. Comparing pretest and posttest scores allows the researcher to assess change over time. These designs come in two main forms:

Design Variant	Groups	Key Strength	Key Weakness
One-group pretest-posttest	One group only; no comparison group	Simple to implement; tracks individual change	Cannot distinguish intervention effect from history, maturation, or other time-based threats
Pretest-posttest with comparison group	Treatment group and nonequivalent comparison group; both measured at baseline and follow-up	Controls for many time-based threats by comparing change scores across groups	Groups may still differ at baseline in unmeasured ways

Posttest-Only Design with a Comparison Group

In this variant, only a posttest is administered. The treatment and comparison groups are assessed after the intervention only, with no baseline measurement. This design is simpler but more vulnerable to selection bias, because it provides no way to verify that the two groups were equivalent at the outset. It is best used when pretest data are unavailable or when testing effects (the act of taking a pretest affecting subsequent scores) would distort results.

Interrupted Time Series Design

The interrupted time series (ITS) design collects outcome data at multiple time points before and after an intervention. The “interruption” is the intervention itself. By examining trends before and after the intervention, researchers can determine whether the intervention produced a change in the level or slope of the outcome beyond what would have been predicted by the pre-intervention trend.

ITS is particularly well-suited to policy evaluation because:

It does not require a comparison group, though adding one strengthens causal inference.
It controls for many time-based threats to validity by modeling the pre-existing trend.
It can use routinely collected administrative data.

A key threat to ITS validity is the history effect: other events occurring at the same time as the intervention may explain the change. Including a comparable region that did not implement the policy as a control time series helps address this.

Regression Discontinuity Design

Regression discontinuity design (RDD) is used when treatment assignment is determined by whether a participant’s score on a continuous variable falls above or below a defined threshold or cutoff. Those just above the cutoff receive treatment; those just below do not. Because participants near the threshold are likely to be very similar to one another, the difference in outcomes between the two groups near the cutoff provides a strong estimate of the causal effect of treatment.

RDD is considered one of the most credible quasi-experimental designs when implemented well, approaching the internal validity of a randomized trial for individuals close to the cutoff. However, findings apply only to the population near the threshold and may not generalize to those far from it.

Difference-in-Differences Design

Difference-in-differences (DiD) is a widely used quasi-experimental method that compares changes in outcomes over time between a group that received an intervention and a group that did not. It effectively “differences out” confounding factors that are constant over time and those that affect both groups equally, isolating the treatment effect.

Instrumental Variables Design

The instrumental variables (IV) method addresses the problem of unmeasured confounders: variables that affect both participation in the intervention and the outcome. An instrument is a third variable that:

Is correlated with the treatment variable (relevance condition).
Affects the outcome only through its effect on treatment, not directly (exclusion restriction).
Is not correlated with unmeasured confounders (independence condition).

IV methods effectively use the instrument to create variation in treatment assignment that is “as good as random,” thereby isolating the causal effect of the treatment. The technique is mathematically demanding and its validity depends heavily on finding a credible instrument.

Natural Experiments

In a natural experiment, an external event, a policy change, a natural disaster, a lottery, or another circumstance outside the researcher’s control produces an assignment to conditions that approximates randomization. Researchers exploit this event after the fact to study the effect of exposure.

Although some natural experiments produce genuinely random or near-random assignment, they are not considered true experiments because the researcher did not design or control the assignment mechanism: they are observational in nature. Natural experiments occupy a privileged position among quasi-experimental designs because the external mechanism of assignment can credibly eliminate selection bias.

Propensity Score Matching

Propensity score matching (PSM) is a statistical technique used to reduce selection bias in nonequivalent groups designs. Each participant’s propensity score is the estimated probability that they would receive the treatment, given their observed background characteristics. By matching treated and untreated participants with similar propensity scores, researchers create comparison groups that are balanced on observed covariates, mimicking what randomization would have achieved.

PSM improves the credibility of causal claims by reducing systematic differences between groups. However, it can only balance groups on measured variables; unobserved confounders remain a threat.

How Do You Choose the Right Quasi-Experimental Design?

Research Situation	Recommended Design
Treatment is assigned based on a numeric threshold or cutoff score	Regression discontinuity design
Data are available from multiple time points before and after an intervention for one group	Interrupted time series (one group)
Data are available from multiple time points for both a treated and an untreated group	Interrupted time series with comparison group, or difference-in-differences
Two or more pre-existing groups are available; baseline data can be collected	Pretest-posttest with nonequivalent comparison group
Two or more pre-existing groups are available; no baseline data	Posttest-only with comparison group
An external event or natural lottery produced assignment that approximates randomization	Natural experiment
Treatment and comparison groups differ markedly on observed background variables	Propensity score matching followed by outcome comparison
An unmeasured confounder is likely to bias results; a valid instrument is available	Instrumental variables

Quasi-Experimental vs True Experimental Design

Understanding the distinctions between quasi-experimental and true experimental design helps researchers make informed methodological choices and interpret results appropriately.

Characteristic	True Experiment	Quasi-Experimental Design
Random assignment	Yes: participants randomly assigned to groups	No: assignment based on pre-existing conditions, cutoffs, or self-selection
Control over treatment	Researcher designs and controls who receives treatment	Researcher often studies pre-existing treatment conditions
Control group	Formal control group required	Comparison group common but not always required
Causal inference strength	Strongest: randomization eliminates most confounders	Moderate: confounders may persist; design features compensate
Internal validity	High	Moderate (lower than RCT; higher than observational)
External validity	Often lower: lab or artificial settings reduce generalizability	Often higher: studies real-world conditions
Ethical constraints	May be unethical to randomize to harmful conditions	Avoids ethical problems by using naturally occurring groups
Cost and feasibility	Expensive and resource-intensive	Often lower cost; can use existing administrative data
Setting	Typically laboratory or highly controlled field	Real-world field settings

Main Threats to Internal Validity in Quasi-Experimental Research

Because random assignment is absent, quasi-experimental studies are more vulnerable to threats to internal validity. Researchers must explicitly address these threats in their study design and analysis.

Threat	Description	Mitigation Strategy
Selection bias	Treatment and comparison groups differ at baseline in ways that affect the outcome, making it unclear whether observed differences reflect the treatment or pre-existing group differences.	Propensity score matching; statistical covariate control; choosing groups with similar baseline characteristics.
History effects	External events occurring during the study period affect the outcome independently of the intervention.	Use of comparison groups that experience the same external events; ITS with multiple pre-intervention data points.
Maturation	Participants naturally change over the course of a study (e.g., children mature, patients recover naturally), making it appear the treatment caused change.	Including a comparison group that matures at the same rate; measuring change over a short period.
Regression to the mean	Participants selected because of extreme scores tend to score closer to the average on subsequent measurement, regardless of treatment.	Avoid selecting participants based on extreme scores; use comparison groups; examine baseline distributions.
Testing effects	Repeated measurement itself influences participants’ responses (e.g., familiarity with a test improves scores).	Use parallel test forms; extend the interval between measurements; use comparison groups that also undergo repeated testing.
Attrition (mortality)	Participants who drop out of the study differ systematically from those who remain, biasing the sample.	Report attrition rates and reasons; perform sensitivity analyses; use intention-to-treat analysis.
Instrumentation	Changes in the measurement tool or data collection procedures over time introduce apparent change unrelated to the intervention.	Standardize measurement procedures; use the same instrument throughout.

Advantages and Disadvantages of Quasi-Experimental Design

Advantages

External validity:

Studies are conducted in real-world settings, making findings more applicable to actual populations and practices than laboratory-based true experiments.

Ethical feasibility:

Allows investigation of causal questions that would be unethical to study through randomization (e.g., exposing participants to harmful substances or withholding life-saving treatments).

Practical and cost-effective:

Often uses existing administrative or routinely collected data, reducing cost and recruitment burden substantially.

Applicability to policy evaluation:

Ideal for evaluating the impact of policies, programs, and interventions implemented at scale, where randomization is neither designed nor possible.

Stronger than observational studies:

Retains some experimental control, providing stronger causal evidence than purely correlational or cross-sectional designs.

Flexibility:

A wide range of design types accommodates many different research questions, data structures, and field contexts.

Disadvantages

Lower internal validity:

Without randomization, it is harder to rule out that differences between groups reflect pre-existing differences rather than treatment effects.

Confounding:

Unmeasured confounders cannot be controlled, and their influence on results can never be fully excluded.

Strong assumptions required:

Methods such as DiD and IV rely on assumptions (parallel trends, instrument validity) that are difficult or impossible to fully verify and must be defended on theoretical grounds.

Limited generalizability of RDD:

Regression discontinuity findings apply only to individuals near the threshold and may not generalize to the broader population.

Data quality dependency:

When using retrospective administrative data, missing values, measurement error, or incomplete records can compromise findings.

Potential for researcher bias:

Without the objectivity conferred by randomization, researchers must be particularly transparent and rigorous in specifying hypotheses and analysis plans in advance.

Reporting Standards and Quality Frameworks for Quasi-Experimental Studies

Several established frameworks guide the reporting and quality assessment of quasi-experimental research. Familiarity with these frameworks is valuable for both authors and readers of quasi-experimental studies.

Framework or Tool	Purpose	Field
TREND (Transparent Reporting of Evaluations with Nonrandomized Designs)	Reporting checklist for quasi-experimental evaluations of behavioral and public health interventions	Public health, epidemiology
CASP Quasi-Experimental Checklist	Critical appraisal tool for assessing methodological quality of quasi-experimental studies	Healthcare, nursing, social science
Cochrane Effective Practice and Organisation of Care (EPOC) criteria	Quality standards for interrupted time series and other quasi-experimental designs in healthcare	Healthcare, health systems
ROBINS-I (Risk of Bias in Non-randomized Studies of Interventions)	Tool for assessing risk of bias in non-randomized studies	Healthcare, systematic reviews
Maryland Scientific Methods Scale (SMS)	Five-level scale ranking study designs by methodological rigor; quasi-experimental designs occupy levels 3 to 5	Criminology, social policy

What Are the Most Common Mistakes in Quasi-Experimental Research?

Ignoring baseline differences: Comparing only posttest scores without measuring or controlling for baseline differences between groups is the single most common source of invalid causal claims in quasi-experimental research.

Overstating causal certainty: Authors sometimes present quasi-experimental findings as definitive causal proof without adequately acknowledging the threats to validity that remain. Using language such as “demonstrates that” rather than “is consistent with the hypothesis that” is misleading.

Failing to test design assumptions: DiD studies that do not examine pre-treatment trends, or RDD studies that do not test for manipulation of the running variable, leave critical assumptions unverified.

Selecting a comparison group for convenience rather than comparability: A comparison group that is demographically or contextually different from the treatment group introduces selection bias that statistical adjustment may not fully correct.

Using a single pre-intervention time point in ITS: An ITS design with only one or two pre-intervention data points cannot reliably estimate the pre-existing trend. More data points before the intervention strengthens inference substantially.

Ignoring attrition: Reporting only results for participants who completed the study without addressing dropout rates or the characteristics of those who dropped out can bias findings significantly.

Not pre-registering the analysis plan: Without pre-registration, there is a risk of reporting only the analyses that yield significant results, inflating false-positive rates.