Pretest-Posttest Experimental Design
- Introduced by
- Donald T. Campbell and Julian C. Stanley
- Year
- 1963 (formalized in Campbell & Stanley)
- Type
- Experimental / quasi-experimental research design
- Family
- Experimental design
What does it do?
The pretest-posttest experimental design measures participants on the outcome variable before and after treatment, typically with random assignment to treatment and control groups.
The difference between pre- and post-scores isolates the treatment effect from baseline variation, making this one of the most widely used frameworks in experimental and quasi-experimental research across education, psychology, medicine, and the social sciences.
The intuition behind the method
Imagine you want to know whether a six-week reading intervention improves literacy scores. Simply measuring scores after the program is unreliable — scores might have risen anyway due to maturation or seasonal effects. The pretest-posttest design adds a measurement before the program starts: now you compare how much each group changed, not where they ended up. With a control group, any change common to both groups (e.g., natural growth) cancels out, leaving behind an estimate of the program's unique contribution. Adding random assignment upgrades this to a true experiment with strong causal warrant.
History
The pretest-posttest design gained systematic methodological treatment through Donald Campbell and Julian Stanley's 1963 monograph Experimental and Quasi-Experimental Designs for Research, which introduced the notation (O1 X O2) and catalogued threats to internal validity. Campbell and Stanley distinguished the true pretest-posttest control group design (with randomization) from the quasi-experimental one-group pretest-posttest design, identifying the latter's vulnerability to history, maturation, and testing threats. Shadish, Cook, and Campbell updated and extended the framework in 2002, cementing the design's place as a cornerstone of applied research methodology.
When to use it
Use a pretest-posttest design when you need to assess change over time due to an intervention and individual baseline differences are likely to introduce noise. It is appropriate when random assignment is feasible (true experiment) or when pretest scores are essential for controlling pre-existing group differences in quasi-experimental settings. Do not use it when a single post-only measurement is sufficient (e.g., outcome variance across people is negligible), when pretesting itself is likely to sensitize participants to the intervention (testing threat — consider a Solomon four-group design instead), or when only one time point of data collection is logistically possible.
Limitations and considerations
- Pretesting can sensitize participants to the treatment content or the researcher's hypothesis, inflating or deflating post-scores (testing threat).
- Mortality/attrition between pretest and posttest can bias results if dropout is related to treatment condition or baseline scores.
- Without randomization, selection bias remains: pre-existing differences between groups may not be fully captured by the pretest score alone.
- Regression to the mean: groups selected on the basis of extreme pretest scores will tend to score closer to the mean at posttest regardless of treatment.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Controls for pre-existing individual differences, increasing statistical power relative to posttest-only designs.
- Allows direct measurement of change, which is often the substantively meaningful quantity (e.g., learning gain, symptom reduction).
- Widely understood across disciplines and compatible with many analytic frameworks (ANCOVA, ANOVA, multilevel modeling).
- When combined with randomization, provides strong causal evidence for the treatment effect.
- Pre- and post-scores enable detection of differential change across subgroups (moderation analysis).
Applications
- Evaluating educational interventions — measuring learning gains before and after a curriculum, tutoring program, or instructional technology.
- Clinical and health research — assessing symptom reduction, functional improvement, or behavioral change following a medical or psychological treatment.
- Organizational training studies — measuring knowledge, skill, or attitude change in employees before and after a training program.
- Public policy evaluation — estimating the effect of a policy change or social program on outcomes such as employment, health, or academic achievement.
- Sport and exercise science — tracking fitness, performance, or psychological indicators across a structured training intervention.
How it works
How it works
All participants complete a measurement of the outcome variable (the pretest, O1) before any treatment is introduced. This baseline serves three purposes: it confirms group equivalence at the start (especially when combined with randomization), allows computation of change scores, and increases statistical power by controlling for individual differences.
Participants are randomly assigned to the treatment group or the control (or comparison) group. Random assignment is what distinguishes a true pretest-posttest experiment from a quasi-experimental variant. Without randomization, pre-existing differences between groups threaten causal inference, and the design is quasi-experimental.
The treatment group receives the intervention while the control group receives either no intervention, a placebo, or treatment as usual. The duration and fidelity of treatment delivery must be standardized and documented to ensure internal validity.
After the intervention period, all participants complete the same outcome measurement again (the posttest, O2). The interval between pretest and posttest should be long enough for the treatment to take effect but not so long that historical events or attrition severely confound results.
The primary analysis computes the change from pretest to posttest for each group and tests whether the treatment group's change is significantly greater than the control group's change. Common analytic approaches include analysis of covariance (ANCOVA) with the pretest as covariate, paired-samples t-tests within each group, or mixed-model repeated-measures ANOVA.
Common pitfalls
- Analyzing posttest scores without adjusting for the pretest — this discards the information the pretest provides and reduces power; use ANCOVA with pretest as covariate rather than a simple posttest comparison.
- Using a pretest instrument that is not the same measure as the posttest, making change scores uninterpretable; both waves must use identical or strictly parallel forms of the outcome measure.
- Ignoring the testing threat in sensitive contexts (e.g., attitude surveys, knowledge tests with transparent items) — if pretesting plausibly changes participants, consider a Solomon four-group design.
- Treating the pretest-posttest design as automatically causal without randomization — in quasi-experimental versions, unmeasured confounders can still explain the observed change difference.
- Applying raw gain scores (O2 - O1) as the dependent variable in regression — this approach has known psychometric problems; ANCOVA or residualized change scores are preferable.
Frequently asked questions
Should I use ANCOVA or a paired t-test to analyze pretest-posttest data?
ANCOVA with the pretest as a covariate is generally preferred for between-group comparisons because it removes pretest variance from the error term, increasing power, and adjusts for any residual baseline imbalance. A paired t-test within one group is appropriate for a one-group pre-post check but cannot isolate the treatment effect from maturation or history. Avoid using raw gain scores (post minus pre) as the dependent variable in ANCOVA — enter pretest and posttest as separate variables.
What is the testing threat and when does it matter?
The testing threat occurs when the act of completing the pretest itself changes participants — for example, by alerting them to what outcomes are valued, rehearsing test items, or reducing anxiety on the posttest. It matters most when the outcome instrument is a knowledge test with transparent content, or when participants are likely to reflect on or discuss their pretest responses before the posttest. If this is a serious concern, add two unpretestedgroups using the Solomon four-group design.
Is the pretest-posttest design quasi-experimental or experimental?
It can be either. With random assignment to treatment and control groups it is a true experiment (the classic pretest-posttest control group design of Campbell & Stanley). Without randomization — for example, when intact classrooms or existing groups are compared — it is a quasi-experimental design and faces stronger threats to internal validity, including selection bias.
How long should the interval between pretest and posttest be?
The interval should be long enough for the intervention to produce its expected effect, but not so long that history effects, attrition, or developmental changes unrelated to the treatment make change scores hard to interpret. There is no universal rule; the appropriate interval is determined by the theory of change underlying the intervention and prior evidence about effect timing.
Can I add more time points to a pretest-posttest design?
Yes — adding one or more follow-up measurements produces a repeated-measures or longitudinal design, which can reveal whether treatment effects persist, fade, or grow over time. This extension is sometimes called a pretest-posttest-follow-up design. Analytic approaches include mixed-model ANOVA or growth curve modeling.
Related methods
Sources
Campbell, D. T., & Stanley, J. C. (1963). Experimental and Quasi-Experimental Designs for Research. Rand McNally.
Open publication referenceCopied.Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin.
ISBN: 978-0395615560Copied.