A true experiment requires three things: the researcher actively manipulates a variable, participants are randomly assigned to groups, and a control group provides a baseline for comparison. These elements work together to do something no other research design can reliably do: establish cause and effect. Without all three, you have a study, but not a true experiment.
The Independent Variable Must Be Manipulated
The defining feature of a true experiment is that the researcher deliberately changes something and measures what happens. The thing being changed is the independent variable, and the researcher must actively intervene to alter it. This is a stricter requirement than it sounds. Studying people who already differ on some characteristic, like comparing smokers to nonsmokers, doesn’t count as manipulation. The researcher has to be the one creating the difference between groups.
In a classic example, researchers studying bystander behavior told participants that either one, two, or five other people were present during a discussion. The researchers created those three conditions on purpose, controlling exactly what each participant experienced. That active intervention is what separates a true experiment from an observational study, where you simply measure things as they naturally occur.
Manipulation matters because it eliminates what researchers call the third-variable problem. If you just compare people who already differ in some way, you can never be sure whether the difference you’re measuring is caused by the variable you care about or by some other trait those people happen to share. When the researcher creates the conditions, that problem disappears.
Random Assignment Creates Equal Groups
Random assignment means every participant has an equal chance of landing in any group. It’s the mechanism that makes the groups comparable before the experiment even begins. If you’re testing whether a specific exercise prevents hamstring injuries in soccer players, you need the group doing the exercise and the group not doing it to be similar in age, fitness level, injury history, and every other factor that could influence the outcome. Random assignment accomplishes this without the researcher needing to identify and match every possible variable.
This is different from random selection, which is about how you recruit participants from a larger population. Random selection helps you generalize your findings to a broader group. Random assignment helps you draw cause-and-effect conclusions within your study. A true experiment requires random assignment. Random selection is valuable but not strictly necessary for the design to qualify.
When random assignment is missing, you get what’s called a quasi-experiment. Quasi-experimental designs still manipulate an independent variable, but they use pre-existing groups instead of randomly formed ones. A study comparing outcomes at two hospitals, one that adopted a new protocol and one that didn’t, is quasi-experimental. The hospitals weren’t randomly assigned to their conditions, so underlying differences between them could explain the results.
The Control Group Sets the Baseline
A control group receives no treatment, a standard treatment, or a placebo. Its purpose is to show what happens in the absence of the thing you’re testing, so you have something meaningful to compare your results against. Without a control group, you can’t tell whether changes in your experimental group were caused by your intervention or by the passage of time, participants’ expectations, or some unrelated event.
The distinction between having a control group and having an effective control group matters more than most people realize. In drug trials, for instance, simply giving one group a sugar pill doesn’t automatically make the trial well controlled. If participants can tell whether they received the real treatment or the placebo (through side effects, taste, or other cues), the comparison breaks down. The groups are no longer experiencing identical conditions except for the variable being tested.
This is where blinding comes in. In a single-blind study, participants don’t know which group they’re in. In a double-blind study, neither participants nor the people administering the treatment know. Blinding prevents expectations from contaminating the results. A participant who knows they received the real drug may report feeling better simply because they expect to. A researcher who knows which group a participant belongs to may unconsciously treat them differently or interpret their results more favorably.
Why These Elements Establish Causation
Cause-and-effect claims require three conditions. First, the cause has to come before the effect in time. Second, there has to be a relationship between the two: when the cause is present, the effect changes. Third, and most critically, there can’t be an alternative explanation for the relationship. True experiments are built to satisfy all three.
Manipulation ensures the researcher controls the timing: the independent variable is introduced before the outcome is measured. The design itself creates the temporal sequence. The comparison between experimental and control groups reveals whether a relationship exists. And random assignment, combined with controlled conditions, rules out alternative explanations by making the groups equivalent in every way except the variable being studied.
This is the logic behind the counterfactual model of causation, which is the standard framework in health and social sciences. The true causal effect of a treatment is the difference between what actually happened to the treated group and what would have happened to those same people if they hadn’t been treated. Since you can’t observe both scenarios for the same individuals, the control group serves as the best possible stand-in for that “what if” scenario. Random assignment is what makes the stand-in trustworthy.
Threats That Undermine an Experiment
Even a well-designed true experiment can produce misleading results if certain problems go uncontrolled. Researcher Donald Campbell identified seven classic threats to an experiment’s internal validity: its ability to accurately link cause to effect.
- History: An outside event occurs during the study that affects the outcome. If you’re testing a stress-reduction technique and a major news event spikes everyone’s anxiety midway through, your results are compromised.
- Maturation: Participants change naturally over time. People may get better, more tired, or more practiced simply because time has passed, not because of the treatment.
- Testing effects: Taking a pretest can influence how participants perform on a posttest, independent of any treatment.
- Instrument decay: The measurement tool changes over the course of the study. Observers may become less attentive, or equipment may drift in accuracy.
- Statistical regression: Participants selected for extreme scores (very high pain, very low performance) tend to score closer to average on retesting, regardless of treatment.
- Selection bias: The groups differ in important ways before the experiment starts. Random assignment directly addresses this threat.
- Attrition: Participants drop out unevenly across groups. If people in the treatment group quit because the intervention is unpleasant, the remaining participants may not represent the original group.
True experimental designs control for most of these threats simultaneously. Having a control group that experiences the same time passage, testing schedule, and measurement tools as the experimental group means that history, maturation, testing, and instrument decay affect both groups equally and cancel out in the comparison.
True Experiments vs. Other Designs
The landscape of research designs is a spectrum, and true experiments sit at one end. Understanding where the line falls helps you evaluate the strength of any study’s conclusions.
Quasi-experiments manipulate an independent variable but lack random assignment. A school district testing a new teaching method in some classrooms but not others is a quasi-experiment if classrooms weren’t randomly selected. These designs can suggest causation, but they can’t rule out the possibility that pre-existing differences between groups drove the results.
Observational studies don’t manipulate anything. Researchers measure variables as they naturally occur and look for relationships. Cohort studies, case-control studies, and surveys all fall here. They can identify strong associations, like the link between smoking and lung cancer, but they can’t prove causation on their own because there’s no way to rule out every possible confounding factor.
Pre-experimental designs are the weakest category. A one-group pretest-posttest design, where you measure a group, apply a treatment, and measure again, has no control group and no random assignment. Any change you observe could be due to the treatment, the passage of time, practice effects, or something else entirely.
What a True Experiment Looks Like in Practice
Consider a randomized controlled trial testing whether a specific exercise prevents hamstring injuries in amateur soccer players. Researchers randomly assigned players to either perform the exercise regularly or continue their normal routine. Over a full year, the group performing the exercise had significantly fewer hamstring injuries. Because players were randomly assigned, the two groups were comparable at the start. Because the researchers controlled the intervention, they could be confident the exercise itself drove the difference.
Another trial tested whether adding Pilates to standard pain medication improved outcomes for people with chronic low back pain. Participants were randomly assigned to either medication alone or medication plus Pilates. The Pilates group experienced significant improvements in pain, daily function, and quality of life, and they used less pain medication. Random assignment ensured the groups were similar before treatment began, and the control group (medication only) provided a clear baseline, making the causal link between Pilates and improvement credible.
Both studies share the same core structure: active manipulation of a variable, random assignment to groups, and a control condition for comparison. That structure is what earned them the label of true experiments and what gives their conclusions real weight.