Experiment vs survey: cause versus description
By Jude Wallis · Published
An experiment imposes treatments on units, and with random assignment it can support a cause-and-effect conclusion. A survey is an observational study that records answers to a fixed set of questions without imposing anything, so it can describe a population and show association but not cause.
AP Statistics: Unit 1 (topics 1.10, 1.13). In the Fall 2026 AP Statistics course, topic 1.10 of Unit 1 defines an experiment as imposing treatments and a survey as an observational study that uses a standard set of questions, and topic 1.13 covers experimental design, where random assignment justifies a cause-and-effect conclusion. Unit 1 is 20% to 30% of the multiple-choice section.
Experiment vs survey: the short answer
The split comes down to one question: did the researcher impose a treatment? An experiment is a study in which a researcher assigns conditions, called treatments, to experimental units to answer a question about a population. A survey is an observational study in which data are collected from people using a standard set of questions, with no treatment imposed.
That one difference decides what you may conclude. Because an experiment controls who receives each treatment, and random assignment balances outside factors across the groups, it can support a cause-and-effect claim. A survey only records what people already are or already do, so it can describe them and reveal associations, but a hidden third variable can always explain the pattern, which is why a survey shows association, not cause.
What an experiment is
An experiment assigns treatments to experimental units and measures a response. The explanatory variable, or factor, is the variable whose levels you impose; those imposed levels are the treatments; and the response variable is the outcome you measure afterward on each unit.
A well-designed experiment has four marks: at least two treatment groups to compare, one of which may be a control group; random assignment of treatments to units; replication, meaning more than one unit per treatment; and direct control of extraneous variables that could otherwise cloud the response. Random assignment is the engine of causation, because it makes the treatment groups as alike as possible on outside variables, so a difference in the response can be pinned on the treatment.
That is the power a survey lacks. When treatments are randomly assigned, a difference between the groups is fairly attributed to the treatment itself. For the wider contrast with studies that only observe, see experiments vs observational studies.
What a survey is
A survey is a specific kind of observational study: it collects data from humans using a standard set of questions and imposes no treatment. You ask, you record, and you summarize. Nothing is assigned, so nothing is controlled.
Surveys are the workhorse of describing a population. With a random sample they estimate parameters (a candidate's support, a mean commute time) and can flag relationships (people who exercise more report better sleep). What a survey cannot do is rule out a confounding variable, one associated with both the explanatory and response variables, that offers an alternative explanation for any relationship it finds.
So a survey answers what and how much and whether two things move together, but not why. To move from a survey's association to a causal claim you need an experiment, or at minimum an honest statement that cause is unproven. The gap between the two is the subject of correlation vs causation.
The differences side by side
| Feature | Experiment | Survey |
|---|---|---|
| Treatment | Imposed by the researcher | None; questions only |
| Study type | Interventional | Observational |
| Key randomness | Random assignment to treatments | Random selection of respondents |
| Supports cause and effect | Yes, with random assignment | No, association only |
| Main threat | Poor design, no control group | Confounding, biased sampling |
| Typical question | Does X change Y? | What is Y, and does it track X? |
The two kinds of randomness sit in different rows on purpose. Random assignment, inside an experiment, earns causation; random selection, in a survey, earns generalization to the population.
When to use which
Run an experiment when the question is causal, such as whether a drug lowers blood pressure or a teaching method raises scores, and when it is ethical and practical to assign treatments. Only random assignment lets you answer a why question, so if cause is the goal, no survey substitutes for it.
Run a survey when the goal is to describe a population or measure how common something is, when you cannot or should not impose a treatment, or when you are exploring which relationships are worth testing later. Surveys are cheaper, faster, and often the only ethical option, since you cannot assign people to smoke or to a hometown. A common research arc is a survey that spots an association, followed by an experiment that tests whether it is causal. To see how the questions themselves are framed, see the AP Statistics FRQ guide.
The classic mix-up and how to avoid it
The frequent slip is reading a strong survey result as proof of cause, especially when the sample is huge. A survey of a million people that finds coffee drinkers sleep worse still cannot show that coffee causes worse sleep, because stress or age could drive both. Sample size sharpens the estimate; it never creates causation.
Two checks keep it clean. First, look for an imposed treatment: if the researcher only asked questions and recorded answers, it is a survey, so the conclusion stops at association. Second, keep the two randomizations apart: random assignment is what an experiment uses to earn cause, while random selection is what a survey uses to generalize, and a study can have one, both, or neither. When a scenario only records what people report, resist any word like causes or reduces in the conclusion.
Same 5-point gap, two very different conclusions
Two studies compare a new study app with no app. In the survey, 100 students report whether they use the app, and app users average 84 on a test while non-users average 79. In the experiment, 100 students are randomly assigned to use the app or not, and the app group averages 84 while the control averages 79. Both find a 5-point gap. What can each conclude?
Compute the gap, which is identical in both studies: points in favor of the app.
Classify the survey. Students were only asked whether they use the app; no treatment was imposed, so this is an observational study. The 5-point gap is an association.
Judge the survey's conclusion. Because app use was self-selected, a confounder such as motivation could explain the gap: motivated students may both choose the app and score higher. The survey supports association only, not cause.
Classify the experiment. The app was randomly assigned, so this is an experiment. Random assignment balances motivation and other extraneous variables across the two groups.
Judge the experiment's conclusion. With those variables balanced, the same 5-point gap can be attributed to the app itself, so a cause-and-effect conclusion is justified.
Read the contrast. The number 5 is the same in both, yet only the experiment licenses the word causes. Design, not the size of the difference, decides what you may claim.
Both studies show a 5-point advantage for the app, but the survey can only say app use is associated with higher scores, while the experiment, thanks to random assignment, can conclude the app causes the higher scores. Identical arithmetic, different conclusions, because only the experiment imposed and randomized the treatment.
Frequently asked questions
Is a survey ever an experiment?
No. A survey collects answers to a standard set of questions and imposes no treatment, so it is always an observational study. An experiment must assign treatments to units. If a study only asks and records, it is a survey, whatever its size.
Can a survey ever prove cause and effect?
No. Without random assignment of a treatment, a confounding variable can always offer an alternative explanation for any association a survey finds. A survey can motivate a causal question and estimate how common something is, but the causal conclusion has to come from an experiment.
Why does random assignment matter more than sample size for cause?
Random assignment balances extraneous variables across the treatment groups, which is what isolates the treatment's effect. A larger sample only makes an estimate more precise; it does not remove confounding, so a huge survey still shows association, not cause.