Not sure which lesson to do first? Find your level and get a plan →

Statistics · Grade 6-8

Experimental Design

Experimental design is the careful planning of experiments to establish cause-and-effect relationships by controlling variables, using comparison groups, and randomly assigning subjects to treatment and control conditions to isolate the effect of interest.


Quick Answer

Experimental design is the careful planning of experiments to establish cause-and-effect relationships by controlling variables, using comparison groups, and randomly assigning subjects to treatment and control conditions to isolate the effect of interest. Use it when you want to know whether something works, not just whether two things go together: split subjects into groups by chance, give one group the treatment and the other a comparison condition, keep everything else the same, and compare the outcomes. Because the groups started out alike, a clear difference at the end points to the treatment.

Why This Matters

Good design is how data can answer "does it work?": whether a medicine helps, a teaching method raises scores, or a fertilizer grows bigger plants. Without a comparison group, ordinary improvement looks like an effect; without random assignment, eager volunteers make a treatment look better than it is; and changing two things at once leaves you unable to say which one mattered.

Intuitive Explanation

Say you want to know whether a study app raises quiz scores. Giving it to the whole class and watching scores rise proves little: scores might have risen anyway, because students learn over time. You need a comparison. Split 40 students into two groups of 20, one using the app and one studying the usual way, give both the same quiz on the same day, and compare. The group without the app shows what would have happened anyway.

Three features make the comparison fair. A control group gives the baseline. Random assignment, such as drawing names from a hat to decide who gets the app, spreads hard workers and less motivated students across both groups, so the groups start out alike. Controlling other variables keeps the quiz, the time limit and the room the same for everyone. Then if the app group averages 78 and the other group 72, the 6-point gap can be credited to the app, allowing for some luck.

Compare a study where students chose whether to use the app. Those who chose it might average 80 against 70 for the rest, a bigger gap, but students who download a study app are likely to be the keener ones to begin with. Nobody assigned the treatment, so this is an observational study, and its gap mixes the app's effect with the students' motivation. Who decided which students got the treatment is what separates the two designs.

Think of it like this. A fair experiment is like a race between two teams on one track: same start line, same weather, and only one thing differs, say the shoes. Then a gap at the finish says something about the shoes. With people, the teams are made alike by assigning members at random, since you cannot line up identical people.

Core idea. To show that a treatment causes an effect, compare groups that differ only in the treatment: give it to one group and not the other, let chance decide who goes where, and keep everything else the same.


When to Use

Use experimental design when the question is whether something causes a change and you are able to assign the treatment yourself: a new app, a fertilizer, a medicine, a different practice routine. Plan a control group, assign subjects by chance, and keep everything else the same. If you can only observe what people already do, it is an observational study, which shows association; if the goal is to describe a population, you need random sampling instead.

Tip. Ask: Are you planning a test where you decide who gets a treatment and keep everything else the same between groups?

How to Recognize It

Before you plan or judge an experiment, find what is being changed on purpose, who gets it, and what gets measured at the end.

  1. Does someone deliberately give a treatment, such as a new app, a fertilizer or a medicine, to some subjects?

    Yes means an experiment can be planned. If researchers only record what people already do, it is an observational study; see Observational vs Experimental Studies.

  2. Is there a comparison group that does not get the new treatment?

    Yes lets you measure what the treatment adds. With no comparison group, any change could come from time or practice, so the plan is missing a Control Group.

  3. Were subjects put into groups by chance rather than by choice?

    Yes keeps the groups alike at the start. If people chose their own group, their differences could explain the result, and the missing step is Random Assignment.

  4. Does only one thing differ between the groups?

    Yes means a difference at the end points to that one thing. If two things changed together, such as a new app and a new teacher, you cannot tell which one caused it.

Experimental Design vs Observational vs Experimental Studies vs Control Group vs Random Assignment

These all come up when someone wants to know whether something works. Experimental Design is the whole plan; the others are a way to sort studies already done, or one part of the plan.

Experimental Design

Meaning
Experimental design is the plan for an experiment: who gets the treatment, how groups are formed, what stays the same, and what is measured.
Key test
Asks you to plan or judge a whole test from start to finish.
Example
Split 40 bean plants by chance, fertilize 20, keep light and water equal, and compare heights after 3 weeks.

Observational vs Experimental Studies

Meaning
This idea sorts studies by whether researchers handed out the treatment or only recorded what already happened.
Key test
Asks you to classify a study that someone already ran.
Example
Comparing students who already eat breakfast with those who skip it is observational.

Control Group

Meaning
A control group is the comparison group that does not get the new treatment, showing what happens without it.
Key test
Names one part of the plan: the baseline group.
Example
The 20 bean plants that get only water.

Random Assignment

Meaning
Random assignment uses chance to decide which group each subject joins, so the groups start out alike.
Key test
Names one step of the plan: how subjects are sorted into groups.
Example
Drawing names from a hat to decide who uses the new study app.

Formula & Notation

Worked Examples

Example 1 — Design a study-app experiment

Easy

A teacher has 40 students and wants to know whether a new study app raises quiz scores. Design an experiment. If the app group then averages 78 and the other group averages 72, what can she conclude?
Show the steps and the answer
  1. Name the parts: treatment = using the app; response = quiz score.

    An experiment changes one thing on purpose and measures what happens to another.

  2. Create a control group: 20 students study the usual way, without the app.

    Scores rise with ordinary practice too, so the control group shows how much they rise without the app.

  3. Assign by chance: put the 40 names in a hat and draw 20 for the app group.

    Chance spreads strong and weak students across both groups, so the groups start out alike.

  4. Keep everything else equal: same quiz, same day, same time limit, same room.

    If anything else differed between the groups, it could explain the result instead of the app.

  5. Compare averages: 78−72=6 points.

    With groups made alike and only the app different, the gap is evidence the app helped, though a small part may be luck.

Answer: Randomly assign 20 students to use the app and 20 to study normally, give both the same quiz under the same conditions, and compare. A 6-point higher average for the app group is evidence that the app raises scores.

Takeaway: Control group, random assignment, everything else equal: with all three, a difference in outcomes points to the treatment.

Example 2 — Same comparison, no experiment

Standard

At another school, students decided for themselves whether to use the study app. The 25 who chose it averaged 80 on the quiz; the 35 who did not averaged 70. Is this an experiment? Does it show that the app adds 10 points?
Show the steps and the answer
  1. Ask who decided the treatment: each student chose.

    If the researcher did not assign the treatment, the study is observational, however neat the comparison looks.

  2. List how the groups may differ before the quiz: students who choose a study app are likely more motivated and may already have higher grades.

    Motivation can raise scores on its own, so it is mixed up with the app's effect.

  3. Compare with the experiment: there the gap was 6 points with assigned groups; here it is 80−70=10 points with chosen groups.

    Self-chosen groups can make a treatment look better than it is, because part of the gap belongs to who chose it.

Answer: No. It is an observational study, so the 10-point gap shows an association that may be partly or wholly due to motivation, not proof that the app adds 10 points.

Takeaway: A comparison is an experiment only when the researcher assigns the treatment, ideally by chance.

Example 3 — Find the mistake: two changes at once

Application

Ella tests a fertilizer. Group A gets fertilizer and sits by the sunny window; Group B gets no fertilizer and sits in a shady corner. After 3 weeks, A's plants grew an average of 12 cm and B's 7 cm. Ella writes: "The fertilizer added 5 cm of growth." Find the mistake and correct it.
Show the steps and the answer
  1. List the differences between the groups: fertilizer and sunlight.

    Two things changed at once, and either one can make plants grow faster.

  2. Test Ella's claim: the 12−7=5 cm gap could come from the fertilizer, the sun, or both.

    When two variables change together, the experiment cannot separate their effects; they are confounded.

  3. Redesign: put all the plants in the same spot, randomly choose half for fertilizer, and keep water and soil the same.

    With sunlight equal for everyone, a growth gap can only come from the fertilizer or chance.

Answer: The 5 cm cannot be credited to the fertilizer, because sunlight also differed. Repeat the test with all plants in the same light and fertilizer assigned at random.

Takeaway: Change only the treatment between groups; every other difference becomes a rival explanation.

Where You'll See It

  • Testing a new fertilizer

    Split identical plants by chance into a treatment group (gets fertilizer) and a control group (doesn't), keep light and water the same, and compare growth — so any growth difference is credited to the fertilizer.

  • Does a new study app raise scores

    Randomly assign students to use the app or not, keep their classes and teachers comparable, then compare test gains; without random assignment and a control, you can't tell if the app or the keener students caused the change.

  • Which paper airplane flies farther

    To fairly test two designs, throw each many times from the same spot with the same force and measure distance, changing only the design — controlling other variables isolates the design's effect.

Common Mistakes

  1. The slip: Giving the treatment to everyone and seeing improvement proves it works.

    Why it's tempting: If everyone improves, the treatment seems clearly responsible.

    Instead: Without a control group you can't know what would have happened anyway (growth, healing, practice). You need a comparison group that doesn't get the treatment.

  2. The slip: Changing several things at once is efficient and still tells you the cause.

    Why it's tempting: It feels faster to improve many things together.

    Instead: If you change multiple variables at once you can't tell which one caused the effect (confounding). Vary one factor at a time, or use a design that separates effects.

  3. The slip: Letting subjects pick their own group is fine as long as both groups exist.

    Why it's tempting: Having two groups looks like a fair comparison.

    Instead: Self-selection makes the groups differ before treatment, so the comparison is unfair. Subjects must be randomly assigned to make the groups comparable at the start.


Mini Practice

Try these on your own. Tap Reveal when you want to check.

  1. Maya tests whether music helps plants grow. 10 plants hear music and 10 sit in silence, with the same soil, water and light. After 3 weeks the music plants average 14 cm and the silent plants 11 cm. Name the treatment, the control group, and the response variable.

    Hint: The response is what you measure at the end; the control is the comparison group.

  2. Ravi tests a plant fertilizer. Group A gets fertilizer and sits by a sunny window; Group B gets none and sits in a dim corner. Group A grows 6 cm more. Can he say the fertilizer caused the difference?

    Hint: Count how many things differ between the two groups.

  3. Twenty volunteers will test whether a memory game improves recall. Describe a fair way to split them into a game group and a no-game group of 10 each.

    Hint: Let chance, not the researcher or the volunteers, decide who goes where.

  4. In an experiment, 30 students are randomly assigned to two groups of 15. One group uses a new study app; the other studies as usual. Mean quiz scores are 78 and 71. What is the difference, and what does the design let you conclude?

    Hint: Subtract the means. Then think about what random assignment does to other differences between the groups.

  5. A survey of 200 students finds that those who eat breakfast have higher grades. Is this an experiment that shows breakfast causes better grades, or a correlation?

    Hint: Did anyone decide who eats breakfast?

  6. Lena gave 12 classmates an energy drink before a 100 m race and wrote: "Their average time was 15.2 seconds, so the drink makes you faster." What is missing from her experiment?

    Hint: Faster than what? What would the times have been without the drink?

Want the full set?

50 practice questions for this concept — free to try, every one with a complete worked solution showing the why, not just the answer.

Frequently Asked Questions

How many subjects does an experiment need?

More than you might expect. With only a few subjects, chance alone can make one group stronger from the start, so a difference at the end could be luck. Giving the treatment to many subjects, called replication, lets those chance differences even out. Scientists also repeat whole experiments to see whether the result holds up. There is no single magic number, but two plants in each group tells you almost nothing.

Is it ever wrong to run an experiment?

Yes. You cannot assign people to something you expect to harm them, such as smoking, skipping sleep for weeks or eating unsafe food, just to measure the damage. Experiments on people need their informed consent and approval from an ethics board. When an experiment would be unethical, scientists rely on observational studies and careful reasoning instead, which is how smoking was linked to lung cancer.

What if my subjects are very different from each other, like fast and slow typists?

Then group similar subjects first and use chance inside each group. To test a typing app, split students into fast and slow typists, then flip a coin within each group to decide who uses the app. This is called a block design. It keeps one big difference between subjects from hiding the treatment's effect, and it lets you compare like with like.

Learning Path

Experimental Design

You are here

Next →

Blinding
You need Correlation vs Causation (why a link is not proof of a cause) and Data Collection (surveys, observations and experiments). Experimental design adds the tools that make a causal claim fair: a control group, random assignment and holding other variables fixed. Next, Blinding removes another big source of bias, people's expectations about which treatment they received.

See Also