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

Math · Statistics & Probability · Grade 6-8

Experimental Design

Experimental design is the deliberate plan for a study where the researcher imposes a treatment and uses a control group, random assignment, replication, and blinding to isolate cause and effect.


Quick Answer

Experimental design is the deliberate plan for a study where the researcher imposes a treatment and uses a control group, random assignment, replication, and blinding to isolate cause and effect. Use it when you want to find out whether something causes a change: does a new fertilizer make plants taller, does music help people memorize words? You let chance decide who gets the treatment, compare with a control group treated the same in every other way, and use enough subjects that one odd result can't decide the answer.

Why This Matters

Without a good design, a study can't separate the treatment's effect from everything else that differs between the groups. If the gardener puts the fertilized plants in the sunniest spot, any extra growth might come from sunlight. Random assignment, a control group and many subjects are what allow a result to say "this caused that," which is why medicines must pass randomized trials before approval.

Intuitive Explanation

A gardener wants to know whether a new fertilizer makes tomato plants taller. She has 30 young plants. If she fertilizes the 15 healthiest-looking ones, they might grow taller simply because they started out healthier. So she numbers the plants and draws 15 numbers from a hat: those plants get fertilizer, the rest get plain water. Chance, not her judgment, decides, so strong and weak plants end up spread across both groups.

Each other part of the design blocks a specific problem. The control group shows how much the plants would grow anyway. Keeping everything else equal (same soil, same sunlight, same watering schedule) leaves the fertilizer as the only planned difference. Using 15 plants per group instead of 1 means a single plant that wilts or shoots up can't decide the result. And if a helper who doesn't know which plants were fertilized does the measuring, hopes can't nudge the numbers.

Having two groups does not make a study an experiment. If the gardener instead compares neighbors who already use the fertilizer with neighbors who don't, she is only observing. Those neighbors might also water more or have sunnier yards, so a height difference can't be pinned on the fertilizer. What separates the two is how each plant ended up in its group: by the researcher's random draw, or by the owner's own choice.

Think of it like this. A fair experiment is like a fair race: same starting line, same track, same clock, so the finish order reflects speed and nothing else. An experiment keeps its groups alike in every way except the treatment, so a difference at the end points to the treatment. Unlike runners, subjects are sorted into groups by chance, because no one can match them up evenly by hand.

Core idea. An experiment deliberately assigns treatments using control, randomization, replication, and blinding so a difference can be pinned on the treatment.


When to Use

Use experimental design when the question asks whether something causes a change and you can assign the treatment yourself: plan a control group, random assignment, enough subjects, and blinding where possible. If you can't or shouldn't assign the treatment (smoking, a person's age, which school someone attends), you can only run an observational study and claim an association. If the question is about choosing whom to survey from a population, that is a sampling method.

Tip. Ask: Does the researcher actively assign subjects to treatments (rather than just observe what they already do)?

How to Recognize It

Before you plan or judge an experiment, look at what the study wants to find out, who decided which subjects got the treatment, and what each group was compared against.

  1. Is the question whether one thing causes a change in another, such as whether a fertilizer makes plants taller?

    If yes, an experiment is the tool that can answer it. If the question only asks what share of people do something, or how two things vary together, a fair survey (sampling methods) or an observational study is enough.

  2. Could the researcher safely and fairly hand out the treatment, or would it mean making people smoke or change schools?

    If the treatment can be handed out, design an experiment. If it cannot, because it is harmful or is a trait like age, you can only run an observational study and report a link, not a cause.

  3. Was a coin flip, a random number or a shuffled list used to decide who got which treatment?

    If yes, the groups should be alike except for the treatment, so a difference can be blamed on it. If subjects chose their own group, or were grouped by a trait they already had, the study is observational even though it has groups.

  4. Is there a comparison group that gets a placebo or the usual treatment, with many subjects in each group?

    If yes, the design has control and replication. With only one group, or one plant per treatment, a confounding variable or a single fluke could explain the result, so the conclusion is weak.

Experimental Design compared with Observational study, Sampling methods, and Confounding variable

These rows all come up when a study has groups, so they blur together. Ask what each step decides: who joins the study (sampling), who gets the treatment (design), nothing at all because you only watch (observational), or which hidden factor randomizing guards against (confounding).

Experimental Design

Meaning
Use this when you are planning a study and need to impose treatments with control, randomization, and replication to claim cause and effect.
Key test
Does the researcher actively assign subjects to treatments (rather than just observe what they already do)?
Example
A gardener wants to know if a new fertilizer makes tomato plants grow taller. How should the study be designed?

Observational study

Meaning
Records data on groups that already exist without imposing any treatment.
Key test
Use when you cannot or should not assign treatments and can only watch.
Example
Comparing smokers to non-smokers as they are

Sampling methods

Meaning
How you SELECT subjects from a population, not how you assign them to treatments.
Key test
Use when the concern is who gets into the study, not what treatment they receive.
Example
Picking 50 students by SRS to enroll

Confounding variable

Meaning
A lurking variable that random assignment is designed to neutralize; the threat, not the design.
Key test
Use when explaining WHY randomization is needed.
Example
Sunequal sunlight across the two trays

Formula & Notation

How to read it: Treatments are often labeled T1,T2,… Control group is C. Randomization is denoted by R.

Worked Examples

Example 1 — Fertilizer test

Easy

A gardener wants to know if a new fertilizer makes tomato plants grow taller. How should the study be designed?
Show the steps and the answer
  1. Choose the treatment and the control: fertilizer mixed in water versus the same amount of plain water.

    The control group shows how tall plants grow without fertilizer, so the gap between groups measures the fertilizer's effect.

  2. Randomly assign 30 similar plants: number them 1 to 30 and draw 15 numbers for the fertilizer group.

    Chance spreads stronger and weaker plants evenly, so neither group starts with an advantage.

  3. Keep conditions equal: the same soil, sunlight and watering schedule for all 30 plants.

    If the fertilized plants also got more sun, extra growth could come from the sun instead.

  4. Use 15 plants per group, and have a helper who doesn't know the groups measure the heights after 6 weeks.

    Many plants per group stop one unusual plant from deciding the result; blinding keeps expectations out of the measurements.

Answer: A randomized, controlled, replicated experiment with blinding

Takeaway: Random assignment, a control group, equal conditions and replication leave the treatment as the only explanation for a difference.

Example 2 — Just watching gardeners

Standard

Instead, the gardener surveys 30 neighbors, recording which already use the fertilizer and how tall their tomatoes are. Is that an experiment?
Show the steps and the answer
  1. Ask how the groups were formed: each neighbor chose whether to use the fertilizer.

    The gardener assigned nothing; she only recorded what was already happening.

  2. List other differences that could explain taller plants: neighbors who buy fertilizer may also water more, weed more or have sunnier yards.

    These confounding variables travel together with fertilizer use.

  3. Conclude that it is an observational study, so a height difference shows association, not cause.

    Without random assignment, the fertilizer's effect can't be separated from the gardening habits that come with it.

Answer: No — it's an observational study

Takeaway: Two groups and a measurement don't make an experiment; the researcher must assign the treatment.

Example 3 — Find the mistake: a control group that got nothing

Application

A class tests whether a sports drink improves sprint times. 20 students are randomly split into two groups of 10. Group A drinks the sports drink and Group B drinks nothing. Group A's average 50 m time improves by 0.4 s and Group B's by 0.1 s. Ben writes: "The drink made runners 0.3 s faster." Find the mistake and correct it.
Show the steps and the answer
  1. Check what already works: random assignment and 10 students per group.

    Those parts are sound, so the problem lies in what the control group received.

  2. List every difference between the groups: Group A got the drink's ingredients, extra fluid, and the belief that they had been given something helpful.

    Group B got none of these, so all three are mixed into the gap of 0.4−0.1=0.3 s.

  3. Fix the control: give Group B a placebo drink with the same color, taste and amount but no active ingredients, and don't tell students which drink they got.

    Now fluid and expectations match across the groups, and the active ingredients are the only difference.

  4. Rerun the test and credit the drink only with the gap between Group A and the placebo group.

    Only a difference measured against a look-alike control isolates the drink's own effect.

Answer: The 0.3 s gap can't be credited to the drink, because the control group got nothing at all. Rerun the test with a look-alike placebo drink for Group B, given blind.

Takeaway: A control group should get everything the treatment group gets except the treatment itself: a placebo, not nothing.

Where You'll See It

  • Clinical trials for new medicines

    Before a medicine is approved, thousands of volunteers are assigned by chance to the medicine or a look-alike placebo, and often neither patients nor doctors know who got which (double-blind). Only if the medicine group does clearly better can the maker claim the medicine caused the improvement.

  • Testing two website layouts

    An online store shows a random half of its visitors a new page layout and the rest the old one. If 4.8% of new-layout visitors buy something against 4.0% of the others, random assignment means the layout, not the time of day or type of shopper, explains the gap, provided both groups are large.

  • Seed trials on farm plots

    Seed companies divide a field into many small plots and assign each plot a variety at random. Spreading every variety across wet and dry, sunny and shady plots stops one variety from looking best just because it landed in the richest soil.

Common Mistakes

  1. The slip: Skipping random assignment but still claiming causation

    Instead: only randomized assignment to treatments justifies a cause-and-effect conclusion.

  2. The slip: Using one subject per treatment

    Instead: replication (many subjects) is needed so one fluke result doesn't drive the conclusion.

  3. The slip: Confusing the control group with 'do nothing'

    Instead: the control gets a placebo or standard treatment so the only difference is the treatment itself.


Mini Practice

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

  1. A teacher wants to test whether background music helps 40 students memorize words. Describe how to split them into two groups of 20 using random assignment.

    Hint: Every student should have the same chance of landing in either group.

  2. 60 bean plants are randomly split: 30 get a new fertilizer, 30 get plain water. Heights are measured after 4 weeks. Name the treatment, the control group, and the response.

    Hint: The response is what gets measured at the end.

  3. Design A gives fertilizer to 1 plant and plain water to 1 plant. Design B uses 25 plants in each group. Which design is better, and why?

    Hint: What could happen if one plant happens to be sick?

  4. In a headache-pill trial, 50 volunteers get the pill and 50 get a sugar pill that looks the same. No one knows which they got. Why give a sugar pill instead of nothing?

    Hint: Think about how people feel just from taking any pill.

  5. A researcher looks up the records of 500 students and compares grades of those who play an instrument with those who do not. Experiment or observational study?

    Hint: Did the researcher decide who plays an instrument?

  6. Jo lets 30 students choose whether to use a new math app. Then Jo says: "App users scored higher, so the app caused the higher scores." Correct Jo.

    Hint: Who decided which students used the app?

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

What is a placebo, and why would anyone give a fake treatment?

A placebo is a dummy treatment, like a sugar pill, that looks just like the real one. People often feel better simply because they believe they are being treated; this is called the placebo effect. If the control group gets nothing at all, you cannot tell whether the real treatment worked or the belief did. When both groups expect help, any extra improvement belongs to the treatment.

What does double-blind mean?

In a single-blind experiment the subjects do not know which group they are in. In a double-blind one, the people who give the treatment and measure the results do not know either. A researcher who knows who got the new pill might, without meaning to, encourage that group or judge their results more kindly. Hiding the group labels from everyone until the end keeps both sides honest.

If random assignment is fair, can the groups still turn out uneven by bad luck?

Yes, especially with small groups. If you randomly split 6 people into two groups of 3, the three tallest could all land in one group. That is one reason replication matters: with 50 people per group, chance differences tend to even out. When one trait is known to matter a lot, researchers can also block: sort subjects by that trait first, such as age group, then assign treatments at random within each block.

Can an experiment be done on plants or objects instead of people?

Yes. The subjects, often called experimental units, can be plants, batteries, paper airplanes or plots of land. A gardener testing a fertilizer might randomly pick 10 of 20 pots to fertilize and water all 20 the same way. The same rules apply: random assignment, a control group and enough units. A tomato plant cannot be fooled by a placebo, but the person measuring the plants should still not know which pots were fertilized.

Learning Path

Experimental Design

You are here

You need Causation, Sampling Bias and Variability first: what it means for one thing to cause another, how a skewed selection distorts results, and why repeated measurements differ. Experimental design adds the tools that rule out other explanations: control, randomization, replication and blinding. Next, Observational vs Experimental Studies uses these ideas to judge which published studies can claim cause and which show only association.

See Also