Skip to main content

Matched Pairs Design in AP Statistics

A matched pairs design is an experiment comparing exactly two treatments, where similar experimental units are paired on a variable that affects the response and a random process decides which member of each pair gets which treatment. In the other version, each unit receives both treatments and the order is randomized. It's a randomized block design with two treatments, where each block is either a pair of matched units or one unit measured under both treatments.

This definition is part of Fiveable's AP Statistics review, which also has guided practice questions.

Verified for the 2027 AP Statistics exam•Last updated November 2026•Fiveable Content Team

The two versions: twin units or one unit used twice

In the first version, you build pairs out of separate units that match on something that matters, such as identical twins, two students with the same prior test score, or two plots of land with the same soil. Then you randomize within each pair. A coin flip for each pair decides which member gets Treatment A, and the other member gets Treatment B.

In the second version, each unit is its own block. A subject tries both drugs, or one person gets Sunscreen A on one arm and Sunscreen B on the other. Here the randomization decides the order of treatments, or which arm gets which sunscreen.

Either way, a matched pairs design has exactly two treatments. A study comparing three fertilizers can still use blocking, but it would be a general randomized block design.

Writing a matched pairs plan for Question 1 of the written section

Suppose a prompt asks you to design a study comparing Sunscreen A and Sunscreen B on a group of volunteers. That task comes from Unit 1, Topic 1.13: Experimental Design, which is about building experiments that can support cause-and-effect conclusions. When Question 1 (Multi-Focus, Practices 1 and 2) or the data collection part of Question 4 asks you to describe an experiment like this, a strong matched pairs answer covers these pieces in order:

  • Name the pairing variable and say how it affects the response (skin type affects how fast a person burns).
  • Say exactly how the pairs are formed (each person's two arms, or twins matched with each other).
  • Describe the randomization so someone could carry it out: for each person, flip a coin, and if it lands heads the left arm gets Sunscreen A and the right arm gets B.
  • Name the response you measure on both members of the pair.
  • Compare within pairs by finding each pair's difference (A minus B) and looking at the mean of those differences.

For the randomization piece, describe a method someone could actually carry out, such as a coin flip, a die roll, or a random number generator, and state what each outcome means. The phrase "randomly assign the treatments" on its own doesn't tell the reader how the assignment happens.

Justifying matched pairs over a completely randomized design

Part of the experimental design topic is explaining why a design fits a study. For matched pairs, the argument has three parts: the units differ a lot on some variable, that variable affects the response, and pairing lets you compare treatments on units that share it.

Take the sunscreen study. If you split people into two random groups, each group would contain a mix of skin types, and that mix adds variability to the burn data. A real difference between sunscreens could get lost in it. When both sunscreens go on the same person, skin type is the same for both arms, so the within-person difference mostly reflects the sunscreen. That's the payoff of blocking: comparisons between treatments get more precise.

A multiple-choice question might describe the units and ask which feature points toward matched pairs. Look for units that come in natural pairs (twins, two arms, the left and right tires on one car) or units that vary widely on something tied to the response. In the tire example, putting Brand A on a randomly chosen side of each car means both brands face the same driver, roads, and mileage.

Pairing still needs randomization, and other mix-ups

A common wrong answer treats pairing as a replacement for random assignment. It isn't one. Pairing controls one extraneous variable, but the coin flip inside each pair is what guards against confounding from everything else, and random assignment is what lets you draw a cause-and-effect conclusion.

Another mix-up calls any blocked design "matched pairs." If a blocked experiment has more than two treatments, or blocks made of more than two units, it is a randomized block design but not a matched pairs design. Matched pairs is the special case with two treatments.

When one unit gets both treatments, watch for carryover, where the first treatment still affects the second measurement. Randomizing the order spreads that effect across both treatments. If carryover is serious, pairing separate units can be the better choice.

Turning the pairs into one sample of differences in Unit 4

The design shows up again in Topics 4.2 and 4.4. Matched pairs data are dependent samples, so you subtract within each pair and analyze one sample of differences with a one-sample t-interval or t-test for the population mean difference, μ_d. The null hypothesis is H₀: μ_d = 0, and you have to state the order of subtraction in context (for example, burn score with A minus burn score with B).

Suppose a study uses 20 pairs of twins. That gives 20 differences, so n = 20 and df = 19. Since 20 is under 30, the conditions check uses a graph of the 20 differences, which should show no strong skewness or outliers. The randomization condition is met by the coin flip inside each pair.

On Question 3 (Inference), a common error is running a two-sample t procedure on paired data. That treats the two columns as independent groups and throws away the pairing that made the design worth using.

Where this sits in AP® Statistics

Keep studying AP® Statistics Unit 1