A matched pairs t-test is a significance test for the mean difference between paired data (like before-and-after measurements on the same subjects). You subtract within each pair, then run a one-sample t-test on those differences using the t-distribution with n − 1 degrees of freedom.
A matched pairs t-test checks whether the true mean difference between two related measurements is something other than zero. The data come in pairs that naturally belong together. Maybe it's the same person measured twice (before and after a treatment), or two subjects matched on key characteristics, with one in each condition. Because the two measurements in a pair aren't independent, you can't treat them as two separate samples.
Here's the trick that makes everything simpler: subtract within each pair to get a single list of differences, then run an ordinary one-sample t-test on that list. The null hypothesis is usually H₀: μd = 0 (no average difference), the test statistic is t = (x̄d − 0)/(sd/√n), and you use a t-distribution with n − 1 degrees of freedom, where n is the number of pairs, not the number of individual measurements. Conditions mirror the one-sample t-test, applied to the differences: random selection or random assignment of treatments within pairs, and Normality of the differences (check with n ≥ 30 or a graph of the differences with no strong skew or outliers).
This term lives in Topic 7.10, Skills Focus: Selecting, Implementing, and Communicating Inference Procedures, which is exactly what it sounds like. Unit 7 hands you a toolbox of t-procedures, and Topic 7.10 tests whether you can pick the right one. Choosing matched pairs versus two-sample is one of the most common selection decisions on the exam, and it's decided by the design of the study, not by the data values. If the data are linked in pairs, you analyze the differences. This also ties Unit 7 back to Unit 3, where matched pairs design first appears as an experimental design strategy for reducing variability. The inference procedure has to match the design that produced the data, and graders check for that.
Keep studying AP Statistics Unit 7
Independent Samples / Two-Sample t-Test (Unit 7)
This is the procedure students confuse with matched pairs most often. Two independent samples means two unrelated groups, so you compare two separate means. Paired data means one list of differences, so you test one mean. The study design tells you which one you're in before you ever look at numbers.
Matched Pairs Design (Unit 3)
The experimental design and the inference procedure are two ends of the same idea. If a Unit 3-style experiment pairs subjects or measures each subject twice, the Unit 7 analysis must be a paired t-test. Pairing removes subject-to-subject variability, which makes the test more sensitive to a real treatment effect.
T-Distribution (Unit 7)
A matched pairs t-test is literally a one-sample t-test in disguise, so it uses the t-distribution with degrees of freedom equal to the number of pairs minus one. If 19 patients each get measured twice, df = 18, not 36.
Confidence Interval (Units 6-7)
Every paired t-test has a sibling, the paired t-interval for the mean difference μd. If the interval misses 0, that matches rejecting H₀: μd = 0 at the corresponding significance level. FRQs sometimes ask for the interval instead of the test, and the setup is identical.
Multiple-choice questions love to describe a study and ask which inference procedure fits, or to ask directly for the difference between a matched pairs t-test and a two-sample t-test. The giveaway phrases for pairs are "each subject was measured before and after," "twins," "the same individuals," or "subjects were matched on..." Random assignment of 40 students to two separate groups, by contrast, is a two-sample setup. On the free-response side, the 2023 FRQ Q4 used exactly this structure, with 19 patients in a study comparing an omega-3 supplement to a placebo where pairing mattered. To earn full credit on an inference FRQ you have to do four things: name the correct procedure (paired t-test for a mean difference), define μd and state hypotheses in context, verify conditions using the differences, and then compute the t-statistic and p-value and write a conclusion linked to the p-value in context. Naming the wrong test usually costs you the whole problem, so the selection step is where the points live.
Both compare two sets of measurements, but the link between observations is the whole difference. A matched pairs t-test handles dependent data, where each value in one group is tied to a specific value in the other (same person, matched partner). A two-sample t-test handles two independent groups with no connection between individual observations. Quick check: if you can sensibly subtract row by row to get one difference per subject or pair, it's matched pairs. If subtracting row by row would pair up strangers, it's two-sample.
A matched pairs t-test is just a one-sample t-test performed on the differences within each pair, testing H₀: μd = 0.
Use it when data are dependent, like before-and-after measurements on the same subjects or subjects matched into pairs; use a two-sample t-test when the groups are independent.
Degrees of freedom equal the number of pairs minus one, not the total number of measurements minus two.
Check the conditions on the differences themselves: randomness in the design, plus Normality of the differences (n ≥ 30 pairs or a graph showing no strong skew or outliers).
The study design from Unit 3 determines the test in Unit 7, so read how the data were collected before picking a procedure.
On FRQs, naming the wrong test (two-sample instead of paired) typically forfeits the question, so the identification step matters as much as the calculation.
It's a significance test for the mean of the differences between paired measurements, like before-and-after scores for the same people. You compute one difference per pair, then run a one-sample t-test on those differences with n − 1 degrees of freedom.
A matched pairs t-test is for dependent data where each observation in one set links to a specific observation in the other, so you analyze one list of differences. A two-sample t-test is for two independent, unrelated groups, so you compare two separate sample means. The study design, not the data values, tells you which to use.
Mechanically, yes. Once you take the difference within each pair, you have a single sample of differences, and the test statistic, conditions, and t-distribution work exactly like a one-sample t-test on that list. The only new part is recognizing that the data are paired in the first place.
Look at how the data were collected. Phrases like "before and after," "each subject received both treatments," or "subjects were matched" signal pairs. Random assignment of subjects into two separate groups, like splitting 40 students between a new teaching method and traditional instruction, signals independent samples.
Degrees of freedom equal the number of pairs minus one. So if 19 patients are each measured under both a supplement and a placebo, you have 19 differences and df = 18.