Read the Design, Not Just the Lists
The previous tutorial, “Paired t Versus Two-Sample t,” showed that the link between observations determines whether a mean comparison is paired or independent. In a word problem, that link may be stated directly—or hidden in a few important details. Your job is to notice who or what each measurement belongs to before choosing a procedure.
Some clues are direct: “the same students were measured twice,” “before and after,” or “each person completed both tasks.” Others describe deliberate matching: “twins were paired,” or “each new device was matched with a standard device from the same production batch.” In both cases, one observation has a meaningful partner because of the study design.
A word problem can present paired data in two columns, two lists, or a table. None of those formats determines the design. Instead, ask: Can I identify which two measurements belong together, and does that connection come from the way the study was conducted? If so, the data may be paired. If the only connection is that the lists happen to be the same length, it is not enough.
Language Clues That Signal Pairing
Look for wording that explains how observations are connected. The following phrases often signal a paired design:
- Repeated measurements: “the same participants,” “each person was measured twice,” “before and after,” or “at the start and end.” One unit contributes both values.
- Matched units: “matched pairs,” “twins,” “siblings,” or “paired by age and starting score.” The wording should explain how the partners were chosen.
- One-to-one assignment within a pair: “within each pair, one unit received A and the other received B.” This is a common matched-experiment structure.
- Repeated conditions for each unit: “each participant tried both settings” or “every machine was tested with both materials.” Check whether the same unit supplies both results.
These phrases are clues, not substitutes for understanding the design. For example, “similar participants” does not necessarily mean they were matched. The problem should tell you that each participant in one group was deliberately linked to a particular participant in the other group. Likewise, the word “pair” matters only if the pairing identifies which observations go together.
A useful test is to cover up the numbers and read the description of the data collection. Could you still say which two observations form each pair? If the answer is yes, the link is likely genuine. If the pairing becomes impossible when you ignore the row order or equal group sizes, the design may not be paired.
From a Pairing Clue to a Parameter
When genuine pairs are present, define the difference in words and keep the subtraction order consistent. For a before-and-after study, one possible definition is \(d=\text{before}-\text{after}\). For a comparison of two treatments, it might be \(d=\text{Treatment A}-\text{Treatment B}\). The sign then has a clear meaning in context.
The paired-data target is \(\mu_d\), the true mean of the pairwise differences for the population of interest. The paired t procedure analyzes the sample of differences, rather than treating the two columns as independent samples. This is the same design-and-parameter distinction introduced in “Identifying the Parameter in a Mean Problem.”
The order is a choice, but it must match the question. If positive differences mean scores improved, say so. Reversing the subtraction order reverses the sign and the interpretation. It does not change which observations are paired.
Recognizing pairs is only the first step, not an automatic approval to conduct inference. As covered in “Paired t Versus Two-Sample t,” paired t inference requires an appropriate random sample or randomized experiment, independent differences from distinct pairs, and a distribution of differences reasonably compatible with a t procedure. For sampling without replacement, assess the 10% condition. When there are few pairs, inspect the differences for strong skewness or outliers.
Worked Example: The Same Cyclists Before and After
Worked Example: A Repeated Measurement on Cyclists
A sports-science class randomly selects 12 cyclists from a club roster. Each cyclist completes the same short course twice: once before a training program and once after it. The response is completion time in minutes. The data table has one before value and one after value for each cyclist.
Identify the design: These are paired observations because each cyclist contributes both a before and an after time. The word “after” alone is not the only clue; the decisive detail is that the two times belong to the same cyclist. There are 12 pairs, not two independent samples of 12 cyclists.
Define the difference and target: Let \(d=\text{before time}-\text{after time}\), in minutes. A positive difference means the cyclist took less time after the program. Let \(\mu_d\) be the true mean before-minus-after time difference for the population represented by the random sample.
Choose the procedure and assess the design information: A paired t procedure is appropriate for inference about \(\mu_d\), provided its conditions are met. The random selection supports inference to the club population if the club roster is the population of interest. If the roster contains at least 120 cyclists, the 10% condition holds because 12 is no more than 10% of 120. The differences from distinct cyclists should be independent. Because the sample has only 12 pairs, the distribution of the 12 differences should be checked for strong skewness or outliers before relying on t inference.
Explain what would be wrong with two-sample t: The before and after values are not from separate groups of cyclists. Comparing them as independent samples would discard the within-cyclist links. No test statistic or conclusion can be determined from the design description alone; the actual differences and their distribution would be needed for the full analysis.
Worked Example: Matching Twin Pairs
Worked Example: A Study App and Memory Scores
A researcher recruits 15 pairs of twins. Within each pair, one twin is randomly assigned to use a study app and the other to use a usual study routine. After one week, both twins take the same memory assessment. The response is the number of items answered correctly.
Identify the design: The observations are paired because the researcher deliberately formed twin pairs before comparing the two routines. Each app score is linked to the usual-routine score from that same twin pair. The fact that the people are twins alone would not make every possible comparison paired; here, the study explicitly uses the twins as matched pairs.
Define the difference and target: For each pair, define \(d=\text{app score}-\text{usual-routine score}\), in correctly answered items. A positive \(d\) means the twin using the app scored higher. Let \(\mu_d\) be the true mean score difference for the population of twin pairs represented by the recruitment process.
Choose the procedure and assess the design information: If the researcher wants to compare mean scores, the paired t procedure analyzes the 15 pairwise differences. The within-pair random assignment supports a comparison of the two routines, but it does not by itself establish that the recruited pairs are a random sample of all twins. For inference to a population of twin pairs, the sampling method must support that generalization. The differences across distinct pairs should be independent; if pairs were sampled without replacement, the 10% condition should be checked. With 15 pairs, inspect the differences for strong skewness or outliers.
Keep the pairing unit clear: The two twins within one pair are deliberately linked, so the relevant independence question concerns differences from separate pairs—not whether the two twins in a pair are unrelated observations. The design supplies 15 differences for the paired analysis, not 30 independent scores for a two-sample procedure.
Worked Example: Two Equal Groups That Are Not Paired
Worked Example: Battery-Life Comparisons
A quality-control team randomly selects 16 devices made with a new battery design and 16 different devices made with a standard design. No device is tested under both designs, and devices are not deliberately matched. The team records battery life in hours. The new-design sample has mean 9.8 hours and standard deviation 1.2 hours; the standard-design sample has mean 9.1 hours and standard deviation 1.0 hour.
Identify the design: The groups are not paired. They contain different devices, and the description provides no one-to-one matching rule. Having 16 devices in each group does not make the first new-design device a partner to the first standard-design device.
Identify the target and procedure: Let \(\mu_N\) and \(\mu_S\) be the true mean battery life, in hours, for the populations represented by the new-design and standard-design samples. The target is \(\mu_N-\mu_S\), so a two-sample t procedure—not paired t—is the relevant mean comparison if the conditions are met. The observed difference in sample means is \(9.8-9.1=0.7\) hour; this sample difference does not create pairs.
Assess the design information: The random selection of devices supports inference to the represented device populations. Each group should be independent of the other, and observations within each group should be independent. If each sample was selected without replacement from a population of at least 160 devices, the 10% condition holds for each sample. With 16 observations per group, inspect each group’s distribution for strong skewness or outliers. The problem description gives no graphs or raw values, so those distribution checks cannot be completed from the summary statistics alone.
Spot the tempting mistake: Pairing devices by their order in a spreadsheet would be arbitrary. A genuine match would require a design-based reason—such as deliberately pairing devices from the same production batch—not simply placing two unrelated values on the same row.
A Quick Routine for Word Problems
Use this sequence before choosing a mean-inference procedure:
Name the quantitative measurement and its units.
Look for the same unit measured twice or an explicit matching rule.
Ask whether each observation has a design-based partner, rather than a partner assigned by list order or equal sample size.
State what is subtracted from what and explain the meaning of a positive difference.
For genuine pairs, the target is \(\mu_d\) and the mean procedure is paired t. For two unlinked groups, the target is a difference between population means and the procedure is two-sample t.
For paired t inference, assess the random process, independence across pairs, the 10% condition when relevant, and the distribution of the differences.
Common Mistakes and AP Exam Tips
- Pairing observations just because two lists have equal length: Equal group sizes do not identify which observations belong together. Full-credit reasoning names a real link, such as the same participant measured twice or an explicit matching rule.
- Assuming “similar” means “matched”: Similar ages or backgrounds do not automatically create pairs. Explain whether the study deliberately assigned each unit a particular partner.
- Ignoring who was measured: Phrases such as “each participant completed both versions” are strong clues even if the word “paired” never appears. State that the same units contribute both measurements.
- Choosing an arbitrary partner: Do not subtract the first value in one list from the first value in another unless the design establishes that those observations form a pair.
- Leaving the difference order unstated: Write a definition such as \(d=\text{before}-\text{after}\), then connect its sign to the context. The parameter \(\mu_d\) depends on that defined order.
- Confusing paired data with independent measurements: In paired t inference, the analysis uses one difference per pair. The independence condition concerns differences from distinct pairs.
- Claiming a procedure is automatically valid once pairs are spotted: Pairing identifies the structure and candidate procedure, but the random process, independence, 10% condition when applicable, and distribution of differences still matter.
A concise AP response might say: “The same participants were measured under both conditions, so the observations are paired. Define \(d=\text{condition A}-\text{condition B}\); the target is \(\mu_d\), and paired t inference is appropriate if the conditions for the differences are met.” For an unmatched comparison, identify the two separate groups and state that no design-based pairing is described.
Check Your Understanding
For each situation, decide whether the observations are paired. If they are paired, state a suitable difference order and the parameter.
- A school records typing speed for each of 20 randomly selected students before and after a practice schedule. What design clue identifies the pairs?
- A researcher compares 12 randomly selected garden plots using Compost A with 12 different plots using Compost B. The plots were not matched. Does having 12 plots in each group create pairs?
- A study matches 10 pairs of employees with similar job roles; within each pair, one employee uses a new scheduling tool and the other uses the usual schedule. What would one paired difference represent?
- In a paired study, a student defines \(d=\text{after}-\text{before}\). If the question is whether the measurement decreased on average, should the alternative for \(\mu_d\) be positive or negative?
- For paired t inference, should the analyst expect the two measurements within a pair to be independent, or assess independence of the differences from separate pairs?