Two Measurements, One Unit
A person can provide two related measurements, such as the grip strength of the left and right hands. A single sample can also be measured with two methods. In both situations, each unit contributes a pair of values. The important feature is not that the measurements appear in two columns; it is that each value in one column is linked to the other value from the same person or sample.
In “What Makes Data Paired” and “Defining the Difference Variable,” you learned to identify that link and give the subtraction order explicitly. This tutorial focuses on recognizing and organizing two-measurement data. A dependable first step is to put one unit in each row, keep its measurements together, and then calculate a difference for that row.
Pairing can arise in different ways. A researcher might measure both hands of each person, use two instruments on each physical sample, or record a person’s measurement at two occasions. The setting changes, but the organizing question stays the same: which two observations belong together?
Once the pairs are identified, choose a subtraction order that fits the comparison. For example, if \(d=\text{right}-\text{left}\), a positive difference means the right-hand measurement is larger for that person, and a negative difference means it is smaller. As in “Reducing Paired Data to One-Sample Differences,” there is one difference for each pair. The list of differences—not either original column by itself—records the within-pair comparisons.
Keep Each Pair Together
A paired data table should make the matching visible. Give each row a unit identifier and place both of that unit’s measurements in the same row. If the measurements have been collected in separate lists, use the identifiers to match them before subtracting. Do not match values simply because they occupy the same position after the lists have been sorted or rearranged.
Equal group sizes do not make two groups paired. If one group consists of one set of people and a second group consists of different people, there is no natural row-by-row link unless the study deliberately matched people. Conversely, data can be paired even if the two measurements have very different values or units, provided the design establishes the link and the comparison makes sense.
Worked Examples: Finding and Using the Pairs
Worked Example: Grip Strength in Left and Right Hands
In an invented classroom demonstration, eight students have grip strength measured in both hands, in kilograms. Let \(d=\text{right-hand strength}-\text{left-hand strength}\). Each student is one unit and contributes one pair of measurements.
| Student | Left hand (kg) | Right hand (kg) | \(d=\text{right}-\text{left}\) (kg) |
|---|---|---|---|
| 1 | 21 | 23 | 2 |
| 2 | 25 | 24 | -1 |
| 3 | 24 | 27 | 3 |
| 4 | 28 | 29 | 1 |
| 5 | 22 | 25 | 3 |
| 6 | 26 | 28 | 2 |
| 7 | 23 | 24 | 1 |
| 8 | 27 | 30 | 3 |
Solution. The observations are paired because each left-hand value and right-hand value in a row come from the same student. There are eight pairs, not sixteen pairs: there are sixteen recorded measurements, but only eight students.
For student 2, the difference is \(24-25=-1\) kg. The negative sign tells us that this student’s right-hand measurement is 1 kg less than the left-hand measurement. For student 3, it is \(27-24=3\) kg, so the right-hand measurement is 3 kg greater. Applying the same right-minus-left order to all eight students gives the difference column shown.
The differences total \(2+(-1)+3+1+3+2+1+3=14\) kg, so their mean is \(\bar d=14/8=1.75\) kg. As a check, the left-hand measurements total \(196\) kg and have mean \(196/8=24.5\) kg. The right-hand measurements total \(210\) kg and have mean \(210/8=26.25\) kg. Their mean difference is \(26.25-24.5=1.75\) kg, the same as \(\bar d\). The check works because every right-hand value is paired with exactly one left-hand value and all differences use right minus left.
This average describes the mean of the eight within-student differences in this demonstration. The positive value does not mean every student had greater right-hand strength: student 2’s difference is negative. The individual differences show the variation from person to person, while \(\bar d\) summarizes their average.
Worked Example: Two Methods on the Same Water Samples
A fictional lab compares two measurement methods for the concentration in six water samples. Each physical sample is tested with both methods. Let \(d=\text{Method B}-\text{Method A}\), measured in milligrams per liter.
| Sample | Method A (mg/L) | Method B (mg/L) | \(d=\text{B}-\text{A}\) (mg/L) |
|---|---|---|---|
| 1 | 4.8 | 5.0 | 0.2 |
| 2 | 5.1 | 5.2 | 0.1 |
| 3 | 4.9 | 5.1 | 0.2 |
| 4 | 5.4 | 5.5 | 0.1 |
| 5 | 5.0 | 5.1 | 0.1 |
| 6 | 5.3 | 5.4 | 0.1 |
Solution. These measurements are paired by sample: the two readings in a row are results for the same water sample. The six samples produce six pairs. The pairing is not based on the readings being close; it comes from using both methods on the same sample.
For sample 1, \(d=5.0-4.8=0.2\) mg/L. For sample 4, \(d=5.5-5.4=0.1\) mg/L. The six differences total \(0.2+0.1+0.2+0.1+0.1+0.1=0.8\) mg/L, so \(\bar d=0.8/6\approx0.1333\) mg/L, rounded to four decimal places.
We can check the average with the two column means. Method A’s readings total \(30.5\) mg/L, giving a mean of \(30.5/6\approx5.0833\) mg/L. Method B’s readings total \(31.3\) mg/L, giving a mean of \(31.3/6\approx5.2167\) mg/L. The difference is \(5.2167-5.0833\approx0.1334\) mg/L using the rounded means. Using the unrounded means gives \((31.3-30.5)/6=0.8/6\approx0.1333\) mg/L, matching the mean of the pairwise differences. Small discrepancies from subtracting displayed rounded means are due only to rounding.
The differences describe how Method B’s reading compares with Method A’s reading on each sample. They do not by themselves establish which method is more accurate: that would require a suitable reference or other information about the true concentrations. Here, the purpose of identifying the pairs is to preserve the sample-by-sample method comparison.
Worked Example: Why the Sample Identifier Matters
A fictional technician records results from two instruments on five labeled specimens. Let \(d=\text{Instrument 2}-\text{Instrument 1}\), in standard measurement units. The labels—not the order of values—identify each pair.
| Specimen | Instrument 1 | Instrument 2 | \(d=\text{2}-\text{1}\) |
|---|---|---|---|
| A | 12 | 13 | 1 |
| B | 15 | 14 | -1 |
| C | 11 | 12 | 1 |
| D | 18 | 20 | 2 |
| E | 14 | 15 | 1 |
Solution. Each row is one specimen measured by both instruments, so the observations are paired. The differences are \(1,-1,1,2,1\). Their total is \(4\), and the mean paired difference is \(\bar d=4/5=0.8\) unit. As a check, Instrument 1 totals \(70\), with mean \(70/5=14\), while Instrument 2 totals \(74\), with mean \(74/5=14.8\). The difference between those means is also \(14.8-14=0.8\) unit.
Suppose someone copied the Instrument 2 values into a new list in the order \(20,12,15,13,14\) and then subtracted Instrument 1’s values in the original order. The resulting calculations would be \(20-12=8\), \(12-15=-3\), \(15-11=4\), \(13-18=-5\), and \(14-14=0\). They still total \(4\), so their average is \(0.8\), but those are not the five within-specimen differences. For example, \(20\) belongs to specimen D, not specimen A. The equality of the averages does not repair the mismatching: column totals are unchanged by rearranging, but the individual pair comparisons are wrong.
This example highlights a useful distinction. If the goal is only to subtract the two column means, matching order does not change the column means. But paired analysis uses the actual within-pair differences, so preserving the specimen identifiers is essential. Never sort or shuffle one measurement column independently before constructing the differences.
Common Mistakes and Clear AP Communication
- Counting measurements instead of pairs: Eight people measured twice provide eight pairs, not sixteen. State the number of units or pairs when describing the paired data.
- Calling any two columns paired: Two columns are not enough. Explain the design link, such as “each student supplied both hand measurements” or “both methods were used on each sample.”
- Matching unrelated observations: Equal group sizes do not justify pairing. Each row must connect measurements from the same unit or from a deliberately matched pair.
- Leaving subtraction order unstated: Write a definition such as \(d=\text{right}-\text{left}\). Then interpret positive and negative values using that order and the measurement’s units.
- Reordering measurements without their identifiers: Keep each unit’s values together. The average of the differences may still equal the difference between column means after a shuffle, but the individual differences no longer describe the real pairs.
- Confusing the average difference with every individual difference: A positive \(\bar d\) summarizes the average; it does not say that every \(d_i\) is positive. Check the signs in the difference list before making a statement about individual units.
A clear response identifies what forms a pair, gives the number of pairs, and defines the subtraction order. For example: “Each of the six water samples was measured by both methods, so there are six paired observations. Define \(d=\text{Method B}-\text{Method A}\), in mg/L; a positive \(d\) means Method B gave the higher reading for that sample.” This communicates the design and makes the meaning of every calculated difference clear.
Check Your Understanding
For each question, focus on the unit that connects the two measurements and the meaning of the difference.
- Eight people have both hands measured once. How many pairs are there, and what makes the observations paired?
- Two groups of twelve different students each have a measurement recorded. Are the groups paired merely because they have the same size? Explain.
- For each of five samples, two instruments record a concentration. If \(d=\text{Instrument 2}-\text{Instrument 1}\), what does a negative difference mean in context?
- Why can the difference between two column means stay the same after rearranging one column, even though the within-pair differences become incorrect?
- A paired data table has one row for each person and two measurements in that row. What does the table’s number of rows represent?