Introduction
A data table can look simple: rows run across the page, columns run down it, and each cell contains a value. But interpreting the table depends on answering a more basic question: what does each row describe? One row might describe a student, while another table’s row describes a visit by a patient. Mistaking one for the other can lead to incorrect statements about how many people were studied and what the data can tell us.
This tutorial focuses on reading the structure of a dataset. The key is to distinguish the observational unit—the person, object, occasion, or other entity described by a record—from the measurements recorded about that unit. A column usually holds values for one characteristic, while a row brings together the recorded values for one case at the table’s chosen level.
Reading Rows and Columns
For example, if a table describes individual students, a row may contain one student’s recorded age, height, and test score. The student is the observational unit; age, height, and score are characteristics measured or recorded about that student. Each number in a cell is one recorded value, not a whole variable or a whole observational unit.
The word “case” is useful because the row’s level can vary. In one table, each case may be a person. In another, it may be a patient visit, a school, a classroom, or an object inspected on a production line. The table’s headings and the way records were collected help reveal that level; the visual appearance of rows and columns alone does not.
Worked Example: Students Measured Once
Suppose a hypothetical instructor records the age in years, height in centimetres, and quiz score out of 20 for four students, each measured once.
| Student | Age (years) | Height (cm) | Quiz score (points) |
|---|---|---|---|
| A | 19 | 165 | 16 |
| B | 20 | 172 | 18 |
| C | 18 | 160 | 14 |
| D | 21 | 178 | 19 |
Each row describes one student, so the observational unit is the student. There are four rows and four students. The measurement columns are age, height, and quiz score. For student B, for instance, the entry 172 cm is a measurement of height; it is not a separate observational unit.
The table contains \(4 \times 3 = 12\) values in the three measurement columns: four students each have three recorded measurements. The multiplication counts measurement entries, not observational units. The number of observational units remains four.
The Row Level Can Change
A row does not have to represent an entire person. If a patient is measured at several visits, a table may contain one row for each patient-visit combination. In that case, rows represent visits for particular patients, and several rows can refer to the same patient. The underlying person and the record in the table are related, but they are not interchangeable.
It is helpful to ask two separate questions: What does this row represent? and What entity is the question about? For a patient-visit table, the row represents a visit record. A question about a patient’s change across visits is about the patient’s measurements over time, not about several distinct people merely because there are several rows.
Worked Example: Patients Measured Repeatedly
Consider a hypothetical clinic dataset that records a patient’s blood pressure at three scheduled visits. The table below shows two patients, each measured at every visit.
| Patient | Visit | Blood pressure (mmHg) |
|---|---|---|
| P1 | 1 | 128 |
| P1 | 2 | 124 |
| P1 | 3 | 121 |
| P2 | 1 | 136 |
| P2 | 2 | 132 |
| P2 | 3 | 130 |
Each row represents one visit record, with the patient, visit number, and blood-pressure measurement recorded together. There are six rows but only two patients. The blood pressure is measured at visits, and each patient contributes three recorded values. A row count of six therefore does not mean that six different patients were observed.
The arithmetic makes the distinction explicit: \(2\) patients each contribute \(3\) visits, so \(2 \times 3 = 6\) visit records. The six records contain measurements from two people, not six. If the question concerns the number of visits recorded, the answer is six; if it concerns the number of patients, the answer is two.
The row level also clarifies what can be described directly. A row gives one blood-pressure value at one visit. To describe a patient’s change, the values from that patient’s different visit rows must be considered together and in visit order. The table structure tells us which rows belong to the same person and which occasion each measurement came from.
Why the Question Matters
The observational unit should be identified in relation to the question, not guessed from the variable names. Suppose a hypothetical dataset records the average waiting time and number of appointments for each of five clinics. Its rows describe clinics, even though the appointments were made by individual patients. A question about differences among clinics concerns clinic-level records. The fact that patients contributed to the clinic totals does not make each row a patient.
Likewise, a table of repeated measurements can be described at more than one level without its recorded rows changing. Each row may represent a measurement occasion, while the question may concern the people whose measurements were collected. Naming both levels prevents a common ambiguity: the unit represented by a record is not always the entity being compared or discussed.
A Short Justification: Why Rows Need Not Count People
Suppose each of \(N\) people contributes exactly \(k\) measurement occasions, and the table stores one row per occasion. Each person contributes \(k\) rows. Adding those contributions across all \(N\) people gives the total row count \(R\):
For example, with \(N=4\) patients and \(k=3\) visits per patient, the row count is \(R=4 \times 3=12\). Twelve rows represent four patients measured repeatedly, not twelve patients. This conclusion depends on the stated structure—one row per occasion and three occasions for each patient. If the number of occasions varies across patients, the row count is the sum of their separate occasion counts, and it still does not by itself give the number of distinct patients.
This is a counting argument, not a statistical model. Its practical point is that a row count answers “How many records are in this table?” Only after the row level is understood can we determine whether that count also answers “How many observational units or people were studied?”
Common Mistakes
- Assuming one row always means one person. That is true in some tables, such as the student example, but not in a table with one row per patient visit.
- Calling a measurement the observational unit. A blood-pressure value is a recorded measurement; the visit record is the row, and the patient is the person measured.
- Using the number of rows as the number of people without checking. Repeated records can make the row count larger than the number of people represented.
- Assuming the people who contributed data define the row level. If records summarize groups, such as clinics, each row may describe a clinic rather than an individual contributor.
- Giving an answer without naming the level. State precisely, for example, “each row represents one patient visit,” rather than saying only “the rows are the observations.”
The correction is to inspect what one complete row describes, then identify what each column records. If the data include repeated occasions, ask whether multiple rows can belong to the same person. Finally, relate that structure to the question being asked; row count, person count, and measurement count may all be different.
Worked Example: Records About Classrooms
A hypothetical school report lists one row per classroom. For each of six classrooms, it records the number of enrolled students, the classroom’s average attendance percentage for a month, and the number of absences recorded during that month. Identify the row level and distinguish the records from the people contributing to the measurements.
Each row describes one classroom, so the observational unit represented by the table’s rows is the classroom. There are six rows and six classroom records. The columns contain measurements or summaries about each classroom: enrollment count, average attendance percentage, and absence count.
The students are the people whose attendance contributes to the classroom summaries, but the table does not contain one row per student. Therefore, the table’s six rows do not mean that six students were observed. Nor does the table by itself give the total number of students; the enrollment values would need to be added to answer that separate question.
Check Your Understanding
For each situation, identify what one row represents and distinguish it from the measurements or entities involved.
- A table has one row for each of 30 students, with columns for age and exam score. What is the observational unit, and what do the rows represent?
- A hypothetical dataset contains 5 patients measured at 4 visits each, with one row per visit. How many rows are there, and how many patients are represented? Show the calculation.
- A report has one row per hospital and gives the hospital’s average length of stay. Does a row represent a patient or a hospital? Explain.
- In a repeated-measurement table, identify the difference between a patient, a visit record, and the blood-pressure value in one cell.
- Why is it unsafe to report the number of rows as the number of people studied before checking what each row represents?