From a Table to a Clear Statement
A two-way table can contain many counts, but a reader needs to know what those counts say about the variables being studied. A useful report names the population or group represented, names both variables, states which group each percentage is calculated within, and summarizes the pattern in the conditional percentages. Without the denominator, a percentage may be numerically correct but answer the wrong question.
For example, “60% used public transit” is incomplete if it is unclear whether 60% refers to all people, people in one neighborhood group, or transit users. A clear statement makes the comparison explicit: “Among adults in the far-distance group, 60% reported using public transit, compared with 30% of adults in the near-distance group.” The denominators are now visible in the wording.
Read the Denominator Before Interpreting
Suppose the rows of a two-way table are categories of variable \(A\), and the columns are categories of variable \(B\). A row conditional percentage in a cell is the cell count divided by its row total. It answers: “Among individuals in this row category, what proportion fall into this column category?” A column conditional percentage instead divides by the column total and answers a different question.
This follows directly from how a proportion is defined. If \(n_{ij}\) is the count in row \(i\), column \(j\), and \(n_{i+}\) is the total count in row \(i\), then each individual in the row belongs to exactly one of the column categories. The share in column \(j\) is therefore the cell count divided by the number of individuals in that row:
The row percentages across all columns add to 100% (apart from rounding), because the column categories together account for everyone in that row. For column conditional percentages, divide each cell by its column total instead; percentages then add to 100% down each column. Choose the denominator that matches the question, not simply the one that is easiest to calculate.
Worked Example: Comparing Transit Use by Distance Group
Consider a hypothetical random sample of 200 adults living in a city. Each adult is classified by residential distance from the city center and whether they usually use public transit for their commute.
| Distance group | Uses transit | Does not use transit | Total |
|---|---|---|---|
| Near | 30 | 70 | 100 |
| Far | 60 | 40 | 100 |
| Total | 90 | 110 | 200 |
To compare transit use between distance groups, use row percentages: the denominator for each rate is the total adults in that distance group. Among adults in the near group, the percentage who use transit is
Among adults in the far group, the corresponding percentage is
A clear report is: “In this hypothetical sample of 200 city adults, 30% of those in the near-distance group usually used public transit for their commute, compared with 60% of those in the far-distance group. Transit use differed between the two distance groups in this sample.” This names the group, both variables, the denominators, and the observed pattern. The table shows an association in the sample; it does not establish that living farther away causes transit use.
The full-sample percentage using transit is \(90/200=45\%\), but that answers a different question. It does not describe transit use separately within either distance group, so it cannot replace the two conditional percentages in this comparison.
Build the Interpretation in Context
A report should usually begin by identifying the population or set of individuals represented by the data. Then name both variables in ordinary language rather than relying only on row and column labels. State the denominator with wording such as “among adults in the far-distance group.” Give the conditional percentages with their comparison groups, and finish by describing the pattern as an observed association or difference in this group.
The word “population” should match what the data actually represent. If the table summarizes every member of a defined group, name that group. If the data come from a sample, identify the sample and be cautious about broader claims. A random sample may support generalization to its target population when the sampling and response conditions are suitable, but a table alone does not guarantee that those conditions hold.
Use the observed percentages to describe the direction and size of the difference. For two groups, one concise comparison is the difference between their conditional proportions. In the transit example, the observed difference in transit-use percentages is \(60\%-30\%=30\) percentage points. This is a difference in percentage points, not a 30% increase. It is often enough to report the two percentages directly, especially when the audience needs to see the group-specific rates.
Worked Example: Naming the Denominator in an Insurance Comparison
In a hypothetical survey, 160 adults completed questions about employment status and health-insurance coverage. The table summarizes those 160 respondents.
| Employment status | Insured | Uninsured | Total |
|---|---|---|---|
| Employed | 72 | 18 | 90 |
| Unemployed | 42 | 28 | 70 |
| Total | 114 | 46 | 160 |
To compare insurance coverage by employment status, calculate the insured percentage within each employment group. Among employed respondents,
Among unemployed respondents,
A precise report is: “Among the 160 adults who completed this hypothetical survey, 80.0% of employed respondents had health-insurance coverage, compared with 60.0% of unemployed respondents. In this respondent group, coverage was more common among employed than unemployed adults.” The wording makes clear that the denominators are 90 employed and 70 unemployed respondents, not all 160 respondents.
The percentage \(72/114\approx63.2\%\) is also mathematically valid, but it answers a different question: among insured respondents, what percentage were employed? It does not answer what percentage of employed respondents were insured. Confusing these statements reverses the conditioning group.
Report Patterns Without Overclaiming
An association describes a pattern in the variables as they appear together in the data. It does not, on its own, establish that one variable caused a change in the other. A table of survey responses, for instance, can show that insurance coverage percentages differ by employment status, but it cannot by itself rule out other explanations or establish a causal effect.
Likewise, describe the scope of the evidence accurately. “In this sample, the percentage was higher in group A” reports what the table shows. “People in group A are always more likely to have the outcome” extends beyond the observed data. A statement about a broader population needs an appropriate sampling basis; a causal statement needs evidence that supports causal conclusions.
When conditional percentages are close, say so without treating closeness as proof that the variables are independent. For example, “the percentages were similar across the listed groups in this table” is a description of the observed pattern. It does not establish that the variables are independent in a wider population.
Worked Example: Describing a Pattern Across Three Groups
A hypothetical program records class-format preference for 150 students in one cohort. The two variables are year in the program and whether the student prefers online classes.
| Year in program | Prefers online | Does not prefer online | Total |
|---|---|---|---|
| First | 20 | 30 | 50 |
| Second | 33 | 27 | 60 |
| Third | 30 | 10 | 40 |
| Total | 83 | 67 | 150 |
Calculate the percentage preferring online classes within each year group, using that year’s total as the denominator. For first-year students, the calculation is
For second-year students, it is
For third-year students, it is
A concise report is: “Among the 150 students in this hypothetical program cohort, the percentage who preferred online classes increased across the listed year groups: 40.0% of first-year students, 55.0% of second-year students, and 75.0% of third-year students. In this cohort, year in the program and online-class preference showed an association.” This describes the observed ordering while avoiding a claim that progressing through the program causes a change in preference.
A Practical Reporting Sequence
Before writing, check that the proposed sentence answers the reader’s likely questions. A useful sequence is:
Identify the population, cohort, or sample represented by the table, and avoid generalizing beyond it without support.
Translate the table labels into the actual characteristics being compared.
Use wording such as “among each employment group” so the conditional percentage is unambiguous.
Report the relevant conditional percentages and summarize their pattern in context.
Describe an observed association; do not infer causation or claim more generality than the data support.
Common Mistake
A common error is to report a percentage without saying what group forms its denominator. “63.2% of insured respondents were employed” and “80.0% of employed respondents were insured” can both be true for the insurance table, but they answer different conditional questions. The correction is to write the conditioning group directly into the sentence and check that the denominator matches it.
Another overstatement is to turn a difference in a table into a causal conclusion. A report should say that the variables were associated in the observed group, not that one variable produced the other. Keep the population, variables, denominator, and conclusion tied to what the table actually contains.
Check Your Understanding
For each question, focus on what the reported percentage’s denominator means.
- In a table of exercise group and reported improvement, what wording would make clear that improvement percentages are calculated within each exercise group?
- A statement says, “Among respondents with insurance, 63% were employed.” Explain why this does not tell you the percentage of employed respondents who had insurance.
- Write a one-sentence interpretation when a hypothetical sample shows 25% of group A and 40% of group B in a specified category. Include the variables and denominator wording.
- Why is “the two variables were associated in this sample” more cautious than “one variable caused the other”?
- If conditional percentages are similar across groups, what can you report, and what would be too strong to conclude from the table alone?