Someone hands you a page of survey results: rows of answers, columns of groups, percentages and small letters next to some numbers. You are asked what it means for the launch plan, and you have five minutes. This guide shows how to read a crosstab the way an experienced researcher does, in a fixed order, so you reach the right conclusion and avoid the usual traps. If you want the vocabulary first (banner, stub, base, net), start with our guide to what a crosstab is and how it is built.
Key takeaways
- Read the question wording and the base before you read any percentage.
- In a column-percentage table, each column adds to 100% going down; compare the groups by reading across a row.
- A letter next to a number means it is significantly higher than the column with that letter. No letter, no reliable difference.
- Flag bases under 50 and treat bases under 30 with caution. This is a common working rule, not a law.
- A crosstab shows that groups differ. It does not tell you why.
Why is a crosstab hard to read at first?
Three things trip people up. Percentages can run down the columns or across the rows, and the same counts give different numbers each way. Every column has its own base, and a striking figure on 40 people looks exactly like one on 400. And the letters next to the numbers are easy to skip, yet they show which differences are bigger than sampling noise.
How do you read a crosstab, step by step?
Here is how to read a crosstab in six steps, in the same order every time.
- Read the title and the question wording. “Would you buy this?” and “Would you buy this at $4.99?” are different questions. Check who was asked, too.
- Find the base. The base row shows how many people each percentage is calculated on. A weighted study often shows two: the unweighted base (real respondents) and the weighted base.
- Check which way the percentages run. If each column adds to 100% going down, they are column percentages; if each row does going across, row percentages. Most market-research crosstabs use column percentages.
- Compare across the columns, not down. With column percentages, the useful comparison is the same answer in different groups: 18–34s against 55+, not “definitely” against “probably” within one group.
- Look at the significance letters or arrows. They mark the differences that passed a statistical test. Our guide to significance testing in crosstabs explains how the letters are calculated.
- Check base sizes before you draw conclusions. On small bases, only large gaps are reliable.
How do you interpret a cross tabulation table?
The quickest way to learn how to interpret a cross tabulation table is to walk through one. In this illustrative banner table, 1,000 respondents rated how likely they are to buy a new product, broken down by gender and three age groups. For how banners are designed, see our guide to banner tables in market research.
| Purchase intent | Total | Men (A) | Women (B) | 18–34 (C) | 35–54 (D) | 55+ (E) |
|---|---|---|---|---|---|---|
| Base: all respondents (unweighted) | 1,000 | 490 | 510 | 300 | 380 | 320 |
| Definitely would buy | 17.1% | 17.3% | 16.9% | 24.0% DE | 16.1% | 11.9% |
| Probably would buy | 29.3% | 28.6% | 30.0% | 32.0% | 30.0% | 25.9% |
| Might or might not buy | 25.4% | 25.3% | 25.5% | 23.0% | 26.3% | 26.6% |
| Probably would not buy | 17.0% | 16.1% | 17.8% | 13.0% | 17.1% | 20.6% C |
| Definitely would not buy | 11.2% | 12.7% | 9.8% | 8.0% | 10.5% | 15.0% C |
| Net: top 2 box (definitely + probably) | 46.4% | 45.9% | 46.9% | 56.0% DE | 46.1% E | 37.8% |
| Mean score (5 = definitely, 1 = definitely not) | 3.24 | 3.22 | 3.26 | 3.51 DE | 3.24 E | 2.99 |
Illustrative data, unweighted. Column percentages: each column’s five answers sum to 100%. A letter means the figure is significantly higher than the column with that letter at 95% confidence (two-proportion test for percentages, t-test for means), tested within gender (A–B) and within age (C–E).
What the table says
Take the 18–34 column. 72 of its 300 people said “definitely”, which is 72 ÷ 300 = 24.0%. Another 96 said “probably” (32.0%), so the top-2-box net is 168 ÷ 300 = 56.0%, the same as 24.0 + 32.0.
Now read across the net row. Overall, 46.4% would definitely or probably buy. Men and women are almost identical (45.9% and 46.9%, no letters). Intent falls with age: 56.0% for 18–34 (DE, higher than both older groups), 46.1% for 35–54 (E, higher than 55+) and 37.8% for 55+. The rejection rows confirm it: 55+ are more likely than 18–34 to say they would not buy.
What the table does not say
It does not say men are more negative than women: the 12.7% versus 9.8% gap on “definitely would not buy” carries no letter. Nor is the 6.1-point gap on “probably” between 18–34 and 55+ a confirmed difference.
It does not say why younger people are keener; age may stand in for income or familiarity with the category. And it does not say most buyers are 18–34, as the next section shows.
What is the difference between column and row percentages?
The same counts can be shown both ways, and each answers a different question. Here is the example’s top-2-box net by age.
Column percentages: how likely is each age group to buy?
| 18–34 | 35–54 | 55+ | Total | |
|---|---|---|---|---|
| Likely to buy (top 2 box) | 168 (56.0%) | 175 (46.1%) | 121 (37.8%) | 464 (46.4%) |
| Not likely to buy | 132 (44.0%) | 205 (53.9%) | 199 (62.2%) | 536 (53.6%) |
| Base | 300 (100%) | 380 (100%) | 320 (100%) | 1,000 (100%) |
Row percentages: where do the likely buyers come from?
| 18–34 | 35–54 | 55+ | Base | |
|---|---|---|---|---|
| Likely to buy (top 2 box) | 168 (36.2%) | 175 (37.7%) | 121 (26.1%) | 464 (100%) |
| Not likely to buy | 132 (24.6%) | 205 (38.2%) | 199 (37.1%) | 536 (100%) |
| All respondents | 300 (30.0%) | 380 (38.0%) | 320 (32.0%) | 1,000 (100%) |
The “not likely” row sums to 99.9% because of rounding.
The column view says 18–34s are the most likely to buy. The row view says 35–54s supply the largest share of likely buyers (37.7%), because there are more of them. Both are true: use column percentages to compare groups, and row percentages to describe who makes up an audience.
How do you interpret crosstabs with nets, means and multi-response questions?
Nets
A net combines answers, such as “definitely” plus “probably” into a top-2-box score. It counts people, so its percentage is the combined count divided by the base. Stakeholders tend to remember the net, so check that it agrees with its parts.
Means
A mean scores each answer (here 5 down to 1) and averages them. It is handy for ranking groups but hides the shape: two groups can share a mean of 3.0 while one is split between love and hate and the other sits in the middle. Read it alongside the distribution.
Multi-response questions
When people can pick several answers, such as “Which of these features would you use?”, a column adds to more than 100%. If respondents choose 1.8 features on average, it sums to about 180%. The base is still respondents, so each figure is the share of people who chose that answer.
These rules are easier to follow when every table shows them. Tools that produce crosstabs and banner tables with weighting and significance testing apply the same bases, nets and tests across hundreds of questions, so you are not checking each one by hand.
How small is too small for a base?
A common working rule is to flag any column with a base under 50 and treat bases under 30 as directional only. It is a practice, not a law, so agree your thresholds before you look at the results.
The reason is the margin of error. For a 50% result at 95% confidence, it is roughly ±3.1 points on 1,000 people, ±5.7 on 300, ±13.9 on 50 and ±17.9 on 30. Bases shrink quickly when you nest banners: split the 300-person 18–34 column by gender and then by four regions, and each cell averages fewer than 40 people.
How do you explain a crosstab to a stakeholder in one sentence?
Lead with the finding, give the numbers, and say whether the difference is real. For the example: “Among 1,000 adults, 46% would definitely or probably buy; that rises to 56% among 18–34s, significantly higher than any older group, and men and women do not differ.” When you are interpreting crosstabs for a busy reader, that sentence plus the base is usually all they need.
What are the common mistakes when reading a crosstab?
- Reading down instead of across. In a column-percentage table, comparing answers within one group answers a different question from comparing groups.
- Ignoring the base. A 70% on 25 people is a hint, not a finding. Look at the base row before the number.
- Treating every difference as real. On a page with dozens of comparisons, some gaps look large by chance. Rely on the letters, and remember that at 95% confidence about one in twenty comparisons with no real difference gets flagged anyway.
- Comparing columns with overlapping groups. Men and 18–34s share people, so the table does not test one against the other. The same goes for any column against Total, which contains it.
What should you do next?
The best way to learn how to read a crosstab is practice. Take one table from your latest study, walk it through the six steps and write the one-sentence summary. If a finding holds up, test it against other variables, as our guide to cross-tabulation analysis explains. When stakeholders have follow-up questions, letting them ask questions in plain language and get answers from your own research, with the source shown saves another round of table requests.