What a base is
The base is the number of people a figure rests on. In a survey table, it is the count of people who were asked a question and answered it, and every percentage for that question is a share of that number. A result of 40% means something very different on a base of 500 than on a base of 15.
Researchers often write it as n, as in n = 847. This guide covers the kinds of base you will meet, why every table should show one, how to treat small bases, and how to read a crosstab.
Why every question and every crosstab shows its base
A percentage on its own hides how many people it came from. The base puts that back, and it changes from question to question for three reasons:
- Routing. A question asked only of smartwatch owners has a smaller base than one asked of everyone.
- Non-response. Some people skip an optional question, or leave the survey part way through.
- Cuts. As soon as you split results by group, each group has its own base, and some groups are small.
So the rule is simple: show the base on every question, every chart and every column of a crosstab. A reader who can see the base can judge the number. A reader without it is left to trust it.
The kinds of base you will meet
- Total base. Everyone who took part in the survey.
- Filtered base. Everyone who was asked a question, after routing. Write it as a condition, such as "Base: smartwatch owners".
- Answering base. Everyone who actually answered. Many tables use it for percentages; whichever you use, say so in the base line.
- Row base. In a grid, each row can have its own base, because people may skip a row.
- MaxDiff item base. In a MaxDiff scored by counts, an item's base is the number of times it was shown, because an item shown less often had fewer chances to be chosen.
Multiple-response questions need one more note. The base is people, not answers, so the percentages can add up to more than 100%. A good table says so.
Small bases and the under-thirty caution
A common rule of thumb treats a figure that rests on fewer than about thirty people as directional: worth reporting as a hint, not strong enough to act on alone. It is a rule of thumb, not a statistical threshold, and some teams set the line higher. What matters is that you set a line, apply it the same way every time, and label what falls below it.
The reason is that uncertainty shrinks slowly as a base grows. For a simple random sample, a common approximation of the 95% margin of error on a percentage near 50% is 1.96 multiplied by the square root of 0.25 divided by n. Worked through:
| Base (n) | Approximate margin of error |
|---|---|
| 30 | about 18 percentage points either way |
| 100 | about 10 percentage points either way |
| 400 | about 5 percentage points either way |
Quadrupling the base only halves the margin. A research community is not a random sample, so treat these figures as the least uncertainty a result carries, not the whole of it.
Below about ten, a common practice is to stop charting a figure at all, because a single person moves it by ten points or more.
Practical habits for small bases:
- Label them. Put "directional" or "low base" next to the figure, not in a footnote nobody reads.
- Report counts. "4 of 12" is clearer than 33% when the base is that small.
- Combine groups. If two small groups answer in a similar way, ask whether they can be reported together.
- Plan for them. If a subgroup matters to the brief, set a quota for it before fieldwork, not after.
How to read a crosstab
A crosstab shows the answers to one question, in the rows, split by groups, in the columns. Here is an illustrative example, not real data.
Q1. How often do you take part in the community? Base: all members. Column percentages.
| Total | Joined this year | Joined earlier | |
|---|---|---|---|
| Base (n) | 74 | 48 | 26 |
| Daily | 19 (26%) | 10 (21%) | 9 (35%) |
| Weekly | 34 (46%) | 24 (50%) | 10 (38%) |
| Monthly or less | 21 (28%) | 14 (29%) | 7 (27%) |
Read it in this order:
- Read the title and the filter. Know the question and who was asked it before you look at a number (our survey codebook template records who was asked each question).
- Read the base row first. Here the Joined earlier column rests on 26 people, under thirty, so anything in that column is directional.
- Know which way the percentages run. These are column percentages: each column adds down to 100%. Row percentages add across instead. The same counts give different percentages depending on the direction, so check the label.
- Compare across a row. Joined earlier shows 35% taking part daily against 21% for Joined this year.
- Weigh the gap against the bases. That 14-point gap rests on 9 people out of 26. It is worth a follow-up question, not a conclusion.
- Check the total. The Total column is the whole base, and it should agree with the topline for the same question.
Many tables also mark which differences pass a significance test. A test tells you whether a gap is bigger than chance would usually produce in a random sample. It never makes a small base reliable, and it is no substitute for reading the base row.
How Survizo shows the base
Survizo is built on the rule that every question shows its base:
- On screen. Each results card shows its question code and n, the number of people who answered. Each row of a grid has its own base, and any chart resting on fewer than ten answers is marked Low base.
- Routing. Answers to questions a person's path skipped are dropped before the response is sent, so they never count toward a question's base.
- Multiple response. Shares are of people, not answers, and the card tells you the bars can add up past 100%.
- Crosstabs. Choose two choice questions in Explore and you get a crosstab with counts and row percentages, and the Excel workbook adds column percentages, the reading used in the example above. Its Topline sheet gives the percent base beside every share.
- Segments. Compare two saved segments side by side on one question, each with its own n.
- Slides. Each chart slide in the PowerPoint deck shows its base under the chart.
- AI write-ups. When the AI writes up a cut, it marks a finding that rests on fewer than thirty people as directional.
See analysis and exports for the workspace and the files, and question types for how each type reports its base. To see it on your own kind of study, book a fit call.
Frequently asked questions
What is a good minimum base size?
There is no single number. A common rule of thumb treats results resting on fewer than about thirty people as directional, and a common practice is to stop charting below about ten. Set your own line before fieldwork and label anything below it.
Why do my percentages add up to more than 100%?
You are probably looking at a multiple-response question. The base is people, not answers, and each person can pick several options, so the shares can add up to more than 100%.
What is the difference between a base and a sample?
The sample is everyone who took part. The base is the number of people a particular figure rests on, after routing, skipped questions and cuts. One survey has one sample and many bases.
Should I show column or row percentages?
Use column percentages when you want to compare groups on the same answer, which is the usual reading of a crosstab. Use row percentages when you want to see how the people giving one answer split across groups. Always label which you are showing.
Why is a gap between two small groups not a finding?
Because a few people changing their answer would move it. With 26 people in a column, each person is almost four percentage points. Report the gap as directional and test it with a follow-up question or a larger base.