Run every question against a saved set of breaks, with significance letters, and download the whole run as an Excel tab book.
A banner table shows every question in a survey against the same set of columns, side by side, the way a research tab book does. Banner tables are on the Business plan and above.
A banner is a saved set of breaks: the columns your tables share, for example gender, age band and region. Each break is a group of columns, taken from a question's answers (a choice question or a scale) or from a respondent attribute (device, browser or country). Every table starts with a Total column, and the columns of each group are lettered A, B, C… within that group. A survey can have up to 20 banners, and a banner up to 8 groups and 40 columns.
Here is the top of a table on a banner of gender and age band, from the worked example. The first three rows are the header: the groups, their columns, and each column's letter. Then the base, and one answer with its count, its column percentage and its letters:
Total | Gender | Age band
| Male Female Non-bin. | 18–34 35–54 55+
| A B C | A B C
Base 424 | 200 220 4* | 150 170 100
Satisfied 223 | 120 99 4 | 90 85 44
53% | 60% 45% 100% | 60% 50% 44%
| B | C
Male carries B: its 60% is significantly higher than Female's 45%. 18–34 carries C: its 60% is significantly higher than 55+'s 44%. Non-binary is a low base (marked *), so it is never tested.
Open a survey's Crosstabs, choose Banner tables, then New banner:
Every question in the rows is then tabulated against the banner. Banners are built here; with an API key you can also read and run them through the REST API and the MCP server.
Each table is one question. Its base is the respondents in each column who answered the question, and every percentage is calculated on it. Each answer then has three lines: the count, the column percentage, and the letters. A letter names another column in the same group that this figure is significantly higher than. Scales also get a mean, lettered the same way. Use Show to switch between column percentages and counts.
In the worked example, 60% of men and 45% of women were satisfied; on 200 and 220 respondents that difference is significant at 95%, so the men's cell carries the letter B.
Percentages are compared with a two-sided pooled z-test, and means with Welch's t-test, between each pair of columns in a group.
A column with fewer than 5 respondents, or an effective base under 5 when weighted, is marked low base. Its figures are shown for reference and never tested: it carries no letters, and no other column earns a letter over it. In the worked example, the four non-binary respondents are a low-base column.
When the survey has an applied weighting scheme and your plan includes weighting (Business and above), tables are weighted by default; switch Weighted off to see them unweighted. A weighted table shows three bases: Base (unweighted), the people; Base (weighted), the weighted total; and Effective base, what a weighted column is worth as evidence, smaller than the number of people when weights vary. Counts are weighted estimates, not people, and every test uses each column's effective base. A respondent with no weight yet counts as 1, and the page says how many there are.
Letters are tested pairwise, column against column within a group, at the banner's confidence level and without correction for multiple comparisons, the standard tabulation convention. A whole banner makes many comparisons, so across a tab book some letters will be chance: at 95%, about one comparison in twenty between columns that do not really differ still earns a letter. A letter says a difference is unlikely to be chance alone, not that it is large or that it matters, and a missing letter does not show that two columns are the same. Columns are never tested against Total.
Export tab book downloads the run as an Excel workbook, with the filter, weighting and confidence level you are viewing. It has three sheets:
Figures are numbers with Excel formats, so they sum and chart; each percentage has its letters on the line beneath. A low-base column is in italics, with an asterisk on its base and a note. Your organisation can export a limited number of tab books an hour.
Banner tables do not yet cover breaks within breaks, new categories built from answers beyond the default scale bands, ranking and constant-sum questions, number and multiple-answer grids, or open-ended answers (those belong to the Codebook). A question inside a loop can be a row but not a break: every loop iteration counts, so its bases count answers, not people, and its columns are not tested against each other. Columns are not tested against Total, and the NPS score itself carries no letters (its three bands are tested). The tab book is an Excel workbook only.
Was this article helpful?
Where to look first in analytics, and how to go from numbers to decisions.
Four ways to go from a blank page to a working survey in a few minutes.
Test different wordings of the same question and let the survey optimize itself.