A crosstab shows how the answers to one question split by the answers to another, for example satisfaction by plan. Open it from Surveys, your survey, Analysis.
Build a table
- Pick a row question and a column question. Choice, rating and NPS questions can be used.
- Choose what each cell shows: counts, or a percentage of the row, the column or the whole table.
- Choose how skipped answers are treated. By default a respondent who skipped either question is left out of the table and counted in a separate line, so you always see how many were left out. Include them to show "Skipped" as its own row and column.
- Add filters if you need them: a date range, link or in-app collection, language, identified or anonymous respondents, one invitation wave, and cohorts such as "plan is Pro".
Above the table you see the number of responses in scope, the number in the table (the valid denominator), and how many were left out and why.
If a question lets people pick more than one answer, the table counts respondents, not selections. Rows can add up to more than the number of respondents and percentages can pass 100 percent. The table says so.
Which responses count
The inclusion policy decides which responses are in scope. Spam never counts. The default counts every other response, including ones flagged by response quality review. Pick a stricter policy to leave flagged or excluded responses out, or to see only those. The policy is printed on the page and in the export.
Compare groups
Turn on Compare groups to test whether the column groups differ in how often a row answer is chosen.
- Two groups are compared with a two-sided two-proportion test. When any expected count is below 5, an exact test is used instead, so small tables are never judged with a method that does not fit.
- Every comparison shows the two proportions, the difference with a 95 percent interval, and an effect size.
- Comparisons inside one table form one family. The p-values are adjusted for the family (Holm correction), so a raw p-value under 0.05 is not called significant when many comparisons were made.
- A comparison is withheld, with the reason, when the groups overlap or are paired (the same respondents appear in both), when a group has fewer than 5 respondents, or when there is no variation to test.
Statistical significance does not measure business impact. A tiny difference in a very large sample can be significant and still not matter, and the other way round.
Export
Export CSV downloads exactly the table you see. The file starts with the survey, the table, the inclusion policy, the missing-value policy, every active filter and the denominators, so a number is never separated from the question it answers. Text that could be read as a spreadsheet formula is neutralised.
Saved views
Save view stores the questions, filters and options. A view is private until you share it with the team; sharing needs permission to edit the survey. Anyone who can open the survey can open and export a shared view. Views of other surveys and workspaces are never visible.
API and agents
Everything on this page is available over the REST API and to agents, with the surveys:read and surveys:write abilities:
POST /surveys/{survey}/analysis/crosstabruns a table,.../crosstab/exportreturns the CSV.GET/POST/PATCH/DELETE /surveys/{survey}/analysis/viewsmanages saved views.- Agent tools:
get_survey_analysis_questions,get_survey_crosstab,list_survey_analysis_views,get_survey_analysis_view,create_survey_analysis_view,update_survey_analysis_view,delete_survey_analysis_view.
