Crosstabs and significance

A crosstab puts a question's answers down the side and a group across the top, so you can see how, say, each department answered the same question and which differences are larger than chance would explain. Open it from a survey's Results tab, in the Crosstab view. Available on the Team and Enterprise plans.

Build a crosstab

1. 1Under Columns, choose the group to compare: a respondent detail such as department, location, source, device or language, or the answers to a single-answer choice question.
2. 2Under Rows, choose one or more questions. Choice, multiple choice, Likert, rating, NPS, matrix, slider and date questions can be crossed. Each matrix row gets its own table, and sliders and dates are grouped into bins. Choose Show every question to see them all.
3. 3Choose what cells show: column percentages, counts, or both.

Filters and the date range above the table apply, and the whole view lives in the URL, so a crosstab can be bookmarked or shared. Questions marked sensitive can't be crossed: they show as Hidden: sensitive.

Read a table

• Each column is a group. Percentages are of the people in that group who answered the question, which the Answered row shows. The Total column is everyone in the view.
• A letter beside a figure means it is statistically significant at 95% higher than the column with that letter. This is the column proportions test familiar from SPSS: a two-proportion z-test for every pair of columns, adjusted with the Bonferroni method for the number of pairs.
• An arrow means a column is significantly higher or lower than everyone else who answered. Percentages use a z-test and means use Welch's t-test, adjusted for the number of columns tested.
• Scale questions (Likert, rating, NPS, slider and matrix rows) add a Mean row, and NPS adds the score for each group.
• The line above each table is a chi-square test of whether the answers differ between the groups at all, with Cramér's V for how strongly (0 is no relationship, 1 is a perfect one). Questions where people could choose several answers get column comparisons but no chi-square.

What significant means, and doesn't

Statistically significant at 95% means a difference this large would show up by chance alone less than 1 time in 20. It does not say why groups differ, and a relationship between a group and an answer is never proof that one causes the other. Run many comparisons and a few will come out significant by chance, so treat a single marked cell as a lead to follow up, not a finding.

Tests are left out where they would mislead. A group is tested only when at least 30 people in it answered. A comparison is skipped when a group would expect fewer than 5 people on either side, and chi-square is skipped when more than a fifth of its cells expect fewer than 5 people, or fewer than 30 people answered. The note under each table says how many comparisons were left out.

Anonymous surveys

• A group with fewer than 5 people, or where fewer than 5 answered a question, is hidden. So is the next smallest group whenever a hidden one could be worked out from the Total column.
• Where fewer than 10 people answered a question, a group shows only how many answered, with no counts, percentages or averages, because a single answer in so small a group could be traced to one person. Enough other groups are held back the same way that the Total can't give their answers away.
• Groups that are hidden or show only their size are not tested, and chi-square uses only the groups whose numbers are shown. Crossing by another question follows the same rules.

Export

Open Export and choose Crosstab as Excel or CSV for what's on screen, or Banner table for every question across the same columns. Excel files have one sheet per question and a Significance key sheet with the columns, filters and how to read the marks; CSV has one line per answer and measure. Exports follow your organization's setting for who can export, apply the same anonymity rules as the page, and are recorded in the audit log.