Skip to main content
A crosstab places one variable in rows and compares it across defined columns. Build it from Charts → New chart → Table by adding a breakdown.

Before you start

Define the row measure and comparison groups. Check that eligible responses exist in the groups you intend to test.

Build the table

1

Choose the row variable

Under Question, select the measure whose answers should appear down the table.
2

Select Table

Under Display as, choose Table.
3

Add the first column variable

Under Break down by, choose the segment or question that should form the columns. Column letters and bases appear in the preview.
4

Configure and save

Select the required cells, summary rows, significance settings and weighting, then select Save to charts.
The row variable and every column variable must be distinct.

Add or nest columns

Additional breakdowns create independent column groups, each with its own letters and statistical result. With exactly two column variables, Nest the second question in the first crosses them into combined columns such as region within age. When the row question was shown in several versions, such as one image per cell, Version seen is offered as a column and is preselected. Each column holds the respondents who saw that version. Nesting supports one additional variable. Remove nesting before adding more independent groups.

Choose the values shown in cells

Cells controls the stacked values: Weighted counts can contain decimals because they are weighted sums rather than respondent headcounts. Decimals of % and Decimals of means set how many decimals percentages and means print with, from 0 to 3; the default is 1. They change the display only: the screen, the Excel export, the copied table and what the assistant reads use the same setting, and tests use the unrounded values.

Add base and summary rows

Additional rows can include unweighted and weighted bases. Numeric row variables can also show mean, standard deviation, median, minimum, and maximum. Multi-select rows can show the mean number of answers. Statistical tests use unweighted bases even when the displayed result is weighted.

Read significance markers

The table uses two separate comparisons: A letter names the column with the lower result. For example, B in column A means A is significantly higher than B at the indicated confidence level. Uppercase represents the primary level; lowercase represents the secondary level. With Also mark at 99%, a letter followed by an asterisk, such as A*, marks a difference that also holds at 99%, and the deepest shading marks the same level. Open Legend beneath the table whenever a marker needs interpretation. Each independent column group has its own comparisons.

Configure statistical settings

Heatmap and significance-cell coloring both use the cell background, so enabling significance coloring disables the heatmap.

Check the base before reading a marker

A large percentage difference in a column below 30 is not tested. Widen the column definition, group compatible options, or describe the result as directional.

Group or hide categories

Open the gear beside the row or column variable.
  • Hide removes an option from the table but keeps it available to restore.
  • Group selected combines selected options under a new label.
  • Keep the detail rows (net) adds a grouped net while retaining its component rows.
  • Ungroup returns the options to separate rows or columns.
  • Where groups sit places nets First, After their answers, or Last. After their answers prints each net under the last of its own answers, the usual tab-book order.

Summarize a grid in one table

When the row variable is a whole categorical matrix, Summary table replaces the full answer scale with one line per matrix row and one figure per column. Use it for the recap tables of an image, agreement or brand-association grid. Add the mean of all rows appends the average of the row figures; it is not tested. Letters and shading on each line come from the same tests as the full table. When the columns are the concepts of a concept loop, a respondent can count in several columns, so the summary carries no significance markers. Lead with MaxDiff adds a first column with each row’s preference share from a MaxDiff question on the same items, and sorts the rows by it. Rows are matched to MaxDiff items by the piped item, otherwise by the same wording ignoring case and spacing; an unmatched row shows —. The lead is the mean preference share of the table’s respondents, each counted once, is never tested, and is empty when none of them answered it. See Choose a chart type for how the share is computed. A mean needs scores. If the scale has neither numeric values nor numbered labels, set each option’s Numeric value in the builder, or create a numeric variable with the scores you need.

Generate several crosstabs

Use Bulk generate, choose the crosstab or table output, select the columns and row variables, then choose shared filters, weighting, and destination folder. The selected column order determines the group order in each table. The dialog reports the number of tables before generation. Open-text and unsupported rows are omitted.

Troubleshooting

Expand Legend and read the note below the table. The table may have no eligible bases, too few rows or columns, or completed tests with no result at the selected levels.
Check its unweighted base. A base below 30 is shown with ** and is not tested.
The questionnaire changed after the chart was saved. Reopen the chart, replace the missing variable, and save the current specification.
PowerPoint includes a crosstab as an editable table when the crosstab is on a slide of the saved deck you download. A long table continues across several slides, so check the continuation slides. Use the Charts Excel export when you need the complete grid and its statistical lines.

Check the result

Reopen the saved table, read its legend, and compare raw bases with the requested groups. In the Excel export, check the numeric cells, separate significance columns, and statistical lines. The export and Copy data to clipboard follow the table as shown, including a summary table and its lead column. To prepare tables with per-question comparisons, use the analysis plan. Its table settings are saved separately: a new plan uses 95% primary confidence with no secondary level. Selected tables are generated by default. Only when Only keep tables with a significant difference is enabled does the plan remove comparisons without a qualifying significant cell on a raw column base of at least 30. Review that option, question eligibility and selected comparisons before diagnosing a missing table. An untested or unmarked difference is not evidence that groups are identical.

Read numeric grids and summary rows

A numeric grid is tabulated cell by cell in numeric bands. Add range nets and per-row statistics where the analysis needs them; reversed bounds and empty ranges must be corrected before saving. A scored categorical scale can have mean, median, standard deviation and minimum/maximum rows, but no sum of its scores. Multiple-select tables can show the mean or median number of answers. The display puts percentages above counts. Each matrix row has its own base and tests, so inspect a row’s denominator rather than assuming the whole grid has one base. See Chart metrics and statistics. When a table counts several concept answers from one person, it shows answer and respondent bases separately. Significance and confidence intervals are not computed for this repeated-record table. A looped column question read against an ordinary row question includes respondents with one concept answer; people with several are excluded from that ambiguous column assignment.