Guide to cross tabulation
Cross tabulation: spot relationships between two questions
A cross tabulation (crosstab) counts how the answers to one question spread across the groups of a second – satisfaction by customer type, for example. Row or column percentages, the chi-square test and Cramér's V show whether a difference is more than chance and how strong it is.
Building and testing is free. To publish, you choose a paid plan.
How a crosstab is built
Rows, columns, cells and marginal totals
The categories of one question form the rows, those of the other the columns. Each cell counts the people who chose both categories – say “new customer” and “very satisfied”. The margins hold the total for each row and each column, with the grand total n in the bottom right corner. Marginal totals tell you how large each group is and are the base of every percentage.
Row percentages
Row percentages divide each cell by its row total, so every row adds up to 100 percent. They tell you how new customers spread across the satisfaction levels, and how regular customers do. When the groups sit in the rows, row percentages compare them directly.
Column percentages
Column percentages divide each cell by its column total, so every column adds up to 100 percent. They answer the reverse question: who makes up the very satisfied? Both readings are correct but answer different questions, so decide beforehand which one you are asking.
Total percentages
Total percentages divide each cell by all respondents, so the whole table adds up to 100 percent. That shows how common a combination is overall – useful when you want to size a segment. For comparing groups they are of little use, because large groups automatically get larger shares.
Which question goes in the rows, which in the columns?
Rule of thumb: the explanatory (independent) variable – age, customer type, channel – forms the groups; the outcome (dependent) variable – satisfaction, purchase intent – forms the distribution. Percentage in the direction of the independent variable, within each group, and compare across. If the groups are in the columns, go for column percentages.
Chi-square test and Cramér's V
The chi-square test
The test compares each observed cell count O with the count E you would expect if the two questions were unrelated: E = row total × column total / n. Summed over all cells this gives χ² = Σ (O − E)² / E. The further observation and expectation drift apart, the larger χ² becomes.
p-value and degrees of freedom
How large χ² may get depends on the size of the table, expressed as degrees of freedom: df = (rows − 1) × (columns − 1). The p-value tells you how likely a χ² at least this large would be if there were no relationship in the population. If it falls below the threshold you chose in advance, usually 0.05, the relationship counts as statistically significant.
Cramér's V: how strong is the relationship?
χ² grows with sample size, so it says little about strength. Cramér's V standardises it: V = √(χ² / (n · (min(r, c) − 1))), where r and c are the numbers of rows and columns. V ranges from 0 (no relationship) to 1 (perfect relationship). In large samples a small V is often significant yet of little practical meaning.
How many responses
For the significance test to hold, almost all cells should have an expected count of at least 5 responses. If there are many answer options, combine categories.
Common pitfalls
Cells that are too small
The chi-square test is an approximation and becomes unreliable when expected counts are small. Merge rare answer categories, such as “somewhat dissatisfied” and “very dissatisfied”, or drop empty ones. For very small two-by-two tables, Fisher's exact test is the usual alternative, for example in SPSS or R.
Many tables, many false alarms
Test twenty crosstabs at the 0.05 threshold and on average one will come out significant even if no real relationship exists anywhere. Decide in advance which relationships you want to test, and with many tests tighten the threshold, for instance with the Bonferroni correction: threshold divided by the number of tests.
Multiple choice breaks the assumption
The test assumes every person sits in exactly one cell. With a multiple-choice question a person can tick several options and then appears in several cells; the observations are no longer independent, and χ² would mislead. You can still compare the counts, just without a significance test.
Significant does not mean important or causal
A small p-value only says that a relationship is probably not down to chance – not how large it is, and not what drives what. A third variable may sit behind it. If you weight your data, also check that percentages and test rest on the same numbers.
Crosstabs in 121surveys
Cross questions and link variables
You pick two dimensions: a question from the survey or a link variable that arrives with the participation link – such as channel, source or branch. That lets you compare answers by characteristics you never asked about.
Test and effect size built in
For every table 121surveys calculates chi-square, degrees of freedom, p-value and Cramér's V, each with an explanation. If one axis is a multiple-choice question, the test is left out and a note tells you why.
Clear and anonymous
Cells show the counts on a colour scale – the darker, the more people – plus the totals for each row and column. Groups below your survey's anonymity threshold are hidden, so nobody becomes identifiable through a small cell.
Filters, export and AI assistants
The filters and date range of your analysis also apply to the crosstab. Export the raw data as CSV, Excel, SPSS, ODS or JSON, and an AI assistant can fetch the same table through the MCP tool get_crosstab.
Frequently asked questions
What is a cross tabulation?
A table that counts two characteristics at once: the categories of one in the rows, those of the other in the columns, and each cell gives the number of people with both. It is also called a contingency table or crosstab.
When do I use row and when column percentages?
Percentage within the groups you want to compare. If the groups are in the rows, take row percentages; if they are in the columns, column percentages. You only need total percentages when you care about a combination's share of all respondents.
Does 121surveys show row and column percentages?
The crosstab in the analysis shows the count for each cell together with the row and column totals. Work out row and column percentages straight from these – cell divided by row or column total – or after exporting to Excel or SPSS.
When is a relationship significant?
A p-value below 0.05 is the usual convention. It means: if there were no relationship in the population, a deviation this clear would turn up in fewer than five out of a hundred samples. Set the threshold before the analysis, not afterwards. Whether the relationship matters in practice is what Cramér's V tells you.
Can I cross more than two questions?
A crosstab has two dimensions. For a third, filter the analysis to one group – new customers only, say – and compare the tables of the groups. For models with many variables at once, such as a logistic regression, export the data to SPSS or R.
Crosstab and test in one step
Create your survey in 121surveys, pick two questions or link variables in the analysis and read counts, chi-square, p-value and Cramér's V at a glance.
