Statistics Glossary → Column proportions z-test
Column proportions z-testconcept
A post-hoc method that compares, within each row of a crosstab, the percentages of the column groups pair by pair with a two-proportion z test and marks those that differ with letters.
The chi-square test says there is an association somewhere in the table but not which groups differ. The column proportions z-test compares, for each row category, the proportions of each pair of columns using a standard error based on the combined (pooled) proportion: z = (p₁ - p₂) / √(p̂(1 - p̂)(1/n₁ + 1/n₂)). Results are usually shown with letter notation: columns sharing a letter do not differ significantly in their proportions.
With C columns in a row there are C(C - 1)/2 comparisons, and the p values are Bonferroni-adjusted for that number. The test treats each row as a separate family, so it does not adjust further for testing across rows. With small cell frequencies the z approximation becomes unreliable. Keep in mind that column percentages are being compared: swapping rows and columns answers a different question.
An everyday example
A survey asks three age groups (18-29, 30-49, 50+) which mode of transport they use most. In the 'Bicycle' row the proportions are 18%, 9% and 4%. The Bonferroni-adjusted z-test shows that the youngest group differs from both others, while the difference between 30-49 and 50+ is not significant; in the table the young group gets the letter 'a' and the other two 'b'.
In MerQur
Available as the 'Column proportions z-test' checkbox in the chi-square independence test and as 'Compare column proportions (z test, Bonferroni)' in crosstabs; both give letter notation.
Analyses where this concept appears
Chi-Square IndependenceCrosstab