chi_square

chi_square builds a contingency table from two categorical columns and tests whether the observed counts depart from independence.

Use it when the question is about association between categories: site by outcome band, treatment by response class, product by region.

Count-test contract

Field	Required	Notes
inputs	yes	Exactly one table.
columnA	yes	First categorical variable.
columnB	yes	Second categorical variable.

How to read the result

default includes the tested columns, n, degrees of freedom, chi-square statistic, p-value, and effect size.

A significant result means the variables are not independent. It does not identify which cells are responsible, so inspect the contingency table or residuals in a follow-up node if needed.

Expected counts

• CHI_SQUARE_EXPECTED_CELL_TOO_LOW is critical when any expected cell count is below 1.
• CHI_SQUARE_EXPECTED_CELL_LOW_FREQUENCY warns when more than 20% of expected cells are below 5.
• For small 2x2 cases where approximation quality matters, use Fisher exact in Python/R.

Example

- id: cs
  kind: chi_square
  inputs: [data]
  columnA: site
  columnB: age_group

Related

• Reports - warning callouts in generated reports