t_test

t_test compares the mean of one numeric outcome across two named groups in one tidy table. It returns a statistical object, not a transformed table.

Use it when a mean comparison is the honest question and the data is roughly compatible with a t-test story. If the outcome is ordinal, heavily skewed, or dominated by outliers, consider mann_whitney_u or a custom Python/R node.

Test contract

Field	Required	Notes
inputs	yes	Exactly one tidy table.
valueColumn	yes	Continuous numeric outcome.
groupColumn	yes	Column containing group labels.
groupA, groupB	yes	The two group values to compare.
equalVariance	no	true by default; set false for Welch-style unequal-variance standard errors.

How to read the result

default reports group sizes, group means, mean difference, t statistic, degrees of freedom, p-value, a 95% confidence interval, and effect size.

equalVariance: false uses Welch-style unequal-variance standard errors. The schema default is true, so set it deliberately when variances may differ.

Warnings that matter

• Small groups, non-normal shape, outlier rates, and high variance ratios can all produce warnings.
• TT_VARIANCE_RATIO_HIGH only fires for the equal-variance variant, because that is where unequal variances undercut the assumption.

Example

- id: tt
  kind: t_test
  inputs: [data]
  valueColumn: outcome
  groupColumn: arm
  groupA: control
  groupB: treatment
  equalVariance: false            # default true

Related

• concat - stack two cohorts before testing
• mann_whitney_u - rank-based alternative