Skip to contents

Guide: Decision Rules

Source: vignettes/guide-decision-rules.Rmd
guide-decision-rules.Rmd

This package follows a deliberately small decision tree.

Step 1: is this a simple group comparison?

If goal is anything other than "difference", the package redirects:

  • "association" -> correlation or regression
  • "adjusted_effect" -> regression or mixed models
  • "time_to_event" -> Kaplan-Meier or Cox workflows
  • "agreement" -> ICC, Bland-Altman, or kappa workflows
  • "equivalence" -> TOST or non-inferiority workflows

Step 2: should the simple branch stop?

Even inside goal = "difference", the package redirects or stops when:

  • adjust = "yes"
  • outcome_type = "count"
  • paired data are requested without id
  • repeated data are requested without id
  • fewer than two group levels are available

Step 3: choose a supported branch

Continuous outcome

  • independent 2 groups -> Welch t-test or Mann-Whitney U test
  • paired 2 groups -> paired t-test or Wilcoxon signed-rank test
  • independent 3+ groups -> Welch ANOVA or Kruskal-Wallis test
  • repeated 3+ groups -> repeated-measures ANOVA or Friedman test

Categorical outcome

  • independent groups -> chi-squared or Fisher exact test
  • paired binary -> McNemar test

Ordinal outcome

  • independent 2 groups -> Mann-Whitney U test
  • paired 2 groups -> Wilcoxon signed-rank test
  • independent 3+ groups -> Kruskal-Wallis test
  • repeated 3+ groups -> Friedman test

Normality handling

When normality = "auto" and the outcome is continuous, the package runs a small Shapiro-based check.

  • "yes" -> parametric branch
  • "no" or "unknown" -> rank-based branch

This is intentionally conservative.