Skip to contents

statsguider

R-CMD-check pkgdown License: MIT

statsguider is a small R package for guided statistical test selection. It is intentionally narrow: the package is designed for users who already have a data.frame, want to describe the data structure through a small set of choice-style arguments, and need either:

  • a recommendation for a simple statistical test, or
  • immediate execution of that test with base R.

The package does not try to cover all of statistics. It deliberately stops or redirects the analysis when the question belongs to a different branch such as adjusted modelling, survival analysis, agreement analysis, equivalence, or count regression.

Design Goals

The repository has been rebuilt around four constraints:

  1. Minimal package footprint Only the package source, tests, documentation source, and GitHub workflows are tracked. Rebuildable outputs such as .Rcheck, tarballs, temporary folders, and generated docs/ files are not kept in the repository.
  2. Minimal argument contract The required identifiers are data, outcome, group, and id when needed. Everything else is a choice-style argument such as paired = "yes" or goal = "difference".
  3. Minimal execution scope The package only executes a small set of base-R tests that match a simple branching workflow.
  4. Detailed documentation without heavy assets Tutorials build their own small data tables in code. No large bundled data and no downloaded external datasets are used.

What the Package Covers

Supported executable branches

  • Independent two-group continuous comparison
    • Welch t-test
    • Mann-Whitney U test
  • Paired two-group continuous comparison
    • Paired t-test
    • Wilcoxon signed-rank test
  • Independent three-or-more-group continuous comparison
    • Welch ANOVA
    • Kruskal-Wallis test
  • Repeated continuous comparison across three or more conditions
    • Repeated-measures ANOVA
    • Friedman test
  • Independent categorical comparison
    • Chi-squared test
    • Fisher exact test
  • Paired binary comparison
    • McNemar test
  • Ordinal outcome comparisons
    • Mann-Whitney U test
    • Wilcoxon signed-rank test
    • Kruskal-Wallis test
    • Friedman test

Redirected branches

  • association questions
  • adjusted effect estimation
  • time-to-event outcomes
  • agreement or reproducibility studies
  • equivalence or non-inferiority studies
  • count outcomes

Installation

Install from GitHub with pak:

install.packages("pak")
pak::pak("dai540/statsguider")

or with remotes:

install.packages("remotes")
remotes::install_github("dai540/statsguider")

Then load the package:

Main Functions

select_test()

This is the shortest entry point. You describe the data properties with choice-style arguments and choose whether to only recommend or to run the test.

dat <- data.frame(
  group = c(rep("control", 8), rep("treated", 8)),
  value = c(5.0, 5.2, 4.8, 5.1, 5.3, 4.9, 5.0, 5.4,
            6.1, 6.0, 5.8, 6.2, 6.3, 5.9, 6.1, 6.0)
)

select_test(
  data = dat,
  outcome = "value",
  group = "group",
  goal = "difference",
  outcome_type = "continuous",
  paired = "no",
  repeated = "no",
  adjust = "no",
  normality = "auto",
  run = "recommend"
)

guided_test()

This function asks the same branching questions interactively. In non-interactive contexts, you can pass a named answers list.

recommend_test()

Returns a structured decision object. Use this when you want to inspect the recommended branch before running anything.

run_test()

Runs the recommended base-R test when the branch is supported.

check_design()

Validates whether the design fits the simple workflow. It reports issues and warnings before recommendation or execution.

Minimal Workflow

The intended sequence is:

  1. Put the analysis data in a data.frame.
  2. Identify the outcome column.
  3. Identify the group column when a group comparison is intended.
  4. Mark design features with simple choices:
    • goal
    • outcome_type
    • paired
    • repeated
    • adjust
    • normality
  5. Use run = "recommend" first.
  6. Only switch to run = "run" after confirming the branch is correct.

Argument Contract

The package intentionally uses a small set of arguments.

Data-identifying arguments

  • data
    • a data.frame
  • outcome
    • the name of the outcome column
  • group
    • the name of the grouping column when needed
  • id
    • the name of the subject identifier column for paired or repeated designs

Choice-style arguments

  • goal
    • "difference"
    • "association"
    • "adjusted_effect"
    • "time_to_event"
    • "agreement"
    • "equivalence"
  • outcome_type
    • "auto"
    • "continuous"
    • "binary"
    • "nominal"
    • "ordinal"
    • "count"
  • paired
    • "yes" or "no"
  • repeated
    • "yes" or "no"
  • adjust
    • "yes" or "no"
  • normality
    • "auto"
    • "yes"
    • "no"
    • "unknown"
  • run
    • "recommend"
    • "run"

Return Objects

Recommendation object

recommend_test() and select_test(..., run = "recommend") return a statsguider_decision object. It contains:

  • the action
  • the selected method
  • an alternative method when relevant
  • the reason for the branch
  • the suggested next step
  • warnings collected during design checking

Execution object

run_test() and select_test(..., run = "run") return a statsguider_result object. It contains:

  • the decision object
  • the base-R test result
  • a short execution summary

Documentation Site

The pkgdown site is organized into four sections:

  • Getting Started
  • Guides
  • Tutorials
  • Reference

The site is published at:

The GitHub repository is:

Repository Policy

This repository is intentionally kept small.

  • No large built-in datasets
  • No downloaded external example datasets
  • No tracked .Rcheck outputs
  • No tracked source tarballs
  • No tracked temporary directories such as tmp_*
  • No tracked generated docs/ directory

Citation

If you use statsguider, cite it as:

Dai (2026). statsguider: Guided Statistical Test Selection with a Minimal Branching Workflow. R package. https://github.com/dai540/statsguider