Package {statwitness}

Type: Package
Title: Model-Aware Validation and Audit Certificates for Statistical Analyses
Version: 0.1.0
Description: Provides model-aware behavioral validation and audit certificates for statistical analyses. Controlled transformations and model-specific diagnostic checks are organized across five domains: computational integrity, numerical stability, design adequacy, assumption screening, and influence stability. Supported workflows include linear models, generalized linear models, classical and repeated-measures analyses of variance, mixed-effects models fitted using 'lme4' or 'glmmTMB', and survival models fitted using 'survival'. Checks are selected according to registered applicability conditions for each model class. The resulting certificates describe computational behavior and selected diagnostic findings; they do not establish causal validity, model correctness, or scientific appropriateness.
License: MIT + file LICENSE
Encoding: UTF-8
Depends: R (≥ 4.1.0)
Imports: graphics, reformulas, stats, tools
Suggests: afex, glmmTMB, knitr, lme4, rmarkdown, survival, testthat (≥ 3.0.0)
VignetteBuilder: knitr
Config/testthat/edition: 3
URL: https://github.com/George33cy/statwitness
BugReports: https://github.com/George33cy/statwitness/issues
Config/roxygen2/version: 8.0.0
NeedsCompilation: no
Packaged: 2026-07-12 10:13:36 UTC; georg
Author: Georgios P. Georgiou [aut, cre]
Maintainer: Georgios P. Georgiou <georgiou.georg@unic.ac.cy>
Repository: CRAN
Date/Publication: 2026-07-21 10:40:08 UTC

Model-Aware Validation and Audit Certificates for Statistical Analyses

Description

Model-aware computational validation for regression, ANOVA, repeated-measures, mixed-effects, and survival workflows.

Details

The package applies registered transformations with known expected behavior. Certificates do not establish causal validity or scientific appropriateness.

Preview an Automatic Audit Plan

Description

Fits the requested model and reports which registered audits are applicable.

Usage

audit_plan(
  formula,
  data,
  focus = NULL,
  family = NULL,
  weights = NULL,
  method = "auto",
  engine = NULL,
  control = NULL,
  audit_level = c("standard", "core", "thorough")
)

Arguments

formula

A model formula.

data

A data frame.

focus

An optional coefficient, variable, or omnibus term.

family

An optional model family.

weights

Optional analysis weights.

method

Requested fitting method.

engine

Optional custom fitting function.

control

Optional model-specific control object.

audit_level

Audit breadth used to construct the plan.

Value

A data frame of selected and skipped tests with reasons.

Generate a Behavioral Validation Certificate

Description

Fits a supported model, chooses mathematically applicable behavioral tests, applies controlled transformations, and checks whether the analysis responds as expected.

Usage

statwitness(
  formula,
  data,
  focus = NULL,
  family = NULL,
  weights = NULL,
  method = "auto",
  engine = NULL,
  control = NULL,
  conf_level = 0.95,
  tolerance = 1e-07,
  seed = 20260712,
  include = NULL,
  exclude = NULL,
  audit_level = c("standard", "core", "thorough")
)

Arguments

formula

A model formula. Random-effect terms using | and survival responses using Surv() are supported when their suggested packages are installed.

data

A data frame containing all analysis variables.

focus

An optional coefficient, predictor, or omnibus term such as "criterion:model".

family

An optional GLM or GLMM family.

weights

Optional analysis weights or the name of a weight column.

method

One of "auto", "lm", "aov", "glm", "lmer", "glmer", "glmmTMB", "coxph", or "survreg".

engine

An optional custom fitting function.

control

An optional model-specific control object.

conf_level

Confidence level for coefficient intervals.

tolerance

Relative numerical tolerance.

seed

Seed for deterministic perturbations.

include

Optional test identifiers to retain.

exclude

Optional test identifiers to omit.

audit_level

Audit breadth: "core", "standard", or "thorough".

Value

An object of class statwitness_audit.

Examples

dat <- transform(mtcars, transmission = factor(am))
audit <- statwitness(mpg ~ transmission + wt, data = dat, focus = "transmission")
print(audit)

Audit a Repeated-Measures ANOVA

Description

Fits a repeated-measures ANOVA using afex::aov_ez() and checks design integrity, sphericity handling, row-order and participant-label invariance, outcome transformations, residual screening, and feasible participant-deletion influence checks.

Usage

statwitness_repeated(
  data,
  outcome,
  id,
  within,
  between = NULL,
  focus = NULL,
  type = 3,
  correction = "GG",
  tolerance = 1e-07,
  seed = 20260712,
  audit_level = c("standard", "core", "thorough")
)

Arguments

data

A long-format data frame.

outcome

Name of the numeric outcome column.

id

Name of the participant identifier column.

within

Character vector of within-subject factor names.

between

Optional between-subject factor names.

focus

Optional omnibus effect name.

type

Sums-of-squares type passed to afex::aov_ez().

correction

Sphericity correction, usually "GG".

tolerance

Relative numerical tolerance.

seed

Seed for row-order testing.

audit_level

Audit breadth: "core", "standard", or "thorough".

Value

An object of class statwitness_audit.

List Registered Behavioral Audit Tests

Description

Returns the built-in statwitness audit registry.

Usage

witness_registry()

Value

A data frame describing registered tests, supported model families, and purposes.

mirror server hosted at Truenetwork, Russian Federation.