factorH: Multifactor rank-based ANOVA utilities

Description

Multifactor nonparametric analysis of variance based on ranks. Builds on the Kruskal-Wallis H test and its 2x2 Scheirer-Ray-Hare extension to handle any factorial designs. Provides effect sizes, Dunn-Bonferroni pairwise-comparison matrices, and simple-effects analyses. Tailored for psychology and the social sciences, with beginner-friendly R syntax and outputs that can be dropped into journal reports. Includes helpers to export tab-separated results and compact tables of descriptive statistics (to APA-style reports).

Details

What this package does (and why)

factorH provides a simple, single-call workflow for multifactor nonparametric, rank-based ANOVA and publication-ready outputs:

• ANOVA-like table based on ranks (rooted in Kruskal-Wallis H and the 2x2 Scheirer-Ray-Hare extension)

• effect sizes computed directly from H

• Dunn–Bonferroni post hoc full comparison matrices

• simple-effects post hocs (pairwise comparisons within levels of conditioning factors)

• compact descriptive tables and a TSV writer for quick formatting in Excel or a manuscript

Why? Popular GUI stats tools do not offer a ready-made, user-friendly multifactor rank-based pipeline that mirrors standard H / SRH analyses in a way that is easy for beginners. factorH aims to fill that gap with clear, R-like formula syntax and a one-command report function.

The package is intentionally small: most users will only ever need:

• srh.kway.full(…) to compute everything

• write.srh.kway.full.tsv(…) to export the results into a single tab-separated file.

Formula syntax at a glance

All high-level functions use standard R model formulas:

response ~ factorA + factorB + factorC

lists main effects - interactions are handled internally. You do not need to write A:B or A*B. The response (left of ~) must be numeric (e.g., a Likert score coded as 1..5 stored as numeric).

Examples below use the included dataset mimicry.

library(factorH)
data(mimicry, package = "factorH")
str(mimicry)

Predictors should be factors. If not, functions will coerce them.

What is allowed?

# One factor (KW-style):  
  liking ~ condition

# Two factors (SRH-style):  
  liking ~ gender + condition

# Three or more factors (k-way):  
  liking ~ gender + condition + age_cat

You do not need to write gender:condition or gender*condition. The package will build all needed interactions internally when relevant.

Numeric response (Likert note)

The response must be numeric. For Likert-type items (e.g., 1 = strongly disagree … 5 = strongly agree), keep them numeric; rank-based tests are robust for such ordinal-like data.

If your Likert is accidentally a factor or character, coerce safely:

# if stored as character "1","2",...:
mimicry$liking <- as.numeric(mimicry$liking)
# if stored as factor with labels "1","2",...:
mimicry$liking <- as.numeric(as.character(mimicry$liking))

Diagnostics at a glance

Most users can cover assumption checks with a single command:

  diag_out <- plan.diagnostics(response ~ factorA + factorB (+ factorC ...), data = your_data)

What it does:

1. Raw normality: Shapiro–Wilk in each subgroup and interaction cell of the specified factors.

2. Residual normality per cell: Shapiro–Wilk on residuals from the corresponding full-factorial ANOVA, tested within each cell.

3. Homogeneity of variances: Levene/Brown–Forsythe across full-plan cells (median by default).

4. Count balance: chi-square homogeneity/independence/log-linear independence across factors.

5. It prints a concise overall summary (share of OK and overall status) and returns all detailed tables in diag_out$⁠results, with per-type OK percentages in diag_out⁠$summary. For most workflows, this single command is enough to document model assumptions alongside rank-based analyses.

The one-call pipeline

The main function srh.kway.full() runs:

1. ANOVA-like table on ranks

2. descriptive summary

3. post hoc matrices (Dunn–Bonferroni; P.adj)

4. simple-effects post hocs (within-family Bonferroni).

For 2 factors:

res2 <- srh.kway.full(liking ~ gender + condition, data = mimicry)
names(res2)
res2$anova
head(res2$summary)
names(res2$posthoc_cells)
names(res2$posthoc_simple)[1:4]

For 3 factors:

res3 <- srh.kway.full(liking ~ gender + condition + age_cat, data = mimicry)
res3$anova

Export full result to a tab-separated file

# you can of course provide your own path to the file outside the temporary folder
f <- file.path(tempdir(), "result.tsv")
write.srh.kway.full.tsv(res3, file = f, dec = ".") # decimal dot
file.exists(f)

If you need comma as decimal mark:

f <- file.path(tempdir(), "result.tsv")
write.srh.kway.full.tsv(res3, file = f2, dec = ",") # decimal comma
file.exists(f2)

The TSV contains clearly separated sections: ## SRH: EFFECTS TABLE, ## SUMMARY STATS, ## POSTHOC CELLS, ## SIMPLE EFFECTS, ## META. and can be easily pasted into the any equivalent Excel or Google spreadsheets.

What is in the example dataset?

mimicry is a real study on the chameleon effect (Trzmielewska, Duras, Juchacz & Rak, 2025): how mimicry vs other movement conditions affect liking of an interlocutor. Potential moderators include gender and age (with dichotomized age_cat, and a 3-level age_cat2). This makes it a natural playground for multifactor rank-based analyses.

table(mimicry$condition)
table(mimicry$gender)
table(mimicry$age_cat)

What the functions compute (high level)

• srh.kway(): rank-based k-way ANOVA table using Type II SS (by default, possible switch to III SS) on ranks; p-values are tie-corrected; H is reported with and without the correction factor; effect sizes from unadjusted H.

• srh.effsize(): 2-way SRH table with effect sizes (eta2H, eps2H) computed from H.

• nonpar.datatable(): compact descriptive tables with global ranks (means of ranks per cell), medians, quartiles, IQR, etc., for all main effects and interactions.

• srh.posthocs(): Dunn–Bonferroni pairwise matrices (P.adj) for all effects (main and interactions).

• srh.simple.posthoc() / srh.simple.posthocs(): simple-effects pairwise comparisons within levels of conditioning factors (SPSS-like “within” scope by default).

• srh.kway.full(): orchestrates all of the above.

• write.srh.kway.full.tsv(): exports everything into one TSV (with dot or comma decimal mark).

• plan.diagnostics(): one-call diagnostics: raw normality, residuals cellwise normality, Levene (median), balance chi-square; prints overall summary and returns full tables.

That is it. For most users, the intro ends here: use srh.kway.full() and export with write.srh.kway.full.tsv().

Author(s)

Maintainer: Tomasz Rak tomasz.rak@upjp2.edu.pl

Authors:

• Szymon Wrzesniowski szymon.wrzesniowski@upjp2.edu.pl

Count-balance chi-square diagnostics across factors

Description

For one factor: chi-square goodness-of-fit vs equal proportions. For two factors: chi-square test of independence. For three or more: log-linear independence (Poisson, main effects only) via deviance and df.

Usage

balance.chisq.datatable(formula, data, force_factors = TRUE, correct = FALSE)

Arguments

Details

Uses stats::chisq.test for 1–2 factors. For 3+ factors, prefers MASS::loglm if available; otherwise falls back to a Poisson GLM on the count table.

Value

A data.frame with one row per factor combination (Effect) and columns: n, ChiSq (4 decimals), df, p.chisq (4 decimals), OK.

See Also

plan.diagnostics

Examples

## Not run: 
balance.chisq.datatable(liking ~ gender + condition + age_cat, data = mimicry)

## End(Not run)

Datasets in factorH

Description

Datasets in factorH

Details

What is in the example dataset?

mimicry is a real study on the chameleon effect by Trzmielewska et al. (2025) doi:10.18290/rpsych2024.0019 about how mimicry vs other movement conditions affect liking of an interlocutor. Potential moderators include gender and age (with dichotomized age_cat, and a 3-level age_cat2). This makes it a natural playground for multifactor rank-based analyses.

table(mimicry$condition)
table(mimicry$gender)
table(mimicry$age_cat)

factorH functions reference

Description

factorH functions reference

Details

Function reference

This document collects call patterns and options for each public function. All formulas follow response ~ A + B (+ C …) with numeric response and factor predictors.

srh.kway.full()

Purpose: one-call pipeline: ANOVA on ranks + descriptives + post hocs + simple effects. Syntax: srh.kway.full(y ~ A + B (+ C …), data, max_levels = 30)

• Automatically chooses the ANOVA engine:

  • 1 factor: srh.kway()

  • 2 factors: srh.effsize()

  • 3+ factors: srh.kway()

• Returns a list: anova, summary, posthoc_cells, posthoc_simple, meta.

• Placeholders:

  • not applicable when a component does not apply (e.g., simple effects with 1 factor),

  • failed… when a sub-step errors out (keeps the pipeline alive).

Example:

res <- srh.kway.full(liking ~ gender + condition + age_cat, data = mimicry)
names(res)
res$anova[1:3]
head(res$summary)
names(res$posthoc_cells)
names(res$posthoc_simple)[1:3]
res$meta

Notes:

• Predictors are coerced to factor internally; levels must be 2..max_levels.

• Missing values are removed pairwise on the variables in the formula.

write.srh.kway.full.tsv()

Purpose: export the srh.kway.full() result into a single TSV file for fast formatting. Syntax: write.srh.kway.full.tsv(obj, file = “srh_kway_full.tsv”, sep = “, na =”“, dec =”.”)

• dec = “.” or “,” controls the decimal mark.

• Numeric fields are written without scientific notation.

• Pretty-printed character tables (e.g., from post hocs) are normalized so that dec=“,” also affects numbers embedded in strings.

Example:

# you can of course provide your own path to the file outside the temporary folder
f <- file.path(tempdir(), "result.tsv")
write.srh.kway.full.tsv(res, file = f, dec = ",")
file.exists(f)

srh.kway()

Purpose: general k-way SRH-style ANOVA on ranks (Type II SS), tie-corrected p-values. Syntax: srh.kway(y ~ A + B (+ C …), data, clamp0 = TRUE, force_factors = TRUE, type = 2, …)

• Reports: Effect, Df, Sum Sq, H, Hadj (tie correction), p.chisq, k, n, eta2H, eps2H.

• eta2H and eps2H are computed from unadjusted H (classical SRH practice).

• force_factors = TRUE coerces predictors to factor (recommended).

• type controls sums of squares. Default type = 2 (Type II SS). Set type = 3 for Type III SS (internally uses sum-to-zero contrasts; no global options changed).

Example:

k3 <- srh.kway(liking ~ gender + condition + age_cat, data = mimicry)
k3

One-factor check (KW-like):

k1 <- srh.kway(liking ~ condition, data = mimicry)
k1

Two factor (Type III SS):

k3_ss3 <- srh.kway(liking ~ gender + condition, data = mimicry, type = 3)
k3_ss3

srh.effsize()

Purpose: 2-way SRH table with effect sizes from H. Syntax: srh.effsize(y ~ A + B, data, clamp0 = TRUE, …)

• Same columns as above but tailored to 2-way SRH.

• clamp0 = TRUE clamps small negatives to 0 for effect sizes.

Example:

e2 <- srh.effsize(liking ~ gender + condition, data = mimicry)
e2

nonpar.datatable()

Purpose: compact descriptive tables (APA-style), with global rank means, medians, quartiles, IQR. Syntax: nonpar.datatable(y ~ A + B (+ C …), data, force_factors = TRUE)

• Returns rows for all main effects and all interaction cells (constructed internally).

• Rank means are computed on global ranks (all observations ranked together), which matches how rank-based ANOVA effects are formed.

Example:

dt <- nonpar.datatable(liking ~ gender + condition, data = mimicry)
head(dt)

srh.posthoc()

Purpose: Dunn–Bonferroni pairwise comparison matrix for a specified effect. Syntax: srh.posthoc(y ~ A (+ B + …), data, method = “bonferroni”, digits = 3, triangular = c(“lower”,“upper”,“full”), numeric = FALSE, force_factors = TRUE, sep = “.”)

• Builds a single grouping variable (cells) from the RHS factors and runs FSA::dunnTest.

• Returns a list of three matrices (as data.frames): Z, P.unadj, P.adj.

• triangular = “lower” (default) shows only the lower triangle; diagonal and upper triangle are blank.

• numeric = FALSE returns pretty-printed character tables; set TRUE to get numeric.

Example:

ph <- srh.posthoc(liking ~ condition, data = mimicry)

srh.posthocs()

Purpose: Dunn–Bonferroni pairwise matrices for all effects (main and interactions). Syntax: srh.posthocs(y ~ A + B (+ C …), data, …)

• Iterates srh.posthoc over: A, B, C, A:B, A:C, B:C, A:B:C, …

• Returns a named list: names are “A”, “B”, “A:B”, etc.; each value is a P.adj matrix.

Example:

phs <- srh.posthocs(liking ~ gender + condition + age_cat, data = mimicry)
names(phs)
phs[["gender:condition"]][1:5, 1:5]

srh.simple.posthoc()

Purpose: Simple-effects post hocs (pairwise comparisons within levels of conditioning factors). Syntax: srh.simple.posthoc(y ~ A + B (+ C …), data, compare = NULL, scope = c(“within”,“global”), digits = 3)

• compare selects the target factor for pairwise comparisons (default: first RHS factor).

• Scope:

  • “within” (default): Bonferroni within each by-table (SPSS-like).

  • “global”: one Bonferroni across all tests from all by-tables combined.

• Returns a data.frame with conditioning columns (BY), Comparison, Z, P.unadj, P.adj, m.tests, adj.note. An “adjustment” attribute describes the correction.

Example:

simp <- srh.simple.posthoc(liking ~ gender + condition + age_cat, data = mimicry, compare = "gender", scope = "within")
head(simp)

srh.simple.posthocs()

Purpose: enumerate all simple-effect configurations for a given design. Syntax: srh.simple.posthocs(y ~ A + B (+ C …), data)

• For each target factor and each non-empty combination of the remaining factors as BY, runs srh.simple.posthoc(…, scope = “within”).

• Returns a named list, names like COMPARE(gender) | BY(condition x age_cat).

Example:

sps <- srh.simple.posthocs(liking ~ gender + condition + age_cat, data = mimicry)
head(names(sps), 6)

normality.datatable

Purpose: Shapiro–Wilk normality tests for the raw response within each subgroup for all non-empty combinations of RHS factors (main effects and interaction cells). Syntax: normality.datatable(y ~ A + B (+ C …), data, force_factors = TRUE)

• Returns Effect, factor columns, count, W, p.shapiro (fixed-format to 4 decimals, no scientific notation), and OK/NOT OK (p < 0.05 => NOT OK).

Example:

normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

residuals.normality.datatable

Purpose: Shapiro–Wilk normality tests on residuals from a classical ANOVA model fitted to the selected RHS factors (full factorial for those factors), one test per model (global residuals). Syntax: residuals.normality.datatable(y ~ A + B (+ C …), data, force_factors = TRUE)

• Returns one row per Effect (A, B, A:B, …), with count, W, p.shapiro (4 decimals), OK/NOT OK. Use the cellwise variant below for the strict per-cell assumption.

• We have retained this feature for the purpose of recording older versions of the software, but according to the newer statistical literature it should not be used to determine the validity of a research plan.

Example:

residuals.normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

residuals.cellwise.normality.datatable

Purpose: Shapiro–Wilk tests of residuals from an ANOVA model fitted to the selected RHS factors (full factorial), but tested separately within each cell defined by those factors. Syntax: residuals.cellwise.normality.datatable(y ~ A + B (+ C …), data, force_factors = TRUE)

• This matches the classical ANOVA assumption of normal errors per cell. Returns rows for every cell across all Effects, with count, W, p.shapiro (4 decimals), OK/NOT OK.

Example:

residuals.cellwise.normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

balance.chisq.datatable

Purpose: Count-balance diagnostics across design factors. Syntax: balance.chisq.datatable(y ~ A + B (+ C …), data, force_factors = TRUE)

• For one factor: chi-square test of homogeneity vs equal proportions. For two factors: chi-square test of independence on the contingency table. For three or more: log-linear independence model (Poisson; main effects only) assessed via deviance and df. Returns Effect, n, ChiSq (4 decimals), df, p.chisq (4 decimals), OK/NOT OK (p < 0.05 => NOT OK).

• Note: The response is ignored; only RHS factors are used to build the tables.

Example:

balance.chisq.datatable(liking ~ gender + condition + age_cat, data = mimicry)

levene.plan.datatable

Purpose: Levene/Brown–Forsythe test for homogeneity of variances across the full-plan cells (highest-order interaction of RHS factors). Syntax: levene.plan.datatable(y ~ A + B (+ C …), data, center = “median”, force_factors = TRUE)

• This is the primary variance-equality diagnostic for factorial ANOVA. Returns F, df.num, df.den, p (4 decimals), and OK/NOT OK (p < 0.05 => NOT OK).

Examples:

levene.plan.datatable(liking ~ gender + condition + age_cat, data = mimicry)  
levene.plan.datatable(liking ~ gender + condition, data = mimicry, center = "mean")

plan.diagnostics

Purpose: Orchestrates all diagnostics in one call. Syntax: plan.diagnostics(y ~ A + B (+ C …), data, force_factors = TRUE)

• Runs raw normality (cellwise on the response), residuals cellwise normality, Levene/Brown–Forsythe for the full plan (median by default), and balance chi-square tests for all factor combinations.

• Prints a concise console summary and returns full tables in a list.

• Console summary: prints overall share of OK and overall status (OK only if 100% OK).

Returned list:

$summary: percent_ok, ok_count, total, overall, plus per-type percentages:  
percent_ok_normality_raw, percent_ok_residuals_cellwise, percent_ok_balance_chisq, percent_ok_levene_full_plan.  
$results: normality_raw, residuals_cellwise_normality, levene_full_plan, balance_chisq.  

Examples:

diag_out <- plan.diagnostics(liking ~ gender + condition + age_cat, data = mimicry)  
diag_out$results$normality_raw  
diag_out$results$residuals_cellwise_normality  
diag_out$results$levene_full_plan  
diag_out$results$balance_chisq  
diag_out$summary

Formula tips and pitfalls

• Do not write A:B or A*B. Use A + B (+ C …); the package computes all necessary interaction structures internally.

• Response must be numeric. For Likert data, keep it numeric 1..k.

• Predictors should be factors. If they are not, they will be coerced.

• Coerce predictors to factor explicitly if needed

Example:

#coercing
mimicry$gender <- factor(mimicry$gender)
mimicry$condition <- factor(mimicry$condition)

Performance and reproducibility

• Functions use ranks and Type II sums of squares (via car::Anova under the hood) and Dunn tests (FSA::dunnTest).

• P-values apply a standard tie correction factor for ranks; effect sizes are derived from unadjusted H (classical SRH practice).

• All outputs are plain data.frames and lists, easy to save and post-process.

Syntax and formula patterns

Description

Syntax and formula patterns

Details

Formula syntax at a glance

All high-level functions use standard R model formulas: response ~ factorA + factorB + factorC

• + lists main effects - Interactions are handled internally. You do not need to write A:B or A*B.

• The response (left of ~) must be numeric (e.g., a Likert score coded as 1..5 stored as numeric).

Examples below use the included dataset mimicry.

library(factorH)
data(mimicry, package = "factorH")
str(mimicry)

Predictors should be factors. If not, functions will coerce them.

What is allowed?

# One factor (KW-style):  
  liking ~ condition

# Two factors (SRH-style):  
  liking ~ gender + condition

# Three or more factors (k-way):  
  liking ~ gender + condition + age_cat

You do not need to write gender:condition or gender*condition. The package will build all needed interactions internally when relevant.

Numeric response (Likert note)

The response must be numeric. For Likert-type items (e.g., 1 = strongly disagree … 5 = strongly agree), keep them numeric; rank-based tests are robust for such ordinal-like data.

If your Likert is accidentally a factor or character, coerce safely:

# if stored as character "1","2",...:
mimicry$liking <- as.numeric(mimicry$liking)
# if stored as factor with labels "1","2",...:
mimicry$liking <- as.numeric(as.character(mimicry$liking))

Levene/Brown-Forsythe test for full-plan cells

Description

Tests homogeneity of variances across the highest-order interaction (all RHS factors combined), using Levene's test (Brown-Forsythe with median by default).

Usage

levene.plan.datatable(
  formula,
  data,
  center = c("median", "mean"),
  force_factors = TRUE
)

Arguments

Details

Internally relies on car::leveneTest. If fewer than two groups or any group has < 2 observations, NA values are returned with a warning.

Value

A one-row data.frame with columns: Effect, n.groups, min.n, df.num, df.den, F, p, OK. Values F and p are formatted to 4 decimals (no scientific notation); OK is "OK" if p >= 0.05, otherwise "NOT OK".

See Also

plan.diagnostics

Examples

## Not run: 
levene.plan.datatable(liking ~ gender + condition + age_cat, data = mimicry)
levene.plan.datatable(liking ~ gender + condition, data = mimicry, center = "mean")

## End(Not run)

Mimicry dataset

Description

A dataset used to demonstrate rank-based (nonparametric) multifactor ANOVA.

Usage

data(mimicry)

Format

A data frame with 533 rows and 7 variables:

condition

factor; 5 levels

gender

factor; 2 levels

age

numeric

age_cat

factor; 2 levels

age_cat2

factor; 3 levels

field

factor; 2 levels

liking

numeric; dependent variable

Details

Factor encodings follow the original SPSS labels converted to R factors.

Source

Converted from an SPSS file as part of the factorH package examples.

References

Trzmielewska, W., Duras, J., Juchacz, A., & Rak, T. (2025). Examining the impact of control condition design in mimicry–liking link research: how motor behavior may impact liking. Annals of Psychology, 4, 351–378. doi:10.18290/rpsych2024.0019

Compact descriptive tables (APA-style) with global rank means

Description

Produces descriptive statistics for all main effects and interaction cells implied by the RHS of formula. Ranks are computed globally (across all observations) and cell-wise mean ranks are reported (recommended for interpreting rank-based factorial effects).

Usage

nonpar.datatable(formula, data, force_factors = TRUE)

Arguments

Details

The function first subsets to complete cases on y and all RHS factors, then computes global ranks of y (ties.method = "average"). For each effect (every non-empty combination of factors up to full order), it returns a row per cell with: count, mean, sd, median, quartiles (q1, q3), IQR, and mean_rank. The column Effect identifies the effect (e.g., "A", "B", "A:B"). Missing factor columns for a given effect are added with NA values but retain the proper factor levels for easy binding.

Value

A base data.frame with columns:

• Effect (character),

• factor columns for all RHS factors (factors, possibly NA in some rows),

• count, mean, sd, median, q1, q3, IQR, mean_rank.

The original call is attached as attribute "call".

Examples

data(mimicry, package = "factorH")

# One factor
nonpar.datatable(liking ~ condition, data = mimicry)

# Two factors: rows for gender, for condition, and for gender:condition
nonpar.datatable(liking ~ gender + condition, data = mimicry)

# Three factors: all mains + 2-way and 3-way cells
nonpar.datatable(liking ~ gender + condition + age_cat, data = mimicry)

Raw normality per subgroup (Shapiro–Wilk) across factor combinations

Description

Runs Shapiro–Wilk tests on the raw response within each subgroup for all non-empty combinations of RHS factors (main effects and interaction cells).

Usage

normality.datatable(formula, data, force_factors = TRUE)

Arguments

Value

A data.frame with rows per subgroup/cell. Columns: Effect, factor columns, count, W, p.shapiro (4 decimals), OK.

See Also

plan.diagnostics

Examples

## Not run: 
normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

## End(Not run)

Plan-level diagnostics for ANOVA/rank-based workflows

Description

Runs all assumption checks in one call: raw normality per subgroup (Shapiro-Wilk), residual normality per cell (from a full-factorial ANOVA on the specified factors), Levene/Brown-Forsythe for the full plan (median by default), and count-balance chi-square tests for all factor combinations. Prints a concise summary and returns all detailed tables in a list.

Usage

plan.diagnostics(formula, data, force_factors = TRUE)

Arguments

Details

Requires helper functions defined in this package: normality.datatable, residuals.cellwise.normality.datatable, levene.plan.datatable, balance.chisq.datatable. Levene's test uses car; if unavailable, the Levene block returns NA rows with a warning.

Value

An invisible list with:

• $summary: overall percent_ok, ok_count, total, overall, plus per-type percentages (percent_ok_normality_raw, percent_ok_residuals_cellwise, percent_ok_balance_chisq, percent_ok_levene_full_plan).

• $results: data.frames for normality_raw, residuals_cellwise_normality, levene_full_plan, balance_chisq.

See Also

normality.datatable, residuals.cellwise.normality.datatable, levene.plan.datatable, balance.chisq.datatable

Examples

## Not run: 
diag_out <- plan.diagnostics(liking ~ gender + condition + age_cat, data = mimicry)
diag_out$summary
diag_out$results$normality_raw

## End(Not run)

Cellwise residual normality (Shapiro–Wilk) from ANOVA models

Description

Fits, for each subset of RHS factors, a full-factorial ANOVA to the response and tests Shapiro–Wilk normality of residuals within each cell defined by those factors. Matches the classical ANOVA assumption of normal errors per cell.

Usage

## S3 method for class 'cellwise.normality.datatable'
residuals(formula, data, force_factors = TRUE)

Arguments

Value

A data.frame with rows per cell across all factor combinations. Columns include: Effect, factor columns (with NA for factors not in the current subset), count, W, p.shapiro (4 decimals), OK.

See Also

normality.datatable, plan.diagnostics

Examples

## Not run: 
residuals.cellwise.normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

## End(Not run)

Global residual normality (Shapiro–Wilk) from ANOVA models

Description

For each subset of RHS factors, fits a full-factorial ANOVA and runs a single Shapiro–Wilk test on the model residuals (global test per model). Use residuals.cellwise.normality.datatable for the stricter per-cell assumption.

Usage

## S3 method for class 'normality.datatable'
residuals(formula, data, force_factors = TRUE)

Arguments

Value

A data.frame with one row per Effect (A, B, A:B, ...), with count, W, p.shapiro (4 decimals), OK.

See Also

residuals.cellwise.normality.datatable

Examples

## Not run: 
residuals.normality.datatable(liking ~ gender + condition + age_cat, data = mimicry)

## End(Not run)

SRH with effect sizes for two-factor designs

Description

Extends rcompanion::scheirerRayHare() by adding popular rank-based effect sizes for each SRH term: eta^2_H and epsilon^2_H, and stores the original function call.

Usage

srh.effsize(formula, data, clamp0 = TRUE, ...)

Arguments

Details

Let H be the SRH H-statistic for a given term, n the sample size used by SRH (complete cases on y and factors), and k the number of groups compared by that term (for interactions, the number of observed combinations).

Effect sizes computed:

• Eta^2_H: (H - k + 1) / (n - k).

• Epsilon^2_H (KW-like): H * (n + 1) / (n^2 - 1).

The original call is stored as an attribute and can be retrieved with getCall().

Value

A data.frame (classed as c("srh_with_call","anova","data.frame")) with the SRH table extended by columns: k, n, eta2H, eps2H.

Examples

data(mimicry, package = "factorH")
res <- srh.effsize(liking ~ gender + condition, data = mimicry)
res
getCall(res)

K-way SRH on ranks with tie-corrected p-values and rank-based effect sizes

Description

Generalizes the Scheirer–Ray–Hare (SRH) approach to k-factor designs by using sums of squares from a linear model on ranks, with a standard tie correction D applied to p-values. The function returns H, tie-corrected H (Hadj), p-values and rank-based effect sizes (eta2H, eps2H) for each main effect and interaction up to the full order (i.e., (A + B + ...)^k).

Usage

srh.kway(formula, data, clamp0 = TRUE, force_factors = TRUE, type = 2, ...)

Arguments

Details

Ranks are computed globally on y with ties.method = "average". Sums of squares are obtained from car::Anova() on the rank model R ~ (A + B + ...)^k. Tie correction:

D = 1 - \frac{\sum (t^3 - t)}{n^3 - n},

where t are tie block sizes and n is the number of complete cases. We report Hadj = H / D and p = P(\chi^2_{df} \ge Hadj).

Rank-based effect sizes are computed from the uncorrected H (classical SRH convention): eta2H = (H - k + 1) / (n - k) and eps2H = H * (n + 1) / (n^2 - 1), where k is the number of non-empty groups compared by the term.

For type = 3, the model is fitted with sum-to-zero contrasts (stats::contr.sum) for RHS factors having at least 2 levels, so that Type III tests have the standard interpretation. Global contrast options are not altered.

Value

A data.frame with class c("srh_kway","anova","data.frame") containing columns: Effect, Df, Sum Sq, H, Hadj, p.chisq, k, n, eta2H, eps2H. The original call is attached as an attribute and can be retrieved with getCall().

See Also

Anova

Examples

## Not run: 
data(mimicry, package = "factorH")
# One factor (KW-style check)
srh.kway(liking ~ condition, data = mimicry)

# Two factors (Type II by default)
srh.kway(liking ~ gender + condition, data = mimicry)

# Three factors
srh.kway(liking ~ gender + condition + age_cat, data = mimicry)

# Type III SS (with sum-to-zero contrasts set locally)
srh.kway(liking ~ gender + condition, data = mimicry, type = 3)

## End(Not run)

Full pipeline: rank-based k-way ANOVA + descriptives + post hocs

Description

Runs a complete nonparametric, rank-based workflow for factorial designs: (1) SRH-style ANOVA table, (2) compact descriptive stats with global ranks, (3) Dunn-Bonferroni post hoc matrices for all effects, and (4) simple-effects post hocs (Bonferroni within each by-table).

Usage

srh.kway.full(formula, data, max_levels = 30)

Arguments

Details

Choice of the ANOVA engine:

• 1 factor: srh.kway() (KW-like),

• 2 factors: srh.effsize() (SRH 2-way + effect sizes),

• 3+ factors: srh.kway() (general k-way on ranks).

Value

A list with elements:

• anova – ANOVA-like table,

• summary – descriptive stats data.frame,

• posthoc_cells – list of p.adj matrices for all effects (from srh.posthocs), or a string when failed,

• posthoc_simple – list of simple-effect tables (from srh.simple.posthocs); for 1 factor: "[not applicable]",

• meta – list with call, n, factor levels, and empty-cell info (if 2+ factors).

Components that cannot be computed for the given design are returned as the string "[not applicable]"; failures are reported as "[failed] <message>".

Examples

data(mimicry, package = "factorH")
# 1 factor
f1 <- srh.kway.full(liking ~ condition, data = mimicry)
# 2 factors
f2 <- srh.kway.full(liking ~ gender + condition, data = mimicry)
# 3 factors
f3 <- srh.kway.full(liking ~ gender + condition + age_cat, data = mimicry)

Dunn post hoc in a symmetric matrix form (one specified effect)

Description

Computes Dunn's rank-based pairwise comparisons for the effect implied by formula and returns symmetric matrices for Z, unadjusted p-values, and adjusted p-values. Cells on one triangle (or both) can be blanked for compact reporting. For multi-factor RHS, factors are combined into a single grouping via interaction() (e.g., "A:B" cells).

Usage

srh.posthoc(
  formula,
  data,
  method = "bonferroni",
  digits = 3,
  triangular = c("lower", "upper", "full"),
  numeric = FALSE,
  force_factors = TRUE,
  sep = "."
)

Arguments

Details

The function subsets to complete cases on y and RHS factors, optionally coerces factors, builds a single grouping variable (._grp) and calls FSA::dunnTest(y ~ ._grp, data = ..., method = ...). The pairwise results are placed into symmetric matrices Z, P.unadj, and P.adj. By default only the lower triangle (excluding diagonal) is shown for compactness.

Value

A list with three data.frames:

• Z – Z statistics,

• P.unadj – unadjusted p-values,

• P.adj – adjusted p-values (per method).

The original call is attached as attribute "call".

Examples

data(mimicry, package = "factorH")

# One factor
ph1 <- srh.posthoc(liking ~ condition, data = mimicry)
ph1$`P.adj`    # gotowa macierz p po korekcji

# Two factors combined (all A:B cells vs all A:B cells)
ph2 <- srh.posthoc(liking ~ gender + condition, data = mimicry)
ph2$`P.adj`

# Upper triangle, numeric frames
ph3 <- srh.posthoc(liking ~ condition, data = mimicry,
                   triangular = "upper", numeric = TRUE)
ph3$Z

Dunn post hoc tables (p.adj only) for all effects in a factorial design

Description

For a given y ~ A (+ B + ...) formula, runs srh.posthoc for every main effect and interaction implied by the RHS (all non-empty combinations of factors) and returns a named list of adjusted p-value matrices (P.adj) for each effect.

Usage

srh.posthocs(
  formula,
  data,
  method = "bonferroni",
  digits = 3,
  triangular = c("lower", "upper", "full"),
  numeric = FALSE,
  force_factors = TRUE,
  sep = "."
)

Arguments

Details

The function enumerates all non-empty subsets of RHS factors (mains, 2-way, ..., k-way) and calls srh.posthoc on each corresponding sub-formula. If a subset has fewer than 2 observed levels (e.g., due to missing data after subsetting to complete cases), that effect is skipped.

Value

A named list where each element is a data.frame of adjusted p-values (P.adj) for an effect. Names use "A", "B", "A:B", ..., matching the effect structure. The original call is attached as attribute "call".

Examples

data(mimicry, package = "factorH")

# Two-factor design: p.adj for 'gender', 'condition', and 'gender:condition'
L2 <- srh.posthocs(liking ~ gender + condition, data = mimicry)
names(L2)
L2$gender
L2$condition
L2$`gender:condition`

# Three-factor design: includes mains, all 2-ways, and the 3-way effect
L3 <- srh.posthocs(liking ~ gender + condition + age_cat, data = mimicry)
names(L3)

Simple-effects post hoc (Dunn) with Bonferroni adjustment

Description

Computes Dunn's pairwise comparisons for simple effects of one target factor (compare) within levels of the remaining conditioning factors (by). Adjustment can be done within each conditioning table (SPSS-like) or globally across all tests.

Usage

srh.simple.posthoc(
  formula,
  data,
  compare = NULL,
  scope = c("within", "global"),
  digits = 3
)

Arguments

Details

The data are subset to complete cases on y and all RHS factors. All RHS variables are coerced to factor. The table is split by all factors except compare and Dunn's test (FSA::dunnTest) is run per split. With scope = "within", the Bonferroni correction is applied separately in each split (with m.tests = choose(k,2) for that split). With scope = "global", P.adj is re-computed once with stats::p.adjust(..., method = "bonferroni") across all pairwise tests from all splits (and m.tests is set to the total number of tests).

Value

A data.frame with columns:

• conditioning factor columns (one value repeated per split),

• Comparison, Z, P.unadj, P.adj,

• m.tests (number of tests used for Bonferroni),

• adj.note (human-readable note).

Attributes: "adjustment" (one-line description) and "call".

Examples

data(mimicry, package = "factorH")

# Two factors: pairwise comparisons for 'gender' within levels of 'condition'.
# By default, compare = first RHS factor ('gender' here).
# p.adj uses Bonferroni within each by-table (scope = "within").
tab1 <- srh.simple.posthoc(liking ~ gender + condition, data = mimicry)
head(tab1); attr(tab1, "adjustment")

# One global family of tests (global Bonferroni across all subgroup tests):
tab2 <- srh.simple.posthoc(liking ~ gender + condition, data = mimicry,
                           scope = "global")
head(tab2); attr(tab2, "adjustment")

# Three factors: compare 'gender' within each condition × age_cat cell.
tab3 <- srh.simple.posthoc(liking ~ gender + condition + age_cat, data = mimicry)
head(tab3)

# Choose a different target factor to compare: here 'condition'
# (within each gender × age_cat cell).
tabA <- srh.simple.posthoc(liking ~ gender + condition + age_cat, data = mimicry,
                           compare = "condition")
head(tabA)

# Global Bonferroni variants (less common, but sometimes requested):
tabG  <- srh.simple.posthoc(liking ~ gender + condition + age_cat, data = mimicry,
                            scope = "global")
tabG2 <- srh.simple.posthoc(liking ~ condition + gender, data = mimicry)
tabG3 <- srh.simple.posthoc(liking ~ condition + gender, data = mimicry,
                            scope = "global")
head(tabG); head(tabG2); head(tabG3)

Simple-effects post hoc tables for all possible effects (within-scope)

Description

For a formula y ~ A + B (+ C ...), enumerates all simple-effect setups of the form COMPARE(target) | BY(other factors) and runs srh.simple.posthoc with scope = "within" for each. Returns a named list of data frames (one per simple-effect configuration).

Usage

srh.simple.posthocs(formula, data)

Arguments

Details

For each choice of the comparison factor target from the RHS, all non-empty combinations of the remaining factors are treated as conditioning sets BY. For each pair (target, BY) we call srh.simple.posthoc() with compare = target and scope = "within". Effects where the conditioning subset has < 2 levels of target are skipped; messages are collected in attribute "skipped".

Labels use ASCII: "COMPARE(A) | BY(B x C)" (plain " x ").

Value

A named list of data.frames. Each element contains the columns produced by srh.simple.posthoc (e.g., Comparison, Z, P.unadj, P.adj, m.tests, adj.note). Attributes: "call" and (optionally) "skipped" with messages.

Examples

data(mimicry, package = "factorH")

# All simple-effect tables for a 2-factor design
tabs2 <- srh.simple.posthocs(liking ~ gender + condition, data = mimicry)
names(tabs2)
# e.g., tabs2[["COMPARE(gender) | BY(condition)"]]

# Three factors: all COMPARE(target) | BY(conditioning) combinations
tabs3 <- srh.simple.posthocs(liking ~ gender + condition + age_cat, data = mimicry)
names(tabs3)
attr(tabs3, "skipped")  # any skipped combos with reasons

Write full SRH pipeline result to a TSV file

Description

Exports the result of srh.kway.full into a single, tab-separated text file, in the order: ANOVA > SUMMARY > POSTHOC CELLS > SIMPLE EFFECTS > META. Supports choosing the decimal mark for numeric values.

Usage

write.srh.kway.full.tsv(
  obj,
  file = "srh_kway_full.tsv",
  sep = "\t",
  na = "",
  dec = "."
)

Arguments

Details

Each section is preceded by a header line (e.g., ## SRH: EFFECTS TABLE). For post hoc sections, each effect/table is prefixed with a subheader (e.g., ### posthoc_cells: gender:condition). For simple-effect tables, the attribute "adjustment" (if present) is written as a comment line beginning with "# ".

Components that are not applicable (e.g., simple effects in 1-factor designs) or failed computations are written as literal one-line messages.

Value

(Invisibly) the normalized path to file.

Examples

data(mimicry, package = "factorH")
res <- srh.kway.full(liking ~ gender + condition, data = mimicry)

# Write to a temporary file (CRAN-safe)
f <- tempfile(fileext = ".tsv")
write.srh.kway.full.tsv(res, file = f, dec = ".")
file.exists(f)