factorH_reference: factorH functions reference

In factorH: Multifactor Nonparametric Rank-Based ANOVA with Post Hoc Tests

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:

f <- tempfile(fileext = ".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, …)

• 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).

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

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)

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.

factorH documentation built on Sept. 11, 2025, 9:09 a.m.