Skip to contents

Compute pairwise statistical tests between groups

Source: R/stat_helper.R
compute_pairwise_stats.Rd

Performs pairwise statistical tests between groups defined by a categorical variable. Supports Wilcoxon rank-sum, t-test, Kruskal-Wallis, and ANOVA. Handles nested grouping (comparing color.by groups within each x-level) and per-facet testing.

Usage

compute_pairwise_stats(
  df,
  x,
  y,
  pairs = NULL,
  test = "wilcox.test",
  p.adjust.method = "holm",
  paired = FALSE,
  group.by = NULL,
  facet.by = NULL,
  per.facet = TRUE,
  sig.threshold = 0.05,
  sig.levels = c(`****` = 1e-04, `***` = 0.001, `**` = 0.01)
)

Arguments

df

Data frame containing the data.

x

Character; column name of the categorical x-axis variable.

y

Character; column name of the numeric response variable.

pairs

List of length-2 character vectors specifying group pairs to test. If NULL (default), tests all unique pairwise combinations.

test

Character; statistical test to use. One of "wilcox.test", "t.test", "kruskal.test", or "anova".

p.adjust.method

Character; method for p-value adjustment via stats::p.adjust(). Default "holm".

paired

Logical; whether to perform paired tests (only for "wilcox.test" and "t.test"). Default FALSE.

group.by

Character or NULL; column for nested grouping. When set, comparisons are made between levels of group.by within each level of x.

facet.by

Character or NULL; column for faceting. When set and per.facet = TRUE, tests run independently per facet panel.

per.facet

Logical; if TRUE and facet.by is set, run tests independently per facet panel. Default TRUE.

sig.threshold

Numeric; significance threshold for * vs ns. P-values at or below this are labeled *; above are labeled ns. Default 0.05. See sig.levels for the multi-star thresholds.

sig.levels

Named numeric vector; upper p-value bounds for multi-star significance symbols. Names are the displayed symbols and values are the thresholds. Default c("****" = 0.0001, "***" = 0.001, "**" = 0.01). Any number of levels can be provided. Evaluated from smallest to largest threshold so the most significant symbol always wins.

Value

A data.frame with columns: group1, group2, p.value, p.adj, p.signif, test, facet_level, x_level (when group.by is set).

Author

Jared Andrews, Jacob Martin

Examples

compute_pairwise_stats(
    df = example_iris,
    x = "Species",
    y = "Sepal.Length",
    test = "wilcox.test"
)
#>       group1     group2      p.value        test facet_level x_level
#> 1     setosa versicolor 8.345827e-14 wilcox.test        <NA>    <NA>
#> 2     setosa  virginica 6.396699e-17 wilcox.test        <NA>    <NA>
#> 3 versicolor  virginica 5.869006e-07 wilcox.test        <NA>    <NA>
#>          p.adj p.signif p.adjust.method
#> 1 1.669165e-13     ****            holm
#> 2 1.919010e-16     ****            holm
#> 3 5.869006e-07     ****            holm

# Custom significance levels: only two-star tiers, lower threshold for *
compute_pairwise_stats(
    df = example_iris,
    x = "Species",
    y = "Sepal.Length",
    test = "wilcox.test",
    sig.threshold = 0.01,
    sig.levels = c("**" = 0.001, "***" = 0.0001)
)
#>       group1     group2      p.value        test facet_level x_level
#> 1     setosa versicolor 8.345827e-14 wilcox.test        <NA>    <NA>
#> 2     setosa  virginica 6.396699e-17 wilcox.test        <NA>    <NA>
#> 3 versicolor  virginica 5.869006e-07 wilcox.test        <NA>    <NA>
#>          p.adj p.signif p.adjust.method
#> 1 1.669165e-13      ***            holm
#> 2 1.919010e-16      ***            holm
#> 3 5.869006e-07      ***            holm