rstatix

You can install rstatix from CRAN or the latest developmental version from GitHub. For easy data visualization, it is recommended to load ggpubr alongside it.

To install from CRAN:

install.packages("rstatix")

To install the latest developmental version from GitHub:

if(!require(devtools)) install.packages("devtools")
devtools::install_github("kassambara/rstatix")

To load the packages in your R session:

library(rstatix)
library(ggpubr)

Once a correlation matrix is created using cor_mat(), you can reshape, subset, and visualize it.

Reshaping:

• cor_gather(): Melt a matrix into long format.
• cor_spread(): Spread long format back to wide format.
• cor_reorder(): Reorder matrix using hierarchical clustering.

Subsetting:

• pull_lower_triangle(), pull_upper_triangle(): Extract triangular parts.
• replace_lower_triangle(), replace_upper_triangle(): Replace triangular parts.
• cor_select(): Subset by variable names.

Visualization:

• cor_plot(): Visualize using base R plots.
• cor_as_symbols(): Replace coefficients with symbols (e.g., *, +, .).
• cor_mark_significant(): Add significance levels to the matrix.

Example workflow:

cor.mat %>%
  cor_reorder() %>%
  pull_lower_triangle() %>% 
  cor_plot()

To visualize statistical results on plots (typically ggplot2 objects), use stat_pvalue_manual() from the ggpubr package in conjunction with rstatix test results. This allows you to pass p-values, significance levels, or custom labels directly onto the plot.

# Example: Adding p-values to a boxplot
stat.test <- df %>% t_test(len ~ supp)

p <- ggboxplot(df, x = "supp", y = "len")
p + stat_pvalue_manual(stat.test, label = "p", y.position = 35)

# Example: Customizing labels with glue
p + stat_pvalue_manual(stat.test, label = "T-test, p = {p}", y.position = 36)

You can visualize test results on plots using stat_pvalue_manual() from the ggpubr integration.

Manual p-value placement:

p <- ggboxplot(df, x = "supp", y = "len", color = "supp", palette = "jco", ylim = c(0,40))
stat.test <- df %>% t_test(len ~ supp, paired = FALSE)
p + stat_pvalue_manual(stat.test, label = "p", y.position = 35)

Customizing labels with glue:

p + stat_pvalue_manual(stat.test, label = "T-test, p = {p}", y.position = 36)

Using significance symbols:

p + stat_pvalue_manual(stat.test, label = "p.adj.signif", y.position = 35)

Removing brackets:

p + stat_pvalue_manual(stat.test, label = "p.adj.signif", y.position = 35, remove.bracket = TRUE)

The anova_test() function is a wrapper around car::Anova() used to perform various ANOVA tests, including independent measures, repeated measures, and mixed ANOVA. It returns a list containing the ANOVA table and assumption checks (like Mauchly's test for sphericity).

One-way ANOVA:

df %>% anova_test(len ~ dose)

Two-way ANOVA:

df %>% anova_test(len ~ supp*dose)

Two-way repeated measures ANOVA:

df %>% anova_test(dv = len, wid = id, within = c(supp, dose))

Using a model object:

.my.model <- lm(yield ~ block + N*P*K, npk)
anova_test(.my.model)

The package provides a suite of tools for correlation analysis.

Correlation test between two variables:

mydata %>% cor_test(wt, mpg, method = "pearson")

Correlation of one variable against all others:

mydata %>% cor_test(mpg, method = "pearson")

Pairwise correlation test between all variables:

mydata %>% cor_test(method = "pearson")

Compute a correlation matrix:

cor.mat <- mydata %>% cor_mat()

Extract p-values from a correlation matrix:

cor.mat %>% cor_get_pval()

The t_test() function performs one-sample, two-sample, and pairwise t-tests. It is pipe-friendly and returns results as a tidy data frame.

One-sample test (comparing a sample mean to a known mu):

# Compare len to mu = 0
df %>% t_test(len ~ 1, mu = 0)

Two-sample independent test:

df %>% t_test(len ~ supp, paired = FALSE)

Paired samples test:

df %>% t_test(len ~ supp, paired = TRUE)

Multiple pairwise comparisons (when grouping variable has >2 categories):

df %>% t_test(len ~ dose)

Comparisons against a reference group:

df %>% t_test(len ~ dose, ref.group = "0.5")

Comparisons against the base-mean (all groups):

df %>% t_test(len ~ dose, ref.group = "all")

The get_summary_stats() function computes summary statistics for one or multiple numeric variables. It supports grouped data and different types of summaries (e.g., type = "common" or type = "mean_sd").

Example for common statistics across variables:

iris %>% 
  get_summary_stats(Sepal.Length, Sepal.Width, type = "common")

Example for grouped data:

iris %>%
  group_by(Species) %>% 
  get_summary_stats(Sepal.Length, type = "mean_sd")

In version 1.1.0, cramer_v() changed its default behavior regarding Yates' continuity correction. The correct argument now defaults to FALSE. This change ensures the function returns the standard sqrt(chi2 / (N * (k - 1))) for 2x2 tables, matching the behavior of DescTools::CramerV() and effectsize::cramers_v(adjust = FALSE).

If you require the previous behavior (applying Yates' correction), you must explicitly set correct = TRUE.

# Standard behavior (new default)
cramer_v(data, correct = FALSE)

# Previous behavior (manual override)
cramer_v(data, correct = TRUE)

As of version 1.1.0, eta_squared() and partial_eta_squared() have been updated to correctly handle tables generated by car::Anova(type = 3). The functions now exclude the (Intercept) row from calculations. Previously, including the intercept inflated the denominator and produced incorrect values.

Note that inputs from standard aov or stats::anova() do not carry an intercept row and remain unaffected by this change.

The cor_test() function performs correlation tests between variables using Pearson, Spearman, or Kendall methods. It can be used for a single pair, one variable against all others, or pairwise correlations across all variables in a data frame.

# Correlation between two specific variables
mydata %>% cor_test(wt, mpg, method = "pearson")

# Correlation of one variable against all others
mydata %>% cor_test(mpg, method = "pearson")

# Pairwise correlation test between all variables
mydata %>% cor_test(method = "pearson")