如何为Kruskal-Wallis的Dunn事后检验添加紧凑字母显示至表格与图表

Here's how to generate CLDs for your post-hoc results and integrate them into a summary table and ggplot:

Step 1: Load Required Packages #

library(FSA)       # For dunnTest
library(dplyr)     # For data manipulation
library(multcompView)  # For generating CLDs
library(ggplot2)   # For plotting

Step 2: Run Kruskal-Wallis and Dunn Post-Hoc Test #

data <- iris

# Kruskal-Wallis test
kruskal <- kruskal.test(Petal.Width ~ Species, data = data)

# Dunn post-hoc test with Bonferroni adjustment
kruskal_ph <- dunnTest(Petal.Width ~ Species, data = data, method = "bonferroni")

Step 3: Extract Pairwise Comparisons and Generate CLD #

First, reshape the Dunn test results into a format compatible with multcompView:

# Extract pairwise comparisons table
comparisons <- kruskal_ph$res %>%
  select(Comparison, Z, P.adj) %>%
  separate(Comparison, into = c("group1", "group2"), sep = " - ") %>%
  mutate(Sig = case_when(
    P.adj > 0.05 ~ "ns",
    P.adj <= 0.01 ~ "**",
    P.adj <= 0.05 ~ "*"
  ))

# Create a square p-value matrix for multcompView
groups <- unique(c(comparisons$group1, comparisons$group2))
p_matrix <- matrix(1, nrow = length(groups), ncol = length(groups), dimnames = list(groups, groups))

# Fill matrix with adjusted p-values
for (i in 1:nrow(comparisons)) {
  g1 <- comparisons$group1[i]
  g2 <- comparisons$group2[i]
  p_matrix[g1, g2] <- comparisons$P.adj[i]
  p_matrix[g2, g1] <- comparisons$P.adj[i]
}

# Generate CLD: groups with the same letter are not significantly different
cld <- multcompLetters(p_matrix, threshold = 0.05)$Letters
cld_df <- data.frame(Species = names(cld), CLD = unname(cld))

Step 4: Create Summary Table with CLD #

Combine group-level statistics with the CLD letters:

summary_table <- data %>%
  group_by(Species) %>%
  summarise(
    Mean = round(mean(Petal.Width), 3),
    SD = round(sd(Petal.Width), 3),
    Median = round(median(Petal.Width), 3)
  ) %>%
  left_join(cld_df, by = "Species")

# Add overall Kruskal-Wallis results as a footnote (optional)
cat("Overall Kruskal-Wallis test: Chi-squared =", round(kruskal$statistic, 2), ", p-value =", round(kruskal$p.value, 4), "\n")
print(summary_table)

Step 5: Add CLD to ggplot #

Place the CLD letters above the mean bars:

# Prepare plot data with position for CLD letters
plot_data <- summary_table %>%
  mutate(y_pos = Mean + SD + 0.1)  # Adjust offset to avoid overlapping error bars

# Generate plot
ggplot(data, aes(x = Species, y = Petal.Width)) +
  geom_bar(stat = "summary", fun = "mean", aes(fill = Species), alpha = 0.7) +
  geom_errorbar(stat = "summary", fun.data = "mean_se", width = 0.2, color = "black") +
  geom_text(data = plot_data, aes(x = Species, y = y_pos, label = CLD), size = 5, fontface = "bold") +
  labs(
    title = "Petal Width by Species (Kruskal-Wallis Dunn Test)",
    x = "Species",
    y = "Petal Width (cm)"
  ) +
  theme_minimal() +
  theme(legend.position = "none")

Applying to Other Tests (e.g., Welch ANOVA pairwise.t.test) #

The same workflow applies to other post-hoc tests:

1. Run the test and extract pairwise p-values
2. Reshape into a square p-value matrix
3. Use multcompLetters() to generate CLDs
4. Merge with summary stats and add to plots

Example for Welch ANOVA:

# Welch ANOVA
welch <- oneway.test(Petal.Width ~ Species, data = data, var.equal = FALSE)

# Pairwise t-test without pooling variance
pairwise_welch <- pairwise.t.test(data$Petal.Width, data$Species, pool.sd = FALSE, p.adjust.method = "bonferroni")

# Extract p-value matrix
p_matrix_welch <- pairwise_welch$p.value

# Generate CLD
cld_welch <- multcompLetters(p_matrix_welch, threshold = 0.05)$Letters

内容的提问来源于stack exchange，提问作者MM1