ggplot2技术求助：facet_wrap(scales="free")下stat_compare_means()添加p值报错

Hey there! Let's break down and fix this issue you're hitting with ggplot2 and stat_compare_means() when using scales="free".

What's causing the error? #

When you add scales="free" to your facet (like facet_wrap() or facet_grid()), each subplot only shows the categories that have actual data for that subset. If any of your facets ends up with only one group (e.g., a specific Host + Clade combination where there's only one Phage type), stat_compare_means() can't calculate a p-value—it needs at least two groups to compare. That's why you get the group1, p must resolve to integer column positions, not NULL warning/error.

Without scales="free", ggplot keeps all categories on every facet (even if some have no data), so stat_compare_means() still sees the full group list and can run calculations (even if they're meaningless for empty groups).

Fixes to try #

1. Filter out subsets with only one group first

First, identify which facets have insufficient groups, then remove those subsets from your data:

library(dplyr)

# Check which Host/Clade combinations have fewer than 2 Phage types
single_group_facets <- burst.data %>%
  group_by(Host, Clade) %>%
  summarise(phage_count = n_distinct(Phage)) %>%
  filter(phage_count < 2)

# Keep only subsets with 2+ Phage types
filtered_data <- burst.data %>%
  anti_join(single_group_facets, by = c("Host", "Clade"))

# Now plot with scales="free"
ggplot(filtered_data, aes(x = Phage, y = Burst.Size)) +
  geom_boxplot() +
  facet_wrap(~ Host + Clade, scales = "free") +
  stat_compare_means() # Add method="t.test" or method="wilcox" if needed

2. Pre-calculate p-values and add them manually (more flexible)

If you don't want to remove facets, precompute p-values only for subsets with valid groups, then add them as text:

# Calculate p-values for valid subsets
precomputed_pvals <- burst.data %>%
  group_by(Host, Clade) %>%
  mutate(phage_count = n_distinct(Phage)) %>%
  filter(phage_count >= 2) %>%
  # Use t-test (swap with wilcox.test if non-parametric)
  do(tidy(t.test(Burst.Size ~ Phage, data = .))) %>%
  ungroup() %>%
  mutate(p_label = paste0("p = ", round(p.value, 3)))

# Plot with all facets, adding p-values only where possible
ggplot(burst.data, aes(x = Phage, y = Burst.Size)) +
  geom_boxplot() +
  facet_wrap(~ Host + Clade, scales = "free") +
  geom_text(
    data = precomputed_pvals,
    aes(x = 1.5, y = max(Burst.Size, na.rm = T) * 1.1, label = p_label),
    inherit.aes = FALSE,
    size = 3
  )

Adjust the x and y values in geom_text() to position the p-label where it looks best for your plots.

3. Explicitly define the group in stat_compare_means()

Sometimes explicitly specifying the group variable helps, though it won't fix the single-group facet issue:

ggplot(burst.data, aes(x = Phage, y = Burst.Size, group = Phage)) +
  geom_boxplot() +
  facet_wrap(~ Host + Clade, scales = "free") +
  stat_compare_means(group = Phage)

内容的提问来源于stack exchange，提问作者Ilia