ggplot2中fill选项导致直方图密度值异常的问题求助

Ah, I've run into this exact issue before! The root cause is how ggplot computes the ..density.. statistic when you add a fill grouping: by default, it calculates density per group instead of across the entire dataset. That's why your filled histogram has inflated density values for the b=FALSE group—its density is scaled to the size of its own subset (~84% of your data) rather than the full 10,000 observations.

Here's How to Fix It #

We need to force the density calculation to use the total number of observations from the entire dataset, not just each subgroup. There are two straightforward methods:

Method 1: Adjust the Y-Aesthetic Directly in geom_histogram

Replace the default ..density.. with a custom calculation that uses the total count of all observations (sum(..count..)) instead of per-group counts:

set.seed(42)
x <- rnorm(10000,0,1)
df <- data.frame(x=x, b=x>1)

# Correct filled density histogram
ggplot(df, aes(x = x)) +
  geom_histogram(
    aes(fill = b, y = ..count../sum(..count..)/..width..),
    position = "stack"  # Keeps bins stacked like your original filled plot
  )

• ..count../sum(..count..) gives the proportion of total observations in each bin segment
• Dividing by ..width.. converts that proportion to density (since density = proportion / bin width)

Method 2: Precompute Stats for Full Control

If you prefer more transparency, precompute the bin counts and densities manually using dplyr, then plot with geom_col:

library(dplyr)

# Match ggplot's default bin width (30 bins total)
bin_width <- diff(range(df$x)) / 30

# Calculate grouped bin stats
df_binned <- df %>%
  # Create bins aligned with ggplot's default behavior
  mutate(x_bin = cut(x, breaks = seq(min(x)-bin_width/2, max(x)+bin_width/2, by = bin_width))) %>%
  # Count observations per bin and group
  group_by(x_bin, b) %>%
  summarise(count = n(), .groups = "drop") %>%
  # Compute density using total dataset size
  mutate(
    total_obs = sum(count),
    density = count / total_obs / bin_width,
    # Extract midpoint of each bin for plotting
    x_mid = as.numeric(substr(x_bin, 2, nchar(x_bin)-1)) %>% 
      strsplit(",") %>% 
      sapply(function(y) mean(as.numeric(y)))
  )

# Plot the precomputed data
ggplot(df_binned, aes(x = x_mid, y = density, fill = b)) +
  geom_col(position = "stack", width = bin_width)

Why Your Original Code Failed #

When you use fill=b in the aes, ggplot automatically groups the data by b. The default ..density.. statistic for grouped data is:
..count.. / (n_group * ..width..)
Where n_group is the total number of observations in that subgroup (e.g., ~8413 for b=FALSE).

In contrast, the ungrouped histogram uses:
..count.. / (n_total * ..width..)
Where n_total is all 10,000 observations. This mismatch is why your filled plot had inflated density values.

Verification #

After applying either fix, your filled histogram will:

• Match the shape and scale of your ungrouped density histogram
• Show correct density values (peak around ~0.4, matching the standard normal distribution's maximum density)
• Keep the color-coded bins you wanted

内容的提问来源于stack exchange，提问作者dixhom