使用ggplot绘制数据集计数差异及geom_count自定义方法咨询

Hey there! Let's tackle your R visualization questions step by step—both comparing count differences across groups and adjusting the stats behind geom_count are totally doable with tidyverse tools.

1. Visualizing Count Differences Between Groups #

Instead of just plotting counts side by side, the cleanest way to show differences is to precompute the gap between your two datasets first. Here's a step-by-step example with simulated data (you can swap in your actual datasets):

library(tidyverse)

# Simulate your two datasets (replace with your real data!)
set.seed(123)
df1 <- tibble(
  gender = sample(c("Male", "Female"), 100, replace = TRUE),
  marital_status = sample(c("Married", "Unmarried"), 100, replace = TRUE)
)
df2 <- tibble(
  gender = sample(c("Male", "Female"), 120, replace = TRUE),
  marital_status = sample(c("Married", "Unmarried"), 120, replace = TRUE)
)

# Combine datasets, calculate counts per group, then compute differences
combined_diff <- bind_rows(
  df1 %>% mutate(dataset = "Dataset 1"),
  df2 %>% mutate(dataset = "Dataset 2")
) %>%
  count(gender, marital_status, dataset) %>%
  pivot_wider(
    names_from = dataset,
    values_from = n,
    values_fill = 0  # Handle groups with zero counts in one dataset
  ) %>%
  mutate(count_difference = `Dataset 2` - `Dataset 1`)

# Plot the differences
ggplot(combined_diff, aes(x = interaction(gender, marital_status), y = count_difference)) +
  geom_col(fill = "#2980b9", alpha = 0.7) +
  labs(
    x = "Group Combination",
    y = "Count Difference (Dataset 2 - Dataset 1)",
    title = "How Counts Differ Between Datasets by Group"
  ) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

This gives you a clear bar chart where positive values mean Dataset 2 has more observations in that group, and negative values mean Dataset 1 has more.

2. Customizing geom_count for Proportions or Predefined Differences #

geom_count is built around counting observations, but you don't have to stick with its default behavior. Here are two common alternatives:

Option 1: Plot Proportions Instead of Counts #

You can either precompute proportions (most flexible) or use stat_count directly to calculate them on the fly:

Precomputed Proportions (Recommended)

proportion_data <- bind_rows(
  df1 %>% mutate(dataset = "Dataset 1"),
  df2 %>% mutate(dataset = "Dataset 2")
) %>%
  group_by(dataset, gender, marital_status) %>%
  summarize(total = n(), .groups = "drop") %>%
  group_by(dataset) %>%
  mutate(proportion = total / sum(total)) %>%
  ungroup()

# Plot proportions side by side
ggplot(proportion_data, aes(x = interaction(gender, marital_status), y = proportion, fill = dataset)) +
  geom_bar(stat = "identity", position = position_dodge(width = 0.8)) +
  scale_y_continuous(labels = scales::percent) +
  labs(
    x = "Group Combination",
    y = "Proportion of Total Observations",
    title = "Proportion Comparison Across Datasets",
    fill = "Dataset"
  ) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Using stat_count Directly

If you want to skip precomputing, you can tell stat_count to calculate proportions instead of raw counts:

ggplot(bind_rows(df1 %>% mutate(dataset = "Dataset 1"), df2 %>% mutate(dataset = "Dataset 2")),
       aes(x = interaction(gender, marital_status), y = stat(prop), group = dataset, fill = dataset)) +
  geom_bar(position = position_dodge(width = 0.8), stat = "count") +
  scale_y_continuous(labels = scales::percent) +
  labs(
    x = "Group Combination",
    y = "Proportion",
    title = "Proportion by Group & Dataset",
    fill = "Dataset"
  ) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Option 2: Plot Predefined Differences #

If you've already calculated custom differences (not just count gaps), you can use geom_point or geom_col to visualize them directly—no need to use geom_count at all. For example, if you have a precomputed custom_diff column in your data:

# Example with custom differences (replace with your own calculated values)
custom_diff_data <- combined_diff %>%
  mutate(custom_diff = count_difference / max(abs(count_difference)))  # Normalized difference

ggplot(custom_diff_data, aes(x = interaction(gender, marital_status), y = custom_diff)) +
  geom_point(size = 5, color = "#e74c3c") +
  geom_hline(yintercept = 0, linetype = "dashed", color = "gray50") +
  labs(
    x = "Group Combination",
    y = "Normalized Custom Difference",
    title = "Custom Difference Metric Across Groups"
  ) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

内容的提问来源于stack exchange，提问作者Joe Pemberton