在R语言中绘制多阶段节点标签相同的桑基图

原代码核心问题 #

1. 数据转换逻辑错误：pivot_stages_longer需要的是每个观测（ID）在各阶段（年份）的状态数据，而非先聚合后的比例数据。原代码提前聚合各年份风险等级占比，导致无法捕捉个体迁移路径，桑基图流量关系混乱。
2. 绘图映射参数错误：geom_sankeyedge和geom_sankeynode参数匹配不当，缺少必要的流量映射，且v_space设置未配合数据调整。

修正后的完整代码 #

library(ggsankeyfier)
library(dplyr)
library(ggplot2)
library(tidyr)

# 原始数据
df <- data.frame(
  ID = c(1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5),
  risk_level = c("High", "High", "High", 
                 "Low", "Low", "Very low", 
                 "Low", "Low", "Low", 
                 "Low", "Moderate", "Low", 
                 "Moderate", "High", "High"),
  Year = c(2022, 2023, 2024, 
           2022, 2023, 2024, 
           2022, 2023, 2024, 
           2022, 2023, 2024, 
           2022, 2023, 2024))

# 步骤1：转换为宽格式（每个ID一行，列对应各年份的风险等级）
df_wide <- df %>%
  pivot_wider(names_from = Year, values_from = risk_level)

# 步骤2：转换为桑基图所需的长格式
df_sankey <- pivot_stages_longer(
  df_wide,
  stages_from = c("2022", "2023", "2024"),  # 指定阶段（年份）列
  values_to = "risk_level",                 # 状态列名
  id_cols = ID                              # 观测ID列
)

# 步骤3：计算各迁移路径的比例（按起始年份分组）
df_sankey <- df_sankey %>%
  group_by(stage, node) %>%
  summarise(count = n(), .groups = "drop") %>%
  group_by(stage) %>%
  mutate(proportion = count / sum(count)) %>%
  ungroup()

# 步骤4：绘制桑基图
ggplot(df_sankey, aes(x = stage, y = proportion, group = node,
                      connector = connector, edge_id = edge_id, fill = node)) +
  geom_sankeyedge(color = "white", alpha = 0.7) +  # 流量边样式
  geom_sankeynode(color = "black", size = 0.2) +   # 节点样式
  scale_fill_viridis_d(option = "plasma") +        # 配色调整
  labs(x = "年份", y = "占比", fill = "风险等级") +
  theme_minimal() +
  theme(axis.text.x = element_text(size = 12),
        legend.position = "bottom")

关键说明 #

• 先转宽格式是让pivot_stages_longer正确识别每个ID的跨年份状态迁移，这是桑基图捕捉路径的核心。
• 比例计算放在格式转换后，确保每个迁移路径的占比基于对应年份的总观测数。
• 调整边和节点样式、添加中文标签，让图表更贴合需求。

内容的提问来源于stack exchange，提问作者ccc