dplyr多分组差异分析：添加置信区间的数据格式适配问题

Hey there! Let's work through this together. I see you're currently only reporting mean values for your analysis, and you want to add confidence intervals to make your results more robust. You know that linear regression with lm() can calculate group difference estimates and their CIs, but you're stuck getting your data into the right format. Let's break this down step by step.

我目前仅报告计算得出的均值，但希望为结果添加置信区间。若数据格式正确，我可通过线性回归lm()计算分组差异估计值及其置信区间，但当前难以将数据调整为正确格式。我通常会执行三类计算，核心是要回答诸如「组之间的差异是多少」这类问题。

Let's start with a common "wide format" fake dataset (since this is often the format people struggle with converting):

# 模拟宽格式数据：每一列代表一个分组，每一行代表一个样本的观测值
fake_data_wide <- data.frame(
  group_X = rnorm(40, mean = 8, sd = 1.5),
  group_Y = rnorm(40, mean = 10, sd = 1.5),
  group_Z = rnorm(40, mean = 9, sd = 1.5)
)

The lm() function requires long-format data—each row should represent a single observation, with one column for the group category and one column for the measured value. If your data is in wide format (like the example above), use the tidyr package to reshape it:

# 先安装tidyr（如果还没装）
# install.packages("tidyr")
library(tidyr)

# 宽格式转长格式
fake_data_long <- pivot_longer(
  data = fake_data_wide,
  cols = everything(),  # 选择所有列作为分组
  names_to = "group",   # 新的分组列名
  values_to = "value"   # 新的观测值列名
)

Once your data is in long format, you can fit the linear model and extract the confidence intervals easily. Let's use group_X as our reference group to compare against:

# 拟合线性回归模型
group_model <- lm(value ~ group, data = fake_data_long)

# 查看模型结果（包含分组差异估计值）
summary(group_model)

# 提取95%置信区间（默认）
confint(group_model)

# 如果需要自定义置信水平（比如90%）
confint(group_model, level = 0.9)

Bonus: Flexible group comparisons with emmeans #

If your three types of calculations involve pairwise comparisons between all groups, the emmeans package makes this straightforward:

# 安装并加载emmeans
# install.packages("emmeans")
library(emmeans)

# 计算每组均值的置信区间
emmeans(group_model, ~ group) %>% confint()

# 计算所有组间两两差异及置信区间
pairs(emmeans(group_model, ~ group)) %>% confint()

This should give you all the confidence intervals you need for your mean values and group differences!

内容的提问来源于stack exchange，提问作者Alex