如何在R语言中为带权重的调查数据应用权重？

我有带权重的调查数据，权重代表样本中单个个体对应的总体人数。现在不对数据加权的话，结果没法准确代表总体，想找个简单的方法把权重纳入分析里。是不是要把观测值乘以权重？我看了相关视频，但视频里都是分层数据，我的是个体数据，没找到解决办法，求帮忙。

原始示例代码 #

ID <-  c(1,2,3,4,5,6,7,8,9,10,11,12,
     13,14,15,16,17,18,19,20,21,22,23,24,
     25,26,27,28,29,30,31,32,33,34,35,36)
year <- c(1980,1980,1980,1981,1982,1982,1980,1980,1981,1981,1982,1982,
      1980,1980,1980,1981,1982,1982,1980,1980,1981,1981,1982,1982,
      1980,1980,1980,1981,1982,1982,1980,1980,1981,1981,1982,1982)
city <- c("NY","NY","NY","NY","NY","NY","NY","NY","NY","NY","NY","NY",
      "NY","NY","NY","NY","NY","NY","NY","NY","NY","NY","NY","NY",
      "CA","CA","CA","CA","CA","CA","CA","CA","CA","CA","CA","CA")
district <- c(1,2,1,2,1,2,1,2,1,2,1,2,
          1,2,1,2,1,2,1,2,1,2,1,2,
          1,2,1,2,1,2,1,2,1,2,1,2)
weight <- c(100,17,25,1,100,52,10,5,90,10,10,15,
        13,1,25,1,6,52,10,5,90,7,10,15,
        1,2,3,4,5,6,10,20,3,40,50,6)

df <- data.frame(ID,year,city,district,weight)

df$year <- as.factor(df$year)
df$district <- as.factor(df$district)

print(df)


# aggregate count by year, city and district
a <- aggregate(ID ~ year + city + district, 
            data = df,  
            FUN = length) 

colnames(a)[colnames(a) == 'ID'] <- 'numObs'

# add new Total column, populate total by year
a[a$year == 1980, "TotalforYear"] <-  with(a, sum(numObs[year == 1980]))
a[a$year == 1981, "TotalforYear"] <-  with(a, sum(numObs[year == 1981]))  
a[a$year == 1982, "TotalforYear"] <-  with(a, sum(numObs[year == 1982]))

# add new column called share and initialize it
a[ , 'share'] <- 0
a$share = (a$numObs/a$TotalforYear)

# add group column
a[a$district == 1 & a$city == 'NY', "group"] <- 'NY district1'
a[a$district == 2 & a$city == 'NY', "group"] <- 'NY district2'
a[a$district == 1 & a$city == 'CA', "group"] <- 'CA district1'
a[a$district == 2 & a$city == 'CA', "group"] <- 'CA district2'

a

g1 <- ggplot(a, aes(x=year, y=share, group=group)) +
  geom_line(aes(linetype=group), color = 'red', size = 0.8) +
  geom_point(aes(shape=group), size = 2) +
  theme_classic() +  
  theme(legend.position="bottom",
    panel.grid.major.x = element_line( linewidth=.1, color="grey93" ), 
    panel.grid.major.y = element_line( linewidth=.1, color="grey93" ) )       +
  ylab("Share") +
  scale_linetype_manual(values=c("solid", "dotted", 'dashed', 'longdash')) +
  theme(legend.title=element_blank()) 
g1

加权处理解决方案 #

核心逻辑 #

你的思路是对的，加权的本质就是用个体权重替代原始观测计数，通过求和分组内的权重得到该分组对应的总体规模，再基于这个规模计算占比，结果就能准确代表总体。不需要手动给每个观测值乘权重，直接对权重求和效率更高。

具体修改步骤 #

1. 计算分组加权总和

替换原代码中的aggregate步骤，不再统计原始观测数，而是对每个year+city+district分组的weight求和，得到该分组对应的总体人数：

# 计算加权后的分组总体人数
a_weighted <- aggregate(weight ~ year + city + district, 
                        data = df,  
                        FUN = sum) 

colnames(a_weighted)[colnames(a_weighted) == 'weight'] <- 'weighted_total'

2. 计算年度加权总人数

基于分组加权总和，计算每年的总体总人数：

# 计算每年的加权总人数（用ave函数更简洁）
a_weighted$TotalforYear <- with(a_weighted, ave(weighted_total, year, FUN = sum))

3. 计算加权占比

用分组加权总和除以年度加权总数，得到能代表总体的占比：

# 计算加权占比
a_weighted$share <- a_weighted$weighted_total / a_weighted$TotalforYear

4. 添加分组标签（和原代码逻辑一致）

# 添加group列
a_weighted[a_weighted$district == 1 & a_weighted$city == 'NY', "group"] <- 'NY district1'
a_weighted[a_weighted$district == 2 & a_weighted$city == 'NY', "group"] <- 'NY district2'
a_weighted[a_weighted$district == 1 & a_weighted$city == 'CA', "group"] <- 'CA district1'
a_weighted[a_weighted$district == 2 & a_weighted$city == 'CA', "group"] <- 'CA district2'

5. 绘制加权后的图表

直接使用加权后的数据集绘图，代码逻辑和原图表一致，仅修改数据源：

g1_weighted <- ggplot(a_weighted, aes(x=year, y=share, group=group)) +
  geom_line(aes(linetype=group), color = 'red', size = 0.8) +
  geom_point(aes(shape=group), size = 2) +
  theme_classic() +  
  theme(legend.position="bottom",
        panel.grid.major.x = element_line( linewidth=.1, color="grey93" ), 
        panel.grid.major.y = element_line( linewidth=.1, color="grey93" ) )       +
  ylab("加权占比") +
  scale_linetype_manual(values=c("solid", "dotted", 'dashed', 'longdash')) +
  theme(legend.title=element_blank()) 
g1_weighted

关键说明 #

• 这个方法完全适配你的个体数据，不需要额外的分层信息，只要每个样本都有对应的权重即可。
• 加权后的占比反映的是总体中各分组的真实比例，而非样本中的比例，解决了样本代表性偏差的问题。

内容的提问来源于stack exchange，提问作者ithoughtso