在R survey包的svyby中为svymean设置自定义置信水平

我使用R的survey和tidyverse包，基于mtcars数据集编写代码，按组计算多个变量的加权均值及置信区间。原有代码可正常生成默认95%置信区间，但尝试将置信区间改为0.99时出现错误。

原正常代码 #

library(survey)
library(tidyverse)
var_list = c("wt","qsec")
  
svy_design <- svydesign(
  ids = ~1, 
  data = mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit(),
  weights =  mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit() |> select(mpg)
)

lapply(var_list, function( x ) svyby(as.formula( paste0( "~" , x ) ) ,
                     by = ~cyl, 
                     design = svy_design, 
                     FUN = svymean,
                     keep.names = FALSE, vartype = "ci")) %>% 
  bind_rows() |> relocate(wt, .after = last_col()) %>%  pivot_longer(!c(cyl,ci_l,ci_u),names_to = "var",values_to = "mean",values_drop_na = TRUE)

错误代码及报错信息 #

错误代码中尝试直接传递FUN = svymean(level = 0.99)，触发如下报错：

Error in svymean(level = 0.99) : 
  argument "design" is missing, with no default

问题原因 #

错误的核心是不能直接在FUN参数中调用带参数的svymean——这样会立即执行svymean函数，但此时缺少design等必要参数，导致报错。正确的做法是将svymean的附加参数（如level）直接作为svyby的额外参数传递，或者用匿名函数包装svymean并指定参数。

正确解决方案 #

方法1：直接在svyby中传递level参数

svyby支持将FUN所需的额外参数直接作为自身的参数传入，无需修改FUN的写法：

library(survey)
library(tidyverse)
var_list = c("wt","qsec")
  
svy_design <- svydesign(
  ids = ~1, 
  data = mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit(),
  weights =  mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit() |> select(mpg)
)

lapply(var_list, function( x ) svyby(as.formula( paste0( "~" , x ) ) ,
                     by = ~cyl, 
                     design = svy_design, 
                     FUN = svymean,
                     keep.names = FALSE, 
                     vartype = "ci",
                     level = 0.99)) %>%  # 直接添加level参数指定置信水平
  bind_rows() |> relocate(wt, .after = last_col()) %>%  pivot_longer(!c(cyl,ci_l,ci_u),names_to = "var",values_to = "mean",values_drop_na = TRUE)

方法2：用匿名函数包装svymean

如果需要更灵活的参数控制，可以用匿名函数包裹svymean，明确指定参数：

library(survey)
library(tidyverse)
var_list = c("wt","qsec")
  
svy_design <- svydesign(
  ids = ~1, 
  data = mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit(),
  weights =  mtcars |> 
    dplyr::select(cyl,all_of(var_list),mpg) |> 
    na.omit() |> select(mpg)
)

lapply(var_list, function( x ) svyby(as.formula( paste0( "~" , x ) ) ,
                     by = ~cyl, 
                     design = svy_design, 
                     FUN = function(formula, design) svymean(formula, design, level = 0.99),  # 匿名函数包装并指定level
                     keep.names = FALSE, 
                     vartype = "ci")) %>% 
  bind_rows() |> relocate(wt, .after = last_col()) %>%  pivot_longer(!c(cyl,ci_l,ci_u),names_to = "var",values_to = "mean",values_drop_na = TRUE)

内容的提问来源于stack exchange，提问作者flâneur