如何基于data.table的group列对vdem_media_bias执行全组合t检验？

需求：对data.table对象trust_news按group列分组，完成vdem_media_bias列的所有组间组合t检验。此前尝试的代码存在结果为空或不完整的问题，最终通过循环索引得到全部6组结果，现寻求更简洁高效的实现方法。

数据与依赖包加载 #

library(tidyr)
library(dplyr)
library(purrr)
library(broom)
library(data.table)

trust_news <- data.table(group  =   c(  1,  2,  3,  5,  1,  2,  3,  5   ),
                         mean   =   c(  2.68,   2.8,    3.22,   2.96,   2.16739457271907,   2.52420459002153,   2.41568263469884,   3.203042892 ),
                         polity2    =   c(  11.0065624592239,   11.5173963819866,   10.9690935682988,   7.4505271137667,    12.5039982916238,   10.2183026540883,   11.5528854439886,   8.951439457 ),
                         web    =   c(  87.2661,    94.8967,    89.7391,    74.3872,    99.429653354408,    86.651554228033,    88.7724076776387,   82.71520214 ),
                         rsf    =   c(  16.4114055821914,   16.7094853023036,   29.5703741065912,   23.8887382488302,   15.6296493773739,   17.6154322886768,   28.7548757022008,   19.67113906 ),
                         civil_liberties    =   c(  0.0825935016192092, 0.775101009726579,  0.0505530810447113, 0.302537382248755,  0.4563913292031,    0.759215853597412,  0.172427084833363,  0.772689046 ),
                         freedom_of_expression  =   c(  0.828613177733605,  0.312013322438901,  0.788670591087342,  0.192519523553583,  0.599717513191781,  0.682591150447179,  0.0926750357710926, 0.777797913 ),
                         vdem_gov_censorship_effort =   c(  0.138795469801477,  0.609063884646593,  0.726359178306941,  0.889397969637812,  0.935589374225135,  0.5665276966565,    0.939565934607974,  0.383976181 ),
                         vdem_self_censorship_effort    =   c(  0.396389956859171,  0.368909768892093,  0.795891881421097,  0.635097442795347,  0.818482469632694,  0.236225542529557,  0.615626792133531,  0.035815441 ),
                         vdem_freedom_of_expression =   c(  0.270293421459824,  0.0390741902396886, 0.0430150919275922, 0.451150387704898,  0.477508000166212,  0.841653725881213,  0.885557800737999,  0.205165414 ),
                         ciri_freedom_of_speech_and_press   =   c(  0.316629196080121,  0.473252307510063,  0.782615827564335,  0.600531400120434,  0.710143072640082,  0.0105031020419695, 0.348075169050439,  0.282137779 ),
                         media_integrity    =   c(  0.763527559879246,  0.32401086386208,   0.72357993656577,   0.00400993472498878,    0.804790656378296,  0.752078255508056,  0.12450323855886,   0.784842369 ),
                         vdem_critical_press    =   c(  0.381934476222942,  0.34180259848916,   0.446445782446012,  0.101622162644166,  0.69273703605482,   0.674447773690032,  0.593981628888233,  0.942639773 ),
                         vdem_media_perspective =   c(  0.0121195504624632, 0.55722585253704,   0.759939717080726,  0.422261093911245,  0.425138202329788,  0.0440513028866172, 0.128858393736017,  0.242947985 ),
                         vdem_media_bias    =   c(  0.0386091435387627, 0.149227444223962,  0.109819706451143,  0.79759885890908,   0.421084356961021,  0.269048253813396,  0.661346417250035,  0.051339441 ),
                         vdem_media_corruption  =   c(  0.188072265767419,  0.70446056263869,   0.988731124980027,  0.659578974943737,  0.924096790161402,  0.465803086735264,  0.0389306657957701, 0.865528541 ),
                         vdem_media_freedom =   c(  0.710150348587615,  0.253187878925348,  0.436362652570953,  0.471900738683268,  0.302672853555399,  0.88559075326591,   0.786474651280836,  0.234080239 )
                         )

过往问题回顾 #

• 初始使用names(combn(condition_vec, 2, FUN = paste))生成列名向量，导致循环无法正确索引分组，结果为空
• 修改为as.numeric(combn(...))后，因combn返回矩阵结构转数值时丢失部分组合，仅得到4组结果
• 最终通过遍历1:ncol(col_vec)的索引方式得到全部6组结果，但代码冗余，不够直观

优化实现方案 #

利用combn生成所有组对，结合purrr::map_dfr批量执行t检验，并用broom::tidy统一整理结果，代码简洁且易维护：

# 获取所有唯一分组
unique_groups <- unique(trust_news$group)

# 生成所有不重复的组间组合（列表形式）
group_pairs <- combn(unique_groups, 2, simplify = FALSE)

# 批量执行t检验并整理结果
ttest_results <- map_dfr(group_pairs, function(pair) {
  # 提取两组的vdem_media_bias数据
  group1_data <- trust_news[group == pair[1], vdem_media_bias]
  group2_data <- trust_news[group == pair[2], vdem_media_bias]
  
  # 执行独立样本t检验
  ttest_output <- t.test(group1_data, group2_data)
  
  # 整理结果并添加分组标识
  tidy(ttest_output) %>%
    mutate(group_1 = pair[1], group_2 = pair[2]) %>%
    select(group_1, group_2, everything())
})

# 查看结果
print(ttest_results)

方案优势 #

• 无需手动管理循环索引，自动覆盖所有组间组合
• map_dfr自动将所有t检验结果合并为一个data.frame，无需手动绑定
• broom::tidy将t检验的统计结果（p值、置信区间、统计量等）结构化输出，便于后续分析
• 代码逻辑清晰，可读性强，易于扩展到其他变量的组间检验

内容的提问来源于stack exchange，提问作者Chris