r - 使用 lapply 或 map over columns 将 McNemar 测试应用于所有列组合

Question

我正在尝试将麦克尼马尔测试应用于我的数据的不同模型组合。我有 N 个模型，但这里我显示了 3 个。我可以执行以下操作。

mcnemar.test(x = d$mod1, y = d$mod2)
mcnemar.test(x = d$mod1, y = d$mod3)
mcnemar.test(x = d$mod2, y = d$mod3)

我希望将结果存储在数据框中，例如:

P 值:

      mod1    mod2    mod3
mod1   -    0.1306   0.7518
mod2   -       -     0.3711
mod3   -       -       -

数据:

d <- structure(list(id = 1:50, mod1 = c("Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Incorrect", "Incorrect", "Correct", "Correct", "Correct", 
"Correct", "Incorrect", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Incorrect", "Correct", "Incorrect", "Incorrect", 
"Correct", "Correct", "Correct", "Correct", "Correct"), mod2 = c("Correct", 
"Correct", "Correct", "Incorrect", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct"), mod3 = c("Correct", "Correct", "Incorrect", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Incorrect", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Incorrect", "Correct", "Correct", 
"Incorrect", "Correct", "Correct", "Correct", "Correct", "Correct", 
"Correct", "Correct", "Correct", "Correct")), row.names = c(NA, 
-50L), class = c("tbl_df", "tbl", "data.frame"))

Answer 1

我们可以使用base R中的outer

outer(d[-1], d[-1], FUN = Vectorize(function(x, y) mcnemar.test(x, y)$p.value))
#          mod1      mod2      mod3
#mod1       NaN 0.1305700 0.7518296
#mod2 0.1305700       NaN 0.3710934
#mod3 0.7518296 0.3710934       NaN

或者与嵌套sapply相同

sapply(d[-1], function(x) sapply(d[-1], function(y) mcnemar.test(x, y)$p.value))

---

或者使用tidyverse中的crossing/pivot_wider

library(dplyr)
library(tidyr)
library(purrr)
crossing(v1 = names(d)[-1], v2 = names(d)[-1]) %>% 
   rowwise() %>% 
   mutate(pval = mcnemar.test(d[[v1]], d[[v2]])$p.value) %>% 
   pivot_wider(names_from = v2, values_from = pval) %>% 
   column_to_rownames('v1')
#          mod1      mod2      mod3
#mod1       NaN 0.1305700 0.7518296
#mod2 0.1305700       NaN 0.3710934
#mod3 0.7518296 0.3710934       NaN

---

或者另一种选择是 reshape 为“长”格式，然后进行交叉

d %>% 
    pivot_longer(cols = -id) %>%
    select(-id) %>%
    nest_by(name) %>% 
    crossing(., set_names(., str_c(names(.), '_new'))) %>% 
    rowwise() %>%
    transmute(name, name_new, pvalue = mcnemar.test(data[[1]], data_new[[1]])$p.value) %>%
    ungroup %>%
    pivot_wider(names_from = name_new, values_from = pvalue) %>%
    column_to_rownames('name')
#          mod1      mod2      mod3
#mod1       NaN 0.1305700 0.7518296
#mod2 0.1305700       NaN 0.3710934
#mod3 0.7518296 0.3710934       NaN