对数据帧中的大量组合执行KruskalWallis测试

发布于01月07日

我在R中有一个DF，有50个唯一的A和B组合. 对于A和B的每一种组合，我想执行Kruskal-Wallis测试:kruskal.test(D,C,data = df)

我想测试A和B需要哪些组合来拒绝零假设.

我如何才能做到这一点，而不作出单独的测试，每一个组合？我的数据示例如下

A     B       C     D
mix1 size1    1     0.2
mix1 size1    2     0.15
mix1 size1    3     0.22
mix1 size1    4     0.215
mix2 size1    1     0.2
mix2 size1    2     0.15
mix2 size1    3     0.2
mix2 size1    4     0.15
mix2 size2    1     0.21
mix2 size2    2     0.11
mix2 size2    3     0.23
mix2 size2    4     0.615
...
mix22 size1    1     0.01
mix22 size1    2     0.18
mix22 size1    3     0.7
mix22 size1    4     0.17

我的预期输出是df/table，其中包含A和B的每种组合的Kruskal-Wallis检验的p值.

A     B    P
mix1 size1 0.005
mix2 size1 0.211

也许是用*apply大家庭里的东西？

推荐答案

在by中使用interaction.

> by(dat, with(dat, interaction(A, B)), \(x) {
+   with(x, kruskal.test(D, C))[c('statistic', 'parameter', 'p.value')]
+ }) |> do.call(what='rbind')
            statistic parameter p.value
mix1.size1  2379      3         0      
mix2.size1  2620      3         0      
mix22.size1 2460      3         0      
mix2.size2  2537      3         0

Data:个

> dput(dat)
structure(list(A = c("mix1", "mix1", "mix1", "mix1", "mix2", 
"mix2", "mix2", "mix2", "mix2", "mix2", "mix2", "mix2", "mix22", 
"mix22", "mix22", "mix22"), B = c("size1", "size1", "size1", 
"size1", "size1", "size1", "size1", "size1", "size2", "size2", 
"size2", "size2", "size1", "size1", "size1", "size1"), C = c(1L, 
2L, 3L, 4L, 1L, 2L, 3L, 4L, 1L, 2L, 3L, 4L, 1L, 2L, 3L, 4L), 
    D = c(0.2, 0.15, 0.22, 0.215, 0.2, 0.15, 0.2, 0.15, 0.21, 
    0.11, 0.23, 0.615, 0.01, 0.18, 0.7, 0.17)), class = "data.frame", row.names = c(NA, 
-16L))