r语言 - 如何编写对数据框中所有组/变量组合执行配对 t 检验的函数

Question

我有一个类似于下面创建的数据框data：

ID <- data.frame(ID=rep(c(12,122,242,329,595,130,145,245,654,878),each=5))
Var <- data.frame(Variable=c("Copper","Iron","Lead","Zinc","CaCO"))
n <- 10
Variable <- do.call("rbind",replicate(n,Var,simplify=F))
Location <- rep(c("Alpha","Beta","Gamma"), times=c(20,20,10))
Location <- data.frame(Location)
set.seed(1)
FirstPt<- data.frame(FirstPt=sample(1:100,50,replace=T))
LastPt <- data.frame(LastPt=sample(1:100,50,replace=T))
First3<- data.frame(First3=sample(1:100,50,replace=T))
First5<- data.frame(First5=sample(1:100,50,replace=T))
First7<- data.frame(First7=sample(1:100,50,replace=T))
First10<- data.frame(First10=sample(1:100,50,replace=T))
Last3<- data.frame(Last3=sample(1:100,50,replace=T))
Last5<- data.frame(Last5=sample(1:100,50,replace=T))
Last7<- data.frame(Last7=sample(1:100,50,replace=T))
Last10<- data.frame(Last10=sample(1:100,50,replace=T))
data <- cbind(ID,Location,Variable,FirstPt,LastPt,First3,First5,First7,
First10,Last3,Last5,Last7,Last10)

这可能是一个由两部分组成的问题，但我想编写一个函数，将所有相同的Variables分组(例如，所有Copper的观察值(，并在数字列的所有可能组合之间进行配对t检验(FirstPt：Last10(。我希望它在数据框中返回 p 值，如下所示：

Test                        P-Value
FirstPt.vs.LastPt             …
FirstPt.vs.First3             … 
ect...                        … 

这可能是第二个函数，但我也想在观察结果按Location分组后执行此操作，以便输出数据框如下所示：

Test                                   P-Value
FirstPt.vs.LastPt.InAlpha
FirstPt.vs.LastPt.InBeta        
ect... 

Answer 1

您可以使用一个函数执行这两项操作：

library(tidyverse)
t.test.by.group.combos <- function(.data, groups){
by <-  gsub(x = rlang::quo_get_expr(enquo(groups)), pattern = "\((.*)?\)", replacement = "\1")[-1]
.data %>%
group_by(!!!groups) %>%
select_if(is.integer) %>%
group_split() %>%
map(.,
~pivot_longer(., cols = (FirstPt:Last10), names_to = "name", values_to = "val") %>%
nest(data = val) %>%
full_join(.,.,by = by) %>%
filter(name.x != name.y) %>%
mutate(test = paste(name.x, "vs",name.y, !!!groups, sep = "."),
p.value = map2_dbl(data.x,data.y, ~t.test(unlist(.x), unlist(.y))$p.value)) %>%
select(test,p.value)%>%
filter(!duplicated(p.value))
) %>%
bind_rows() 
}

t.test.by.group.combos(data, vars(Variable))
#> # A tibble: 225 x 2
#>    test                    p.value
#>    <chr>                     <dbl>
#>  1 FirstPt.vs.LastPt.CaCO    0.511
#>  2 FirstPt.vs.First3.CaCO    0.184
#>  3 FirstPt.vs.First5.CaCO    0.494
#>  4 FirstPt.vs.First7.CaCO    0.354
#>  5 FirstPt.vs.First10.CaCO   0.893
#>  6 FirstPt.vs.Last3.CaCO     0.496
#>  7 FirstPt.vs.Last5.CaCO     0.909
#>  8 FirstPt.vs.Last7.CaCO     0.439
#>  9 FirstPt.vs.Last10.CaCO    0.146
#> 10 LastPt.vs.First3.CaCO     0.578
#> # … with 215 more rows
t.test.by.group.combos(data, vars(Variable, Location))
#> # A tibble: 674 x 2
#>    test                          p.value
#>    <chr>                           <dbl>
#>  1 FirstPt.vs.LastPt.CaCO.Alpha    0.850
#>  2 FirstPt.vs.First3.CaCO.Alpha    0.822
#>  3 FirstPt.vs.First5.CaCO.Alpha    0.895
#>  4 FirstPt.vs.First7.CaCO.Alpha    0.810
#>  5 FirstPt.vs.First10.CaCO.Alpha   0.645
#>  6 FirstPt.vs.Last3.CaCO.Alpha     0.870
#>  7 FirstPt.vs.Last5.CaCO.Alpha     0.465
#>  8 FirstPt.vs.Last7.CaCO.Alpha     0.115
#>  9 FirstPt.vs.Last10.CaCO.Alpha    0.474
#> 10 LastPt.vs.First3.CaCO.Alpha     0.991
#> # … with 664 more rows

这是一个冗长的函数，但通常我们按groups参数分组，然后选择组和任何整数列，然后按组拆分数据帧。之后，我们映射所有变量组合并为每个组合执行 t.tests。最后，我们将所有组重新加入到一个数据帧中。

Answer 2

我认为这就是你想要的。关键是使用group_by并从整洁中做。

df <- NULL
for(i in (4:(ncol(data)-1))){
for(j in ((i+1):ncol(data))){
df <- rbind(df,data %>% 
group_by(Location) %>% 
do(data.frame(pval = t.test(.[[i]],.[[j]], data = .)$p.value)) %>% 
ungroup() %>% 
mutate(Test = paste0(colnames(data)[i],'.vs.',colnames(data)[j]))
)
}
}
df$Test <- paste0(df$Test,'.In',df$Location)

Answer 3

也许，您可以使用以下代码实现您想要的：

library(dplyr)
library(tidyr)
data %>%
pivot_longer(cols = FirstPt:Last10) %>%
group_by(Variable) %>%
summarise(p_value = list(combn(name, 2, function(x) 
t.test(value[name == x[1]], value[name == x[2]])$p.value)), 
test = list(combn(name, 2, paste, collapse = "_"))) %>%
unnest(cols = c(test, p_value))

#  Variable p_value test           
#   <fct>      <dbl> <chr>          
# 1 CaCO       0.915 FirstPt_LastPt 
# 2 CaCO       0.529 FirstPt_First3 
# 3 CaCO       0.337 FirstPt_First5 
# 4 CaCO       0.350 FirstPt_First7 
# 5 CaCO       0.395 FirstPt_First10
# 6 CaCO       0.765 FirstPt_Last3  
# 7 CaCO       0.204 FirstPt_Last5  
# 8 CaCO       0.873 FirstPt_Last7  
# 9 CaCO       0.479 FirstPt_Last10 
#10 CaCO       1     FirstPt_FirstPt
# … with 24,740 more rows

要按Location分组，您可以将其添加到命令group_by并保持其余代码不变。