I have a dataset with the following structure:
df <- data.frame(id = 1:5,
study = c("st1","st2","st3","st4","st5"),
a_var = c(10,20,30,40,50),
b_var = c(6,5,4,3,2),
c_var = c(3,4,5,6,7),
d_var = c(80,70,60,50,40))
I would like to calculate the difference between each column that has _var in its name and the mean of all other columns containing _var in their names, like this:
mean_deviated_value <- function(data, variable) {
md_value = data[,variable] - rowMeans(data[,names(data) != variable])
md_value
}
df$a_var_md <- mean_deviated_value(dplyr::select(df, contains("_var")), "a_var")
df$b_var_md <- mean_deviated_value(dplyr::select(df, contains("_var")), "b_var")
df$c_var_md <- mean_deviated_value(dplyr::select(df, contains("_var")), "c_var")
df$d_var_md <- mean_deviated_value(dplyr::select(df, contains("_var")), "d_var")
Which gives me my desired output:
id study a_var b_var c_var d_var a_var_md b_var_md c_var_md d_var_md
1 1 st1 10 6 3 80 -19.666667 -12.33333 -9.80 83.80000
2 2 st2 20 5 4 70 -6.333333 -16.91667 -10.35 70.76667
3 3 st3 30 4 5 60 7.000000 -21.50000 -10.90 57.73333
4 4 st4 40 3 6 50 20.333333 -26.08333 -11.45 44.70000
5 5 st5 50 2 7 40 33.666667 -30.66667 -12.00 31.66667
How do I do it in one go, without repeating the code, preferably with dplyr/purrr?
I tried this:
df %>%
mutate(across(contains("_var"), ~ list(md = .x - rowMeans(select(., contains("_var") & !.x)))))
And got this error:
Error: Problem with `mutate()` input `..1`.
ℹ `..1 = across(...)`.
x no applicable method for 'select' applied to an object of class "c('double', 'numeric')"
We can use map_dfc with transmute to create *_md columns, and glue syntax for the names.
library(tidyverse)
nms <- names(df) %>%
str_subset('^.*_')
bind_cols(df, map_dfc(nms, ~transmute(df, '{.x}_md' := mean_deviated_value(select(df, contains("_var")), .x))))
#> id study a_var b_var c_var d_var a_var_md b_var_md c_var_md d_var_md
#> 1 1 st1 10 6 3 80 -19.666667 -25.00000 -29.00000 73.66667
#> 2 2 st2 20 5 4 70 -6.333333 -26.33333 -27.66667 60.33333
#> 3 3 st3 30 4 5 60 7.000000 -27.66667 -26.33333 47.00000
#> 4 4 st4 40 3 6 50 20.333333 -29.00000 -25.00000 33.66667
#> 5 5 st5 50 2 7 40 33.666667 -30.33333 -23.66667 20.33333
Note that if you use assigment. The first time rowMeans will compute with b_var, c_bar and d_bar. But the second time, contains("_var") will also capture the previously created a_var_md and use it to compute the means. I don't know if this is intended behaviour but it is worth mentioning.
df$a_var_md <- mean_deviated_value(dplyr::select(df, contains("_var")), "a_var")
select(df, contains("_var"))
#> a_var b_var c_var d_var a_var_md
#> 1 10 6 3 80 -19.666667
#> 2 20 5 4 70 -6.333333
#> 3 30 4 5 60 7.000000
#> 4 40 3 6 50 20.333333
#> 5 50 2 7 40 33.666667
We can avoid this by replacing contains("_var") with matches("^.*_var$")
Created on 2021-12-20 by the reprex package (v2.0.1)