mapping over nested tibbles and run regressions - r

Question

I have some data which looks similar to the following:
Data:
rank_A <- sample(x = 1:5, size = 100, replace = TRUE)
rank_B <- sample(x = 1:5, size = 100, replace = TRUE)
Y_A <- rnorm(n = 100)
Y_B <- rnorm(n = 100)
X <- rnorm(n = 100)
df <- data.frame(rank_A, rank_B, Y_A, Y_B, X)
Which looks like:
> tibble(df)
# A tibble: 100 x 1
df$rank_A $rank_B $Y_A $Y_B $X
<int> <int> <dbl> <dbl> <dbl>
1 5 1 0.128 -0.833 1.15
2 2 1 0.165 0.325 0.225
3 1 3 0.525 -0.632 0.390
4 5 3 -1.32 0.718 -0.377
5 4 2 -0.900 -0.364 -0.259
6 5 3 -1.17 0.556 -0.0702
7 4 1 -0.560 -1.66 -1.64
8 5 1 -2.01 -0.898 0.306
9 1 1 -2.56 0.693 1.34
10 4 2 0.338 -0.733 -0.481
# ... with 90 more rows
So I have two rank columns and a corresponding Y for each rank. A and B here are just two categories. I want to run two regressions both on Y_A and Y_B both regressed on X. I have the following code which works for rank_A.
df_regs <- df %>%
group_by(rank_A) %>%
nest() %>%
mutate(
Reg_A = map(data,
~lm(Y_A ~ X,
data = .)
),
Reg_A_summary = map(Reg_A,
~summary(.)
)
)
df_regs$Reg_A_summary
I would like to construct it such that I do not have to write a new piece of code for rank_B. I have many different regressions (different X variables I will give to each Y_A and Y_B). I thought about doing something like the following (which doesn't work).
df %>%
mutate(Class_A = group_by(rank_A) %>%
nest() %>%
mutate(
Reg_A = map(data,
~lm(Y_a ~ X,
data = .)
)
),
Class_B = group_by(rank_B) %>%
nest() %>%
mutate(
Reg_B = map(data,
~lm(Y_b ~ X,
data = .)
)
)
)
If you know of a cleaner method, please also let me know your thoughts as I will have many regressions.

Answer 1

Here is one approach, using the new pivot functions from the latest release of tidyr...
df2 <- df %>% pivot_longer(cols=-X,
names_to = c(".value", "Class"),
names_sep = "_") %>%
group_by(Class, rank) %>%
summarise(Reg = list(lm(Y ~ X)))
This creates a tibble with columns Class, rank, and the list output Reg of the lm model for Y~X for each combination. You can then map through it to extract the summary, coefficients, or whatever.

Answer 2

We can do this with map2 for corresponding 'group_by` and dependent columns
library(dplyr)
library(purrr)
library(stringr)
library(broom)
map2(c('rank_A', 'rank_B'), c("Y_A", "Y_B"), ~ {
grp <- .x
predCol <- .y
newCol <- str_replace(predCol, "Y", "Reg")
df %>%
group_by_at(grp) %>%
nest %>%
mutate(!! newCol:= map(data,
~lm(reformulate('X', response = predCol))))
}
)