Search code examples
rggplot2tidyversepurrr

How to plot leave-one-out regression lines with ggplot / tidyverse

I'd like to use ggplot to plot a regression line and also show the regression lines resulting from omitting each of the observations in turn.

Here is a basic plot, using the anscombe dataset:

data(anscombe)

ggplot(anscombe, aes(x = x2, y = y2)) +
  geom_point(color = col, size = 4) +
  geom_smooth(method = "lm", formula = y ~ x, se = FALSE,
              color = "red", linewidth = 1.5) +
  labs(title = "Dataset 1" ) +
  theme_bw(base_size = 14)

I can do what I want manually, but it is very clunky. There is probably a tidyverse solution, most likely using purrr, but I can't think of it.

ggplot(anscombe, aes(x = x2, y = y2)) +
  geom_point(color = col, size = 4) +
  geom_smooth(data = anscombe |> slice(-1), method = "lm", formula = y ~ x, se = FALSE, color = "grey") +
  geom_smooth(data = anscombe |> slice(-2), method = "lm", formula = y ~ x, se = FALSE, color = "grey") +
# ...
  geom_smooth(data = anscombe |> slice(-10), method = "lm", formula = y ~ x, se = FALSE, color = "grey") +
  geom_smooth(data = anscombe |> slice(-11), method = "lm", formula = y ~ x, se = FALSE, color = "grey") +
  geom_smooth(method = "lm", formula = y ~ x, se = FALSE,
              color = "red", linewidth = 1.5) +
  labs(title = "Dataset 2" ) +
  theme_bw(base_size = 14)

enter image description here

Solution

  • Using e.g. lapply you could loop over the rows of your data like so:

    library(ggplot2)

ggplot(anscombe, aes(x = x2, y = y2)) +
  geom_point(color = "blue", size = 4) +
  lapply(seq(nrow(anscombe)), function(x) {
    geom_smooth(
      data = anscombe[-x, ],
      method = "lm", formula = y ~ x, se = FALSE,
      color = "grey", linewidth = 1.5
    )
  }) +
  geom_smooth(
    method = "lm", formula = y ~ x, se = FALSE,
    color = "red", linewidth = 1.5
  ) +
  labs(title = "Dataset 1") +
  theme_bw(base_size = 14)