Supplying custom knot values to geom_smooth by facet

[muswellbrook]

Hi there,

I'm trying to plot some piecewise/broken-stick regressions using geom_smooth. For example:

ggplot(mtcars, aes(x = mpg, y = hp)) +
  geom_point() +
  geom_smooth(method = lm, 
              formula = y ~ splines::bs(x, df = 2, degree = 1, knots = 20)) +
  facet_wrap(~gear)

However this uses a single value (knots=20) for the breakpoint in each facet. I would like to pass a unique value for each facet (that I have calculated earlier). Is this possible?

I tried adding a variable to the data frame and passing the variable name (e..g, psi) as an aesthetic, but geom_smooth wouldn't recognize it. I suspect the answer has something to do with defining a custom function as described at the end of this page: https://ggplot2.tidyverse.org/reference/geom_smooth.html

Thanks so much for this package - it is amazing and I love using it.

Best,

Rich

[nirgrahamuk]

I would do that like this

library(tidyverse)
mtcars2 <- select(mtcars, gear, mpg)

lm18k <- lm(hp ~ splines::bs(mpg, df = 2, degree = 1, knots = 18), mtcars)
mtcars2$lm18k <- predict(lm18k, newdata = mtcars)
lm19k <- lm(hp ~ splines::bs(mpg, df = 2, degree = 1, knots = 19), mtcars)
mtcars2$lm19k <- predict(lm19k, newdata = mtcars)
lm20k <- lm(hp ~ splines::bs(mpg, df = 2, degree = 1, knots = 20), mtcars)
mtcars2$lm20k <- predict(lm20k, newdata = mtcars)

mtcars3 <- pivot_longer(mtcars2,
  cols = (lm18k:lm20k)
) %>%
  mutate(keepme = case_when(
    name == "lm18k" & gear == 3 ~ TRUE,
    name == "lm19k" & gear == 4 ~ TRUE,
    name == "lm20k" & gear == 5 ~ TRUE,
    TRUE ~ FALSE
  )) %>%
  filter(keepme)

ggplot(mtcars, aes(x = mpg, y = hp)) +
  geom_point() +
  geom_smooth(
    data = mtcars3,
    method = lm, mapping = aes(y = value)
  ) +
  facet_wrap(~gear)

[muswellbrook]

Thanks for your response but it seems to miss the problem which is to provide a custom knot to each facet for the (broken) line of best fit. As far as I can tell your code just overlays a linear fit on the predictions from custom piecewise models. Maybe I missed something...?

[nirgrahamuk]

However this uses a single value (knots=20) for the breakpoint in each facet. I would like to pass a unique value for each facet (that I have calculated earlier). Is this possible?

My example shows how to have 18 knot param for one facet, 19 for another, and 20 for the third.

[muswellbrook]

My example shows how to have 18 knot param for one facet, 19 for another, and 20 for the third.

Ok, but where does your code plot the fitted broken-stick? I can see the call to geom_smooth, but it looks like a linear fit. Thanks for your time.

[nirgrahamuk]

You are right to point out that I incorrectly smoothed over the actual broken stick line. I'll fix that for you

ggplot(mtcars, aes(x = mpg, y = hp)) +
  geom_point() + 
  geom_line(
    data = distinct(mtcars3),
  mapping = aes(y = value)
  ) +
  facet_wrap(~gear)