I quite often come across plots from check_model() where the y-axis is scaled ludicrously (presumably because of the loess function creating a curve that is wildly curved). This code breaks reprex() but gives an example:
library(performance)
library(see)
library(ggplot2)
# remotes::install_github("profandyfield/discovr")
display_tib <- discovr::display
fb_glm <- glm(display ~ fb, family = binomial(), data = display_tib)
check_model(fb_glm, check = "outliers")
The obvious solution in these cases is to explore the residuals/influence stats more manually, or adjust the scales manually with sopmething like
fb_resid <- check_model(fb_glm, check = "outliers")
fb_gg <- plot(fb_resid)
fb_gg[[1]] + coord_cartesian(ylim = c(-4, 4))
but from a teaching perspective it's a bit of a nightmare - you just want students to see plots they can say something sensible about.
Is there a way to limit to the scaling to sensible maximum/minimum values? I've seen this behaviour with the homoscedasticity plots too.
I quite often come across plots from
check_model()where the y-axis is scaled ludicrously (presumably because of the loess function creating a curve that is wildly curved). This code breaksreprex()but gives an example:The obvious solution in these cases is to explore the residuals/influence stats more manually, or adjust the scales manually with sopmething like
but from a teaching perspective it's a bit of a nightmare - you just want students to see plots they can say something sensible about.
Is there a way to limit to the scaling to sensible maximum/minimum values? I've seen this behaviour with the homoscedasticity plots too.