Skip to content

Non-sensical scaling for plots #894

Description

@profandyfield

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.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Bug 🐛Something isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions