easystats
check_model() linearity & variance for categorical predictors
For models with a categorical predictor (3 levels here):
star <- read.csv("https://drmankin.github.io/disc_stats/star.csv") |>
dplyr::mutate(dplyr::across(dplyr::starts_with("star"), forcats::as_factor))
star_maths_lm <- lm(math2 ~ star2, data = star, na.action = na.exclude)
performance::check_model(star_maths_lm)
plot(star_maths_lm, which = 1)
The variance and linearity plots could arguably be more informative. In particular the one to compare the variance across the different groups.
In a way the very large CI masks the variability in the data points. But I'm not s ure what's the best way to adress it:
- Add an option to simply drop the CI ribbon?
- Have bespoke plots for categorical predictors?
- Alternatives?
Sir @DominiqueMakowski I have raised a PR regarding this issue can you please review it and give your feedback
https://github.com/easystats/performance/pull/874
As mentioned here, maybe we should first try to disable the confidence bands. I think this already makes the plot more readable, especially since other plotting functions also show the data points (but no CI). I think the CI are problematic here, not necessarily the data points (at least not for all plots).
