Odd plots for categorical variable only models

[jebyrnes]

It attempting to look at lm() objects with only categorical predictors and evaluating fitted v. residuals (for linearity) or sqrt(abs(std residuals)), I get some very odd plots - particularly relative to good old plot(). Also some very interesting errors. This seems like non-ideal behavior.

library(palmerpenguins)
library(performance)

mod <- lm(body_mass_g ~ sex, data = penguins)

check_model(mod, check = "linearity") |> plot()
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : pseudoinverse used at 3858.9
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : neighborhood radius 686.83
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : reciprocal condition number 1.3398e-15
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : There are other near singularities as well. 4.7173e+05
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : pseudoinverse used at
#> 3858.9
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : neighborhood radius
#> 686.83
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : reciprocal condition
#> number 1.3398e-15
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : There are other near
#> singularities as well. 4.7173e+05

check_model(mod, check = "homogeneity") |> plot()
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : pseudoinverse used at 3858.9
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : neighborhood radius 686.83
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : reciprocal condition number 1.3398e-15
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric =
#> parametric, : There are other near singularities as well. 4.7173e+05
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : pseudoinverse used at
#> 3858.9
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : neighborhood radius
#> 686.83
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : reciprocal condition
#> number 1.3398e-15
#> Warning in predLoess(object$y, object$x, newx = if
#> (is.null(newdata)) object$x else if (is.data.frame(newdata))
#> as.matrix(model.frame(delete.response(terms(object)), : There are other near
#> singularities as well. 4.7173e+05

^{Created on 2022-10-13 with reprex v2.0.2}

Compare this to plot()

library(palmerpenguins)
library(performance)

mod <- lm(body_mass_g ~ sex, data = penguins)

plot(mod, which = 1)

^{Created on 2022-10-13 with reprex v2.0.2}

[bwiernik]

The issue is trying to fit a smoother line to only two fitted values.

We should detect all categorical predictors for a model and then do something like raincloud/violin/boxplots instead

[jebyrnes]

Agree. But, I'm curious why the points didn't show up properly as well.

[mattansb]

They do, but the scale on the y axis is HUGE....

[jebyrnes]

OH! It's due to the smoother. Ah. That makes sense. Is there a way to remove it?

[bwiernik]

Not currently.