Partial-effect and two-predictor surface plots (visreg / plotmo style)

Publication-ready ggplot replacements for the visreg and plotmo workflow used in Moses et al. (2021). plot_partial() draws a one-predictor conditional (partial) effect with optional residuals and confidence band. plot_surface() draws a two-predictor response surface as a filled contour (the pattern that paper built by converting visreg::visreg2d() into ggplot by hand). No visreg or plotmo dependency.

Usage

plot_partial(
  model,
  x,
  data = NULL,
  by = NULL,
  residuals = FALSE,
  se = TRUE,
  level = 0.95,
  n = 100,
  ...
)

plot_surface(
  model,
  x,
  y,
  data = NULL,
  n = 40,
  type = c("contour", "persp"),
  points = TRUE,
  ...
)

Arguments

model: A fitted model (lm, glm, negbin, gam, or model_count() / td_count_model).
x, y: Focal predictor names (character) or bare names.
data: Optional data frame used to build the prediction grid. Defaults to model.frame(model).
by: Optional conditioning variable for panels (partial plots).
residuals: Overlay partial residuals?
se: Draw a pointwise confidence band (partial) or skip (surface)?
level: Confidence level for the band.
n: Number of grid points per axis.
...: Passed to predict methods.
type: "contour" (filled contours; default) or "persp" (base graphics::persp() drawn to the current device and returned as a recorded plot list; prefer "contour" for publication ggplot output).
points: Overlay observed (x, y) points on the surface?

Value

A ggplot object (plot_partial, plot_surface with type = "contour"). For type = "persp", a list with the persp transformation matrix and the prediction grid (still no plotmo).

References

Moses, J., Fayle, T. M., Novotny, V. and Klimes, P. (2021). Elevation and leaf litter interact in determining the structure of ant communities on a tropical mountain. Biotropica, 53, 906-919. doi:10.1111/btp.12914

Examples

ants <- png_ants |>
  dplyr::group_by(site, elevation) |>
  dplyr::summarise(richness = dplyr::n_distinct(species), .groups = "drop")
ants$litter <- 1 + ants$elevation / 2000 + stats::rnorm(nrow(ants), 0, 0.3)
fit <- model_count(ants, richness ~ poly(elevation, 2) * litter,
                   family = "poisson")
plot_partial(fit, elevation, residuals = TRUE)
#> Warning: prediction from rank-deficient fit; attr(*, "non-estim") has doubtful cases

plot_surface(fit, elevation, litter)
#> Warning: prediction from rank-deficient fit; attr(*, "non-estim") has doubtful cases