This function provides several plots of the result of fitting a two-predictor model.

# S3 method for lm
       which = 1, 
       plane.col = "gray", plane.alpha = 0.5,
       sharedMouse = TRUE, 
       do_grid = TRUE,
       grid.col = "black", 
       grid.alpha = 1,
       grid.steps = 5,
       sub.steps = 4,
       vars = get_all_vars(terms(x), x$model),
       clip_to_density = 0,



An object inheriting from class "lm" obtained by fitting a two-predictor model.


Which plot to show? See Details below.

plane.col, plane.alpha

These parameters control the colour and transparency of a plane or surface.


If multiple plots are requested, should they share mouse controls, so that they move in sync?


Use the surface3d function to plot the surface rather than planes3d. This allows curved surfaces to be shown. The default is FALSE if the model looks like a simple 2 parameter linear fit, otherwise TRUE.


Plot a grid.

grid.col, grid.alpha, grid.steps

Characteristics of the grid.


If use_surface3d is TRUE, use an internal grid of grid.steps*sub.steps to draw the surface. sub.steps > 1 allows curvature within facets. Similarly, if do_grid is TRUE, it allows curvature within grid lines.


A dataframe containing the variables to plot in the first three columns, with the response assumed to be in column 1. See the Note below.


If positive, the surface, plane or grid will be clipped to a region with sufficient data.


Other parameters to pass to the default plot3d method, to control the appearance of aspects of the plot other than the plane.


Three plots are possible, depending on the value(s) in which:

  1. (default) Show the points and the fitted plane or surface.

  2. Show the residuals and the plane at z = 0.

  3. Show the predicted values on the fitted plane or surface.

If clip_to_density is positive, then the surface, plane or grid will be clipped to the region where a non-parametric density estimate (using MASS::kde2d), normalized to have a maximum value of 1, is greater than the given value. This will suppress parts of the plot that aren't supported by the observed data.


Called for the side effect of drawing one or more plots.

Invisibly returns a high-level vector of object ids. Names of object ids have the plot number (in drawing order) appended.


The default value for the vars argument will handle simple linear models with a response and two predictors, and some models with functions of those two predictors. For models that fail (e.g. models using poly), you can include the observed values as in the third example below.

If clip_to_density > 0,

  1. The clipping is approximate, so it may not agree perfectly between surfaces, planes and grids.

  2. This option requires the suggested packages MASS and akima, and will be ignored with a warning if either is not installed.


Duncan Murdoch


ids <- plot3d(lm(mpg ~ wt + qsec, data = mtcars), which = 1:3)
#>  [1] "data.1"  "axes.1"  "xlab.1"  "ylab.1"  "zlab.1"  "plane.1" "grid.1" 
#>  [8] "data.2"  "axes.2"  "xlab.2"  "ylab.2"  "zlab.2"  "plane.2" "grid.2" 
#> [15] "data.3"  "axes.3"  "xlab.3"  "ylab.3"  "zlab.3"  "plane.3" "grid.3" 

plot3d(lm(mpg ~ wt + I(wt^2) + qsec, data = mtcars))
open3d() # Specify vars in the order: response, pred1, pred2. plot3d(lm(mpg ~ poly(wt, 3) + qsec, data = mtcars), vars = mtcars[,c("mpg", "wt", "qsec")])
open3d() # Clip parts of the plot with few (wt, qsec) points plot3d(lm(mpg ~ poly(wt, 3) + qsec, data = mtcars), vars = mtcars[,c("mpg", "wt", "qsec")], clip_to_density = 0.1)