plot.regvec3d: Plot method for regvec3d objects

Description

The plot method for regvec3d objects uses the low-level graphics tools in this package to draw 3D and 3D vector diagrams reflecting the partial and marginal relations of y to x1 and x2 in a bivariate multiple linear regression model, lm(y ~ x1 + x2).

The summary method prints the vectors and their vector lengths, followed by the summary for the model.

Usage

# S3 method for regvec3d
plot(
  x,
  y,
  dimension = 3,
  col = c("black", "red", "blue", "brown", "lightgray"),
  col.plane = "gray",
  cex.lab = 1.2,
  show.base = 2,
  show.marginal = FALSE,
  show.hplane = TRUE,
  show.angles = TRUE,
  error.sphere = c("none", "e", "y.hat"),
  scale.error.sphere = x$scale,
  level.error.sphere = 0.95,
  grid = FALSE,
  add = FALSE,
  ...
)
# S3 method for regvec3d
summary(object, ...)
# S3 method for regvec3d
print(x, ...)

Value

None

Arguments

x: A “regvec3d” object
y: Ignored; only included for compatibility with the S3 generic
dimension: Number of dimensions to plot: 3 (default) or 2
col: A vector of 5 colors. col[1] is used for the y and residual (e) vectors, and for x1 and x2; col[2] is used for the vectors y -> yhat and y -> e; col[3] is used for the vectors yhat -> b1 and yhat -> b2;
col.plane: Color of the base plane in a 3D plot or axes in a 2D plot
cex.lab: character expansion applied to vector labels. May be a number or numeric vector corresponding to the the rows of X, recycled as necessary.
show.base: If show.base > 0, draws the base plane in a 3D plot; if show.base > 1, the plane is drawn thicker
show.marginal: If TRUE also draws lines showing the marginal relations of y on x1 and on x2
show.hplane: If TRUE, draws the plane defined by y, yhat and the origin in the 3D
show.angles: If TRUE, draw and label the angle between the x1 and x2 and between y and yhat, corresponding respectively to the correlation between the xs and the multiple correlation
error.sphere: Plot a sphere (or in 2D, a circle) of radius proportional to the length of the residual vector, centered either at the origin ("e") or at the fitted-values vector ("y.hat"; the default is "none".)
scale.error.sphere: Whether to scale the error sphere if error.sphere="y.hat"; defaults to TRUE if the vectors representing the variables are scaled, in which case the oblique projections of the error spheres can represent confidence intervals for the coefficients; otherwise defaults to FALSE.
level.error.sphere: The confidence level for the error sphere, applied if scale.error.sphere=TRUE.
grid: If TRUE, draws a light grid on the base plane
add: If TRUE, add to the current plot; otherwise start a new rgl or plot window
...: Parameters passed down to functions [unused now]
object: A regvec3d object for the summary method

Details

A 3D diagram shows the vector y and the plane formed by the predictors, x1 and x2, where all variables are represented in deviation form, so that the intercept need not be included.

A 2D diagram, using the first two columns of the result, can be used to show the projection of the space in the x1, x2 plane.

The drawing functions vectors and link{vectors3d} used by the plot.regvec3d method only work reasonably well if the variables are shown on commensurate scales, i.e., with either scale=TRUE or normalize=TRUE.

References

Fox, J. (2016). Applied Regression Analysis and Generalized Linear Models, 3rd ed., Sage, Chapter 10.

Examples

Run this code

if (require(carData)) {
   data("Duncan", package="carData")
   dunc.reg <- regvec3d(prestige ~ income + education, data=Duncan)
   plot(dunc.reg)
   plot(dunc.reg, dimension=2)
   plot(dunc.reg, error.sphere="e")
   summary(dunc.reg)

   # Example showing Simpson's paradox
   data("States", package="carData")
   states.vec <- regvec3d(SATM ~ pay + percent, data=States, scale=TRUE)
   plot(states.vec, show.marginal=TRUE)
   plot(states.vec, show.marginal=TRUE, dimension=2)
   summary(states.vec)
}

Run the code above in your browser using DataLab