R: Get the "earth" basis matrix

model.matrix.earth {earth}

R Documentation

Get the "earth" basis matrix

Description

Get the basis matrix of an earth object.

Usage

## S3 method for class 'earth'
model.matrix(object = stop("no 'object' arg"),
    x = NULL, subset = NULL, which.terms = NULL,
    ...,
    env = parent.frame(),
    trace = 0,
    Callers.name = "model.matrix.earth")

Arguments

`object`	An `earth` object. This is the only required argument.
`x`	An input matrix with the same number of columns as the `x` matrix used to construct the original `earth` object. Default is NULL, meaning use the original `x` matrix after taking the original `subset`, if any.
`subset`	Which rows to use in `x`. Default is NULL, meaning use all of `x`.
`which.terms`	Which terms to use. Default is NULL, meaning use `object$selected.terms`.
`...`	Unused, but provided for generic/method consistency.
`env`	For internal use.
`trace`	Default 0. Set to non-zero to see which data `model.matrix.earth` is using.
`Callers.name`	For internal use (used by earth in trace messages).

Value

A basis matrix bx of the same form returned by earth.

If x, subset, and which.terms are all NULL, this function returns the object's bx. In this case, it is perhaps easier to simply use object$bx.

The format of bx is described in earth. The matrix bx can be used as the input matrix to lm or glm, as shown below in the example. In fact, that is what earth does internally after the pruning pass — it calls lm.fit, and additionally glm if earth's glm argument is used.

Examples

data(trees)
a <- earth(Volume ~ ., data = trees)
summary(a, decomp = "none") # "none" to print terms in same seq as a.lm below

# yields:
#  Call: earth(formula=Volume~., data=trees)
#
#               coefficients
#  (Intercept)        27.246
#  h(Girth-14)         6.177
#  h(14-Girth)        -3.266
#  h(Height-72)        0.491
#
#  Selected 4 of 6 terms, and 2 of 2 predictors
#  Importance: Girth, Height
#  Number of terms at each degree of interaction: 1 3 (additive model)
#  GCV 10.6    RSS 197    GRSq 0.962    RSq 0.976

bx <- model.matrix(a)               # equivalent to bx <- a$bx
a.lm <- lm(trees$Volume ~ bx[,-1])  # -1 to drop intercept
summary(a.lm)                       # yields same coeffs as above summary
                                    # displayed t values are not meaningful
# yields:
#  Call:
#  lm(formula = trees$Volume ~ bx[, -1])
#
#  Residuals:
#     Min     1Q Median     3Q    Max
#  -4.882 -1.770  0.281  1.646  4.983
#
#  Coefficients:
#                       Estimate Std. Error t value Pr(>|t|)
#  (Intercept)            27.246      1.123   24.26  < 2e-16
#  bx[, -1]h(Girth-14)     6.177      0.354   17.44  3.2e-16
#  bx[, -1]h(14-Girth)    -3.266      0.335   -9.76  2.4e-10
#  bx[, -1]h(Height-72)    0.491      0.123    3.99  0.00045
#
#  Residual standard error: 2.7 on 27 degrees of freedom
#  Multiple R-squared: 0.976,   Adjusted R-squared: 0.973
#  F-statistic:  361 on 3 and 27 DF,  p-value: <2e-16