AIC and BIC selection of number of spline knots

Calculates AIC and BIC for the selection of knots in a spline over values (potentially including polynomials) up to a user-defined maximum.

NKnots(
  form,
  var,
  data,
  degree = 3,
  min.knots = 1,
  max.knots = 10,
  includePoly = FALSE,
  plot = FALSE,
  criterion = c("AIC", "BIC", "CV"),
  cvk = 10,
  cviter = 10
)

Arguments

form

A formula detailing the model for which smoothing is to be evaluated.
var

A character string identifying the variable for which smoothing is to be evaluated.
data

Data frame providing values of all variables in form.
degree

Degree of polynomial in B-spline basis functions.
min.knots

Minimum number of internal B-spline knots to be evaluated.
max.knots

Maximum number of internal B-spline knots to be evaluated.
includePoly

Include linear and polynomial models up to, and including degree-th order polynomials.
plot

Logical indicating whether a plot should be returned.
criterion

Statistical criterion to minimize in order to find the best number of knots - AIC, BIC or Cross-validation.
cvk

Number of groups for cross-validation
cviter

Number of iterations of cross-validation to average over. 10 is the default but in real-world applications, this should be somewhere around 200.

Value

A plot, if plot=TRUE, otherwise a data frame with the degrees of freedom and corresponding fit measure.

Examples


data(Prestige, package="carData")
NKnots(prestige ~ education + type, var="income", data=na.omit(Prestige), plot=FALSE)