sb {wsbackfit}

Specify a nonparametric and/or a varying coefficient term in a wsbackfit formula

Description

Function used to indicate nonparametric terms and varying coefficient terms in a sback formula.

Usage

sb(x1 = NULL, by = NULL, h = -1)

Arguments

x1

the univariate predictor

by

numeric predictor of the same dimension as x1. If present, the coefficients of this predictor depend, nonparametrically, on x1, i.e., a varying coefficient term.

h

bandwidth (on the scale of the predictor) for this term. If h = -1, the bandwidth is automatically selected using k-fold cross-validation (see sback). A value of 0 would indicate a linear fit. By default -1.

Value

A list with the following components:

cov

character vector with the name(s) of the involved predictor(s).

h

numeric value with the specified smoothing parameter.

Author(s)

Javier Roca-Pardinas, Maria Xose Rodriguez-Alvarez and Stefan Sperlich

See Also

sback, summary.sback, plot.sback

Examples

library(wsbackfit)
set.seed(123)
###############################################
# Gaussian Simulated Sample
###############################################
set.seed(123)
# Define the data generating process
n <- 1000
x1 <- runif(n)*4-2
x2 <- runif(n)*4-2
x3 <- runif(n)*4-2
x4 <- runif(n)*4-2
x5 <- as.numeric(runif(n)>0.6)

f1 <- 2*sin(2*x1)
f2 <- x2^2
f3 <- 0
f4 <- x4
f5 <- 1.5*x5

mu <- f1 + f2 + f3 + f4 + f5
err <- (0.5 + 0.5*x5)*rnorm(n)
y <- mu + err

df <- data.frame(x1 = x1, x2 = x2, x3 = x3, x4 = x4, x5 = as.factor(x5), y = y)

# Fit the model with a fixed bandwidth for each covariate
m0 <- sback(formula = y ~ x5 + sb(x1, h = 0.1) + sb(x2, h = 0.1) 
  + sb(x3, h = 0.1) + sb(x4, h = 0.1), kbin = 30, data = df)

summary(m0)

op <- par(no.readonly = TRUE)

par(mfrow = c(2,2))
plot(m0)


# Fit the model with the bandwidths selected by k-fold cross-validation.
m1 <- sback(formula = y ~ x5 + sb(x1, h = -1) + sb(x2, h = -1) 
  + sb(x3, h = -1) + sb(x4, h = -1), kbin = 30, bw.grid = seq(0.01, 0.99, length = 30),
  data = df)

summary(m1)

par(mfrow = c(2,2))
plot(m1)


par(op)