predict.sback {wsbackfit}

Predict method for sback fits

Description

Predicted smooth functions and values based on an sback object

Usage

## S3 method for class 'sback'
predict(object, newdata, newoffset = NULL, ...)

Arguments

object

an object of class sback.

newdata

a data frame containing the values of the covariates at which predictions are wanted. If not provided then the predictions correspond to the original data.

newoffset

an optional numerical vector containing an a priori known component to be included in the linear predictor for the predictions (offset associated with the newdata).

...

not yet implemented.

Value

A list with the following components:

pdata

the original supplied newdata argument.

poffset

the original supplied newoffset argument.

coeff

a numeric vector with the estimated coefficients. This vector includes both the parametric effects as well as the coefficients associated with the linear component of the nonparametric functions.

peffects

matrix with the estimated nonparametric functions (only the nonlinear component) for each covariate value in the original supplied newdata.

pfitted.values

a numeric vector with the fitted values for the supplied newdata.

Author(s)

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

See Also

sback, summary.sback

Examples

library(wsbackfit)
data(infect)

# Generalized varying coefficient model with binary response
m3 <- sback(formula = inf ~ sb(gluc, h = 10) + sb(gluc, by = linf, h = 10), 
  data = infect, family = "binomial", kbin = 15)

summary(m3)

# Plot both linear and non linear 
# components of nonparametric functions: composed = FALSE
op <- par(no.readonly = TRUE)
par(mfrow = c(1,3))
plot(m3, composed = FALSE)

# Personalized plots
# First obtain predictions in new data
# Create newdata for prediction
ngrid <- 30
gluc0 <- seq(50, 190, length = ngrid)
linf0 <- seq(0, 45, length = ngrid)
df  <- expand.grid(gluc = gluc0, linf = linf0)

m3p <- predict(m3, newdata = df)

par(mfrow = c(1,2))
ii <- order(df[,"gluc"])

## Parametric coefficients
names(m3p$coeff)

# Nonlinear components
colnames(m3p$peffects) 

# Include the linear component
plot(df[ii,"gluc"], m3p$coeff[["gluc"]]*df[ii,"gluc"] + 
  m3p$peffects[ii,"sb(gluc, h = 10)"], 
  type = 'l', xlab = "Glucose (mg/dl)", ylab = "f_1(gluc)", 
  main = "Nonparametric effect of Glucose")

# Include the linear component
plot(df[ii,"gluc"], m3p$coeff[["gluc:linf"]]*df[ii,"gluc"] + 
  m3p$peffects[ii,"sb(gluc, h = 10, by = linf)"], 
  type= 'l', xlab = "Glucose (mg/dl)", ylab = "f_2(gluc)",
  main = "Varying coefficients as a function of Glucose")

# Countour plot of the probability of post-opererational infection
n <- sqrt(nrow(df))
Z <- matrix(m3p$pfitted.values, n, n)
filled.contour(z = Z, x = gluc0, y = linf0, 
  xlab = "Glucose (mg/dl)", ylab = "Lymphocytes (%)",
  main = "Probability of post-opererational infection")

par(op)