npregCv: Cross-Validation Nonparametric Models

In micEconNP: Microeconomic Analysis with Non-Parametric Methods

Description

Leave-one-out cross-validation of nonparametric kernel regression models, partial linear models, and categorical regression splines.

Usage

   npregCv( x )

   npplregCv( x )
   
   crsCv( x )

Arguments

x	an object of class npregression (returned by npreg) or an object of class plregression (returned by npplreg) or an object of class crs (returned by crs).

Value

A single numeric value: the mean squared prediction error over all observations, where the predicted values are obtained based on models that are estimated without the respective observation. This value has an attribute err that is the vector of the prediction errors of all observations.

Author(s)

Arne Henningsen

See Also

npreg, npplreg, crs.

Examples

set.seed(42)
n <- 250
x1 <- rnorm(n)
x2 <- rnorm(n)
y <- 1 + x1 + x2^2 + rnorm(n)

model <- npreg( y ~ x1 + x2, regtype="ll" )

cv <- npregCv( model )
c( cv )
all.equal( cv, model$bws$fval, check.attributes = FALSE )


# partially linear model
set.seed(42)
n <- 250
x1 <- rnorm(n)
x2 <- rbinom(n, 1, .5)
z1 <- rbinom(n, 1, .5)
z2 <- rnorm(n)
y <- 1 + x1 + x2 + z1 + sin(z2) + rnorm(n)

model <- npplreg( y ~ x1 + factor(x2) | factor(z1) + z2, regtype="ll" )

cv <- npplregCv( model )
c( cv )


# categorical regression splines
set.seed(42)
n <- 250
num.eval <- 50
x1 <- runif(n)
x2 <- runif(n)
z <- round( runif( n, min = 0, max = 3 ) )
dgp <- cos( 2 * pi * x1 ) + sin( 2 * pi * x2 ) + z/5
z <- factor(z)
y <- dgp + rnorm( n, sd = 0.5 )

model <- crs( y ~ x1 + x2 + z, deriv = 1, nmulti = 1 )

cv <- crsCv( model )
c( cv )
all.equal( cv, model$cv.min, check.attributes = FALSE )

micEconNP documentation built on May 2, 2019, 6:30 p.m.