nlshc: Heteroscedastic Nonlinear Regression

In regtools: Regression and Classification Tools

Heteroscedastic Nonlinear Regression

Description

Extension of nls to the heteroscedastic case.

Usage

nlshc(nlsout,type='HC')

Arguments

nlsout	Object of type 'nls'.
type	Eickert-White algorithm to use. See documentation for nls.

Details

Calls nls but then forms a different estimated covariance matrix for the estimated regression coefficients, applying the Eickert-White technique to handle heteroscedasticity. This then gives valid statistical inference in that setting.

Some users may prefer to use nlsLM of the package minpack.lm instead of nls. This is fine, as both functions return objects of class 'nls'.

Value

Estimated covariance matrix

Author(s)

Norm Matloff

References

Zeileis A (2006), Object-Oriented Computation of Sandwich Estimators. Journal of Statistical Software, 16(9), 1–16, https://www.jstatsoft.org/v16/i09/.

Examples

# simulate data from a setting in which mean Y is 
# 1 / (b1 * X1 + b2 * X2)
n <- 250
b <- 1:2
x <- matrix(rexp(2*n),ncol=2)
meany <- 1 / (x %*% b)  # reg ftn
y <- meany + (runif(n) - 0.5) * meany  # heterosced epsilon
xy <- cbind(x,y)
xy <- data.frame(xy)
# see nls() docs
nlout <- nls(X3 ~ 1 / (b1*X1+b2*X2),
   data=xy,start=list(b1 = 1,b2=1))
nlshc(nlout)

regtools documentation built on March 31, 2022, 1:06 a.m.