A new R‐package, called
nlr, has been used to implement the methods detailed in in this book. It can be download free from the Comprehensive R Archive Network (CRAN).
nlr is an object‐oriented package system that fits nonlinear regression models using robust methods. It includes features to handle heteroscedasticity of errors, autocorrelated errors, and outlier detection. The object system can define a nonlinear regression model, a robust loss rho function, and a heteroscedastic variance in an object called
nlr also includes features that can compute a parameter covariance matrix, plot a fitted model, predict responses, compute prediction intervals, and perform further inferences. The final fit results are accessible through output objects that are flexible, so that researchers can apply them to their own particular problems.
To use the
nlr package, we first have to define a nonlinear regression function model as an
nl.form object. This is an extension of the initial object definition by Bunke et al. (1995b), with extra functionality. The data set for which the model is going to be fitted has to be adjusted to use the names of the response and predictor variables defined in a nonlinear function model object. A robust function should also be defined ...