VarReg: An R package for semi-parametric variance regression

Variance regression is used to model heteroscedasticity in terms of covariates. Algorithms already exist but they may face computational instability due to the required parameter constraints. Therefore, having a suite of tools based on a variety of computational methods is essential. Here we present VarReg, an R package that fits variance regression models using a stable expectation-maximisation approach, which can incorporate flexible splines. VarReg accommodates censored and skewed data and allows regression models for shape parameters as well as location and scale. VarReg’s functions are demonstrated with real-world examples, illustrating a valuable, stable package for fitting these complex regression models.