varIdent

varIdent is a variance function class used primarily in the R nlme package to model heteroscedasticity by allowing different residual variances for different levels of a grouping factor. It is one of several varFunc structures provided to flexibly describe how the variance of observations changes with covariates or groups when fitting linear mixed-effects (lme) or generalized least squares (gls) models.

The typical use is to specify varIdent(form = ~1 | group) as the weights argument in lme or gls; this requests separate multiplicative variance factors for each level of the grouping factor. One level is taken as a reference (with factor 1) and the remaining levels have estimated variance ratios relative to that reference. These ratios are estimated jointly with model parameters, usually by maximum likelihood or restricted maximum likelihood. The resulting object supplies methods for extracting coefficients, fitted variance weights, and updating the structure.

varIdent is useful when residual variability is not constant across strata but does not depend smoothly on a covariate. It complements other variance functions in nlme (for example, varPower or varExp) which model variance as a continuous function of covariates. By permitting level-specific variances, varIdent helps improve inference and standard-error estimates when heteroscedasticity follows group structure.