Variance-covariance matrix of a LaMaModel object

Computes the approximate covariance matrix of the estimated parameters coef(object), as the inverse Hessian of the negative log-likelihood at the estimate.

Usage

# S3 method for class 'LaMaModel'
vcov(object, ...)

Arguments

object

model object of class "LaMaModel"

...

additional arguments (not used)

Value

covariance matrix with rows and columns named like coef(object)

Details

The Hessian is

• the exact Hessian obj$he() for models without random effects,

• for models with random effects, the Hessian of the Laplace-approximated marginal likelihood w.r.t. the fixed parameters, by finite differences of obj$gr() (as in sdreport),

• for models from qreml, the Hessian of the penalised negative log-likelihood, including the penalised coefficients. The covariance is then conditional on the estimated smoothing parameters.

If the Hessian is not positive definite, which indicates that the model has not converged, a warning is issued.

See also

coef.LaMaModel