Skip to contents

Computes a heteroskedasticity-robust variance-covariance matrix.

Usage

# S3 method for class 'ral'
vcov(object, fit_idx = 1, type = "HC1", ...)

Arguments

object

An object inheriting from class ral.

fit_idx

Integer index of the fit. Defaults to 1.

type

Character. One of "HC1" (default), "HC0", or "HC3".

...

Currently unused.

Value

A \(p \times p\) variance-covariance matrix.

Details

Let \(\hat\phi_i\) denote the estimated influence function at observation \(i\). Three variance estimators are available:

HC0: $$V_{\mathrm{HC0}} = \frac{1}{n^2}\sum_i \hat\phi_i\,\hat\phi_i'$$

HC1 (default): $$V_{\mathrm{HC1}} = V_{\mathrm{HC0}} \times \frac{n}{n - p}$$

HC3: $$V_{\mathrm{HC3}} = \frac{1}{n^2}\sum_i \frac{\hat\phi_i\,\hat\phi_i'} {(1 - \hat{h}_{\theta,i})^2}$$

where \(\hat{h}_{\theta,i}\) is the leverage; see hatvalues.ral.

Cluster-robust inference. When cluster_variable is non-NULL and identifies fewer groups than observations, the observation-level influence functions are aggregated to cluster-level influence functions

$$\hat\Phi_g = \frac{G}{n} \sum_{i \in C_g} \hat\phi_i$$

and the variance is computed as

$$V_{\mathrm{HC0}} = \frac{1}{G^2} \sum_{g=1}^{G} \hat\Phi_g\,\hat\Phi_g'.$$