Joint Mahalanobis influence on the fitted \((\hat{\beta},
\hat{\sigma}, \hat{\theta})\). With groups = NULL (default) the
unit is the observation: the per-observation influence vectors are
stacked into a \((p + 1 + L) \times n\) matrix \(W\), and the
result is \(\sqrt{w_i^T V^{-1} w_i}\) with \(V = (1/n) W W^T\).
With groups set the unit is the cluster: the
per-cluster influence functions \(\Xi = -J_{par}^{-1} S\) (\(S\)
the per-cluster score contributions from .scoreByCluster),
restricted to the \((\beta, \sigma, \theta)\) rows, are scored the
same way with \(V = (1/J) \Xi \Xi^T\). This flags whole-cluster
outliers that no single observation reveals -- e.g. before relying on
the bootstrap variance-component test of anova, which
is anti-conservative under group contamination. If \(V\) is
singular (a variance-component boundary) the Moore-Penrose
pseudo-inverse is used.
# S3 method for rlmerMod
cooks.distance(model, groups = NULL, IF = NULL, ...)Named numeric vector: one entry per observation
(groups = NULL) or per cluster level.
An rlmerMod object.
Cluster grouping for cluster-level Cook's distance:
NULL (default) for per-observation; TRUE for the
top-level grouping factor; or a factor / grouping-factor name to
aggregate over. Cluster-level requires a single or nested grouping
structure (crossed designs error, as for the variance-parameter
covariance).
Optional pre-computed implicitIF_full(model);
computed on demand if NULL.
Currently unused.
The full IF computation is the expensive part; pre-compute it once
via IF = implicitIF_full(fit) and pass it in if you need
cooks.distance and influence together.
implicitIF_full,
influence,
hatvalues