The (return value) class of an estimator is of class ALEstimate
if it is asymptotically linear; then it has an influence function
(implemented in slot pIC
) and so all the diagnostics for influence
functions are available; in addition it is asymptotically normal, so
we can (easily) deduce asymptotic covariances, hence may use these
in confidence intervals; in particular, the return values of kStepEstimator
oneStepEstimator
(and roptest
, robest
, RMXEstimator
,
MBREstimator
, OBREstimator
, OMSEstimator
in package
'ROptEst') are objects of (subclasses of) this class.
As the return value of CvMMDEEstimator
(or MDEstimator
with
CvMDist
or CvMDist2
as distance) is asymptotically linear,
there is class MCALEstimate
extending MCEstimate
by
extra slots pIC
and asbias
(only filled optionally with
non-NULL
values). Again all the diagnostics for influence
functions are then available. Classes ML.ALEstimate
and
class CvMMD.ALEstimate
are nominal subclasses of class
MCALEstimate
, nominal in the sense that they have no extra slots,
but they might have particular methods later on.
Helper method getPIC
by means of the estimator class, and, in
case of estimators of class CvMMDEstimate
, also the name
(in slot name
) produces the (partial) influence function:
calling .CvMMDCovariance
-- either directly or through wrapper
.CvMMDCovarianceWithMux
. This is used in the corresponding
.checkEstClassForParamFamily
method, which coerces object
from
class "MCEstimate"
to "MCALEstimate"
.