3. Mean squared error
Isabel Molina, J.N.K. Rao and Gauri Sankar Datta
Previous | Next
Note that the BLUP
of the small area mean
is a linear function of
Hence, its MSE can be easily calculated and it
is given by the sum of two terms:
where
is due to the estimation of the random area
effect
and
is due to the estimation of the regression
parameter
with
However, the EBLUP
given in (2.7) is not linear in
due to the estimation of the random effects
variance
Using a moments estimator of
Prasad and Rao (1990) obtained a second order
correct approximation for the MSE of the EBLUP. Later, Datta and Lahiri (2000)
and Das, Jiang and Rao (2004) obtained second order correct MSE approximations
under ML and REML estimation of
When using the REML estimator of
their approximation to the MSE, for large
is given by
where
Note that as
and
so
is the leading term in the MSE for large
However, for small
is approximately zero and then
might be the leading term for small
For example, taking only one covariate
with constant values
and constant sampling variances
and letting
we obtain
and
that is,
is twice as large as
Datta and Lahiri (2000) obtained an estimator of the MSE
of the EBLUP
given by
The MSE estimator (3.2) is second-order unbiased in the
sense that
In the case that
the BLUP
of
becomes the regression-synthetic estimator
But surprisingly, the approximation to the MSE
of the EBLUP given in (3.1) can be very different from the MSE of the synthetic
estimator. Note that the latter is
because
is strictly positive even for
In fact, in the simple example with
only one covariate
with constant values
and constant sampling variances
we have
whereas the approximation to the
MSE of the EBLUP given in (3.1) with
gives
three times larger. It turns out
that (3.1) is not a good approximation of the MSE of the EBLUP when
and, instead, we should use
Moreover, since for
this quantity does not depend on
any unknown parameter, we can take it also as MSE estimator, i.e., we can take
In practice, the true value of
is not known but we have the consistent
estimator
When
the EBLUP becomes the regression-synthetic
estimator for all areas, that is
In this case,
for all areas and the MSE estimator given in
(3.2) reduces to
Thus, the MSE estimator given in
(3.2) can be seriously overestimating the MSE for
To reduce the overestimation, we consider a
modified MSE estimator of
given by
where
In fact, for
close to zero, it may happen that
is closer to the true MSE than the full MSE
estimator
but the question of when is
close enough to zero arises. This question
motivates the use of a preliminary testing procedure of
to define alternative MSE estimators of the
EBLUP in Section 4.
Previous | Next