(b) Let
A
and
D
be nonsingular matrices of orders
k
and
m
, respectively,
B
be
k\times m
, and
C
be
k\times m
. Then, provided that the inverses exists
(A+BDC^('))^(-1)=A^(-1)-A^(-1)B(D^(-1)+C^(')A^(-1)B)^(-1)C^(')A^(-1)
. The inverses of the scatter matrices can be expressed as
(x^(')x)^(-1)=(x_((i))^(')x_((i))+x_(i)x_(i)^('))^(-1)
and
(x_((i))^(')x_((i)))^(-1)=(x^(')x-x_(i)x_(i)^('))^(-1)
, where
x_((i))
denotes the design matrix with the
i
th observation deleted. Denote the predicted leverage as
h_(i(i))=x_(i)^(')(x_((i))^(')x_((i)))^(-1)x_(i)
and deduce the relationship between the leverage values
h_(ii)=x_(i)^(')(x^(')x)^(-1)x_(i)
and the predicted leverage
h_(i(i))
, i.e., express
h_(i(i))
as a function of
h_(ii)
. (c) Consider the studentized residuals (
t_(i)^(**)
) and the deleted (externally) studentized residuals
t_(i)
diagnostics. (i) Name the data aberration(s) diagnosed by
t_(i)^(**)
and
t_(i)
, respectively. (ii) Give the respective expression (formula) for
t_(i)^(**)
and
t_(i)
. (iii) Argue which one is a better measure between
t_(i)^(**)
and
t_(i)
. (6)