Biomedical Engineering Reference
In-Depth Information
Some minor modification is needed for 0 < t < h; but as h is the band-
width and will go to 0 with increasing n, the modification is on a vanishing
neighborhood of 0. These smoothed versions of theG
n;i
's lead to a smoothed
version of the empirical measure given by
d P
n
(u;) = d G
n;1
(u) + (1 ) d G
n;0
(u) :
This can be used to define a smoothed version of the log-likelihood function,
namely,
Z
f log F(u) + (1 ) log(1 F(u))gd P
n
(;u) :
l
n
(F) =
The MSLE, denoted by F
M
n
, is simply the maximizer of l
n
over all sub-
distribution functions and has an explicit characterization as the slope of a
convex minorant as shown in Theorem 3.1 of Groeneboom et al. (2010). The-
orem 3.5 of that paper provides the asymptotic distribution of F
M
n
(t
0
) under
certain assumptions, which, in particular, require F and G to be three times
differentiable at t
0
; under a choice of bandwidth of the form h h
n
= cn
1=5
,
it is shown that n
2=5
( F
M
n
(t
0
)F(t
0
)) converges to a normal distribution with
a non-zero mean. Explicit expressions for this asymptotic bias as well as the
asymptotic variance are provided. The asymptotics for f
M
n
, the natural es-
timate of f which is obtained by differentiating F
M
n
, are also derived; with
a bandwidth of order n
1=7
, n
2=7
( f
M
n
(t
0
) f(t
0
)) converges to a normal
distribution with non-zero mean.
The construction of the SMLE simply alters the steps of smoothing and
maximization. The raw likelihood, l
n
(F) is first maximized to get the MLE
F
n
, which is then smoothed to get the SMLE:
Z
F
S
n
(t) =
K
h
(tu) d F
n
(u) :
Again, under appropriate conditions and, in particular, twice differentia-
bility of F at t
0
, n
2=5
( F
M
n
(t
0
) F(t
0
)) converges to a normal limit with a
non-zero mean when a bandwidth of order n
1=5
is used and the asymptotic
Search WWH ::
Custom Search