7.2Maximum Likelihood Estimators 231
In determining the maximizing value ofθ, it is often useful to use the fact that
f(x 1 ,...,xn|θ) and log[f(x 1 ,...,xn|θ)]have their maximum at the same value ofθ.
Hence, we may also obtainθˆby maximizing log[f(x 1 ,...,xn|θ)].
EXAMPLE 7.2a Maximum Likelihood Estimator of a Bernoulli ParameterSupposethatninde-
pendent trials, each of which is a success with probabilityp, are performed. What is the
maximum likelihood estimator ofp?
SOLUTION The data consist of the values ofX 1 ,...,Xnwhere
Xi=
{
1 if trialiis a success
0 otherwise
Now
P{Xi= 1 }=p= 1 −P{Xi= 0 }
which can be succinctly expressed as
P{Xi=x}=px(1−p)^1 −x, x=0, 1
Hence, by the assumed independence of the trials, the likelihood (that is, the joint
probability mass function) of the data is given by
f(x 1 ,...,xn|p)=P{X 1 =x 1 ,...,Xn=xn|p}
=px^1 (1−p)^1 −x^1 ···pxn(1−p)^1 −xn
=p
n 1 xi
(1−p)n−
1 nxi
, xi=0, 1, i=1,...,n
To determine the value ofpthat maximizes the likelihood, first take logs to obtain
logf(x 1 ,...,xn|p)=
∑n
1
xilogp+
(
n−
∑n
1
xi
)
log(1−p)
Differentiation yields
d
dp
logf(x 1 ,...,xn|p)=
∑n
1
xi
p
−
(
n−
∑n
1
xi
)
1 −p