234 Chapter 7: Parameter Estimation
pˆi,i=1,...,m, as preliminary estimates of thepi. Now, letnf be the number of errors
that are found by at least one proofreader. Becausenf/Nis the fraction of errors that are
found by at least one proofreader, this should approximately equal 1−
∏m
i= 1 (1−pi), the
probability that an error is found by at least one proofreader. Therefore, we have
nf
N
≈ 1 −
∏m
i= 1
(1−pi)
suggesting thatN≈Nˆ, where
Nˆ = nf
1 −
∏m
i= 1 (1−pˆi)
(7.2.1)
With this estimate ofN, we can then reset our estimates of thepiby using
pˆi=
ni
Nˆ
, i=1,...,m (7.2.2)
We can then reestimateNby using the new value (7.2.1). (The estimation need not stop
here; each time we obtain a new estimateNˆofNwe can use (7.2.2) to obtain new estimates
of thepi, which can then be used to obtain a new estimate ofN, and so on.) ■
EXAMPLE 7.2c Maximum Likelihood Estimator of a Poisson ParameterSuppose X 1 ,...,Xn
are independent Poisson random variables each having meanλ. Determine the maxi-
mum likelihood estimator ofλ.
SOLUTION The likelihood function is given by
f(x 1 ,...,xn|λ)=
e−λλx^1
x 1!
···
e−λλxn
xn!
=
e−nλλn^1 xi
x 1 !...xn!
Thus,
logf(x 1 ,...,xn|λ)=−nλ+
∑n
1
xilogλ−logc
wherec=
∏n
i= 1 xi!does not depend onλ, and
d
dλ
logf(x 1 ,...,xn|λ)=−n+
∑n
1
xi
λ