8.7Tests Concerning the Mean of a Poisson Distribution 331
REMARK
If Program 5.2 is not available, one can use the fact that a Poisson random variable
with meanλis, for largeλapproximately normally distributed with a mean and variance
equal toλ.
8.7.1 Testing the Relationship Between Two Poisson Parameters.............
LetX 1 andX 2 be independent Poisson random variables with respective meansλ 1 andλ 2 ,
and consider a test of
H 0 :λ 2 =cλ 1 versus H 1 :λ 2 =cλ 1
for a given constantc. Our test of this is a conditional test (similar in spirit to the Fisher-
Irwin test of Section 8.6.1), which is based on the fact that the conditional distribution
ofX 1 given the sum ofX 1 andX 2 is binomial. More specifically, we have the following
proposition.
PROPOSITION 8.7.1
P{X 1 =k|X 1 +X 2 =n}=
(
n
k
)
[λ 1 /(λ 1 +λ 2 )]k[λ 2 /(λ 1 +λ 2 )]n−k
Proof
P{X 1 =k|X 1 +X 2 =n}
=
P{X 1 =k,X 1 +X 2 =n}
P{X 1 +X 2 =n}
=
P{X 1 =k,X 2 =n−k}
P{X 1 +X 2 =n}
=
P{X 1 =k}P{X 2 =n−k}
P{X 1 +X 2 =n}
by independence
=
exp{−λ 1 }λk 1 /k!exp{−λ 2 }λn 2 −k/(n−k)!
exp{−(λ 1 +λ 2 )}(λ 1 +λ 2 )n/n!
=
n!
(n−k)!k!
[λ 1 /(λ 1 +λ 2 )]k[λ 2 /(λ 1 +λ 2 )]n−k
It follows from Proposition 8.7.1 that, ifH 0 is true, then the conditional distribution of
X 1 given thatX 1 +X 2 =nis the binomial distribution with parametersnandp=1/(1+c).
From this we can conclude that ifX 1 +X 2 =n, thenH 0 should be rejected if the observed
value ofX 1 , call itx 1 , is such that either
P{Bin(n, 1/(1+c))≥x 1 }≤α/2