11.4Tests of Independence in Contingency Tables 495
we obtain, after some computation, that with the five regions as given in the beginning of
this section,
ˆp 1 =.04214
ˆp 2 =.13346
ˆp 3 =.43434
ˆp 4 =.28841
ˆp 5 =.10164
Using the data valuesX 1 = 6,X 2 = 5,X 3 = 8,X 4 = 6,X 5 = 5, an additional
computation yields the test statistic value
T=
∑^5
i= 1
(Xi− 30 ˆpi)^2
30 ˆpi
=21.99156
To determine thep-value, we run Program 5.8.1a. This yields
p-value≈P{χ 32 >21.99}
= 1 −.999936
=.000064
and so the hypothesis of an underlying Poisson distribution is rejected. (Clearly,
there were too many weeks having 0 accidents for the hypothesis that the underlying
distribution is Poisson with mean 3.167 to be tenable.) ■
11.4 Tests of Independence in Contingency Tables
In this section, we consider problems in which each member of a population can
be classified according to two distinct characteristics — which we shall denote as the
X-characteristic and theY-characteristic. We suppose that there arerpossible values for
theX-characteristic andsfor theY-characteristic, and let
Pij=P{X=i,Y=j}
fori=1,...,r,j =1,...,s. That is,Pijrepresents the probability that a randomly
chosen member of the population will haveX-characteristiciandY-characteristicj.