10.3One-Way Analysis of Variance 449
The value of the test statistic is thus
TS=
863.3335/2
1991.5785/12
=2.60
Now, from Table A4 in the Appendix, we see thatF2,12,.05=3.89. Hence, because
the value of the test statistic does not exceed 3.89, we cannot, at the 5 percent level of
significance, reject the null hypothesis that the gasolines give equal mileage. ■
Let us now show that
E[SSb/(m−1)]≥σ^2
with equality only whenH 0 is true. So, we must show that
E
[m
∑
i= 1
(Xi.−X..)^2 /(m−1)
]
≥σ^2 /n
with equality only whenH 0 is true. To verify this, letμ.=
∑m
i= 1 μi/mbe the average
of the means. Also, fori=1,...,m, let
Yi=Xi.−μi+μ.
BecauseXi.is normal with meanμi and varianceσ^2 /n, it follows thatYi is normal
with meanμ.and varianceσ^2 /n. Consequently,Y 1 ,...,Ymconstitutes a sample from
a normal population having varianceσ^2 /n. Let
Y ̄ =Y.=
∑m
i= 1
Yi/m=X..−μ.+μ.=X..
be the average of these variables. Now,
Xi.−X..=Yi+μi−μ.−Y.
Consequently,
E
[m
∑
i= 1
(Xi.−X..)^2
]
=E
[m
∑
i= 1
(Yi−Y.+μi−μ.)^2
]
=E
[m
∑
i= 1
[(Yi−Y.)^2 +(μi−μ.)^2 +2(μi−μ.)(Yi−Y.)
]
=E
[m
∑
i= 1
(Yi−Y.)^2
]
+
∑m
i= 1
(μi−μ.)^2 + 2
∑m
i= 1
(μi−μ.)E[Yi−Y.]