374 Chapter 9: Regression
the future response whose input level isx 0 and consider the probability distribution of the
response minus its predicted value — that is, the distribution ofY−A−Bx 0. Now,
Y∼N(α+βx 0 ,σ^2 )
and, as was shown in Section 9.4.3,
A+Bx 0 ∼N
(
α+βx 0 ,σ^2
[
1
n
+
(x 0 −x)^2
Sxx
])
Hence, becauseYis independent of the earlier data valuesY 1 ,Y 2 ,...,Ynthat were used
to determineAandB, it follows thatYis independent ofA+Bx 0 and so
Y−A−Bx 0 ∼N
(
0,σ^2
[
1 +
1
n
+
(x 0 −x)^2
Sxx
])
or, equivalently,
Y−A−Bx 0
σ
√
n+ 1
n
+
(x 0 −x)^2
Sxx
∼N(0, 1) (9.4.6)
Now, using once again the result thatSSRis independent ofAandB(and also ofY) and
SSR
σ^2
∼χn^2 − 2
we obtain, by the usual argument, upon replacingσ^2 in Equation 9.4.6 by its estimator
SSR/(n−2) that
Y−A−Bx 0
√
n+ 1
n
+
(x 0 −x)^2
Sxx
√
SSR
n− 2
∼tn− 2
and so, for any valuea,0<a<1,
p
−ta/2,n− 2 <
Y−A−Bx 0
√
n+ 1
n
+
(x 0 −x)^2
Sxx
√
SSR
n− 2
<ta/2,n− 2
= 1 −a
That is, we have just established the following.