9.4Statistical Inferences About the Regression Parameters 371
Confidence Interval Estimator ofα
The 100(1−a) percent confidence interval forαis the interval
A±
√√
√√
∑
i
x^2 iSSR
n(n−2)Sxx
ta/2,n− 2
Hypothesis tests concerningαare easily obtained from Equation 9.4.3, and their
development is left as an exercise.
9.4.3 Inferences Concerning the Mean Responseα+βx 0
It is often of interest to use the data pairs (xi,Yi),i=1,...,n, to estimateα+βx 0 , the
mean response for a given input levelx 0. If it is a point estimator that is desired, then the
natural estimator isA+Bx 0 , which is an unbiased estimator since
E[A+Bx 0 ]=E[A]+x 0 E[B]=α+βx 0
However, if we desire a confidence interval, or are interested in testing some hypothesis
about this mean response, then it is necessary to first determine the probability distribution
of the estimatorA+Bx 0. We now do so.
Using the expression forBgiven by Equation 9.3.1 yields that
B=c
∑n
i= 1
(xi−x)Yi
where
c=
1
∑n
i= 1
xi^2 −nx^2
=
1
Sxx
Since
A=Y−Bx
we see that
A+Bx 0 =
∑n
i= 1
Yi
n
−B(x−x 0 )
=
∑n
i= 1
Yi
[
1
n
−c(xi−x)(x−x 0 )
]
Since theYi are independent normal random variables, the foregoing equation shows
thatA+Bx 0 can be expressed as a linear combination of independent normal random