368 Chapter 9: Regression
It now follows from Equation 9.4.2 that whenβ=1, the test statistic
TS=
√
8 Sxx/SSR(B−1)
has at-distribution with 8 degrees of freedom. The significance levelαtest will rejectH 0
when the value ofTSis sufficiently small (since this will occur whenB, the estimator of
β, is sufficiently smaller than 1). Specifically, the test is to
reject H 0 if
√
8 Sxx/SSR(B−1)<−tα,8
Program 9.2 gives that
√
8 Sxx/SSR(B−1)=30.2794(.4646−1)=−16.21
Sincet.01,8=2.896, we see that
TS<−t.01,8
and so the null hypothesis thatβ≥1 is rejected at the 1 percent level of significance. In
fact, thep-value is
p-value=P{T 8 ≤−16.213}≈ 0
and so the null hypothesis thatβ≥1 is rejected at almost any significance level, thus
establishing a regression toward the mean (see Figure 9.7).
A modern biological explanation for the regression to the mean phenomenon would
roughly go along the lines of noting that as an offspring obtains a random selection of
Father’s Height
72
70
68
66
64
62
60
Son’s Height
60 62 64 66 68 70 72 74
x
y
FIGURE 9.7 Example 9.4c for x small, y>x. For x large, y<x.