This preview shows page 1. Sign up to view the full content.
Unformatted text preview: 195 SECTION 8.3
250 (d) 200
R4 The correlation coefﬁcient between R3 and R4
is 0.997. 1 100
0 20 R3 40 60 1 (e) R3 and R4 are nearly collinear.
(f) The cubic model is best. The quadratic is inappropriate because the residual plot exhibits a pattern. The residual
plots for both the cubic and quartic models look good, however, there is no reason to include R 4 in the model
since it merely confounds the effect of R3 .
1 Section 8.3
1. (a) False. There are usually several models that are about equally good.
(c) False. Model selection methods can suggest models that ﬁt the data well.
(d) True. 3. (v). x2 , x1 x2 , and x1 x3 all have large P-values and thus may not contribute signiﬁcantly to the ﬁt.
2 5. (i). Rest, Light*Rest, and Sit*Rest all have large P-values and thus may not contribute signiﬁcantly to the ﬁt. 7. The four-variable model with the highest value of R2 has a lower R2 than the three-variable model with the
highest value of R2 . This is impossible. ...
View Full Document