---

Page 47 Exercise 2.8: The proof I had in mind for this I no longer think is valid.

Page 105 Exercise 3.9.7: the trick is that b_1 - b_2 has to be in C(U) so X(b_1-b_2) is in C(X_0), so if you premultiply by (I-M_0) you get 0.

Page 110 Fourth display, alpha's should have hats

Sec. 14.1.1 No reason the fitting method needs to be least squares. You could do forward selection to minimize a penalized loss function or information criterion.

Page 432 Line -7, "of squares," rather than "for squares"

Page 472 The last three displayed lines should begin with equal signs.

Page

Thanks to