
Linear Regression ◾ 199
10.8.1 Confidence Intervals on Regression Parameters
Condence intervals can be calculated for the k (= p + 1) regression param-
eters (including the intercept) using the formula
ˆ
(/,)
,/
βα
α
j
tnkt
nk
±− −
−−
12
12
(10.24)
More advanced textbooks, such Johnson and Wichern (2007), give formulas
for condence ellipsoids. Condence ellipsoids give joint condence regions
for the p-estimated regression parameters that have a minimum area.
10.8.1.1 Extrapolation
It is extremely dangerous to use predictors outside of the joint range of
experimental data. For observational studies where data are typically col-
linear, this is ver