STAT3012 Lecture Notes - Lecture 5: Standard Deviation, Forb, Sampling Distribution
Lecture 5 - Inference and prediction in simple linear regres-
sion
New concepts
✷Hypothesis testing for β0and β1
✷Confidence intervals for β0and β1
✷Confidence interval for σ2
✷confint() command
Applied Linear Models: Lecture 5 1
find more resources at oneclass.com
find more resources at oneclass.com
Theory – Sampling distribution of b
β1
✷Standardizing the normal sampling distribution of b
β1gives
b
β1−β1
σ/√Sxx ∼N(0,1).(1)
✷Replacing σ2with the estimate s2=RSS/(n−2) changes the sampling distri-
bution of b
β1(for a proof see Kutner et al; 2005, p45):
b
β1−β1
SE(b
β1)∼tn−2,(2)
where
◦the notation tn−2signifies a t-distribution with (n−2) degrees of freedom
(df);
◦the term SE(b
β1) = s
√Sxx is the standard error of b
β1; i.e. the (estimated)
standard deviation of the sampling distribution of b
β1.
Applied Linear Models: Lecture 5 2
find more resources at oneclass.com
find more resources at oneclass.com
Theory – Sampling distribution of b
β1
✷Equation (2) forms the basis for testing hypotheses and constructing confidence
intervals for β1.
✷For testing H0:β1=β0
1, against H1:β16=β0
1, use the t-test statistic
t=b
β1−β0
1
SE(b
β1).(3)
✷Find the p-value corresponding to t. As usual, smaller p’s indicate more/stronger
evidence against H0.
✷A(1 −α)confidence interval for β1is
b
β1±tn−2(1 −α/2) ×SE(b
β1).(4)
Applied Linear Models: Lecture 5 3
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Lecture 5 - inference and prediction in simple linear regres- sion. Standardizing the normal sampling distribution of b 1 gives b 1 1. Replacing 2 with the estimate s2 = rss/(n 2) changes the sampling distri- bution of b 1 (for a proof see kutner et al; 2005, p45): b 1 1. Theory sampling distribution of b 1 intervals for 1. Equation (2) forms the basis for testing hypotheses and constructing con dence. For testing h0 : 1 = 0. 1, against h1 : 1 6= 0. Find the p-value corresponding to t. as usual, smaller p"s indicate more/stronger evidence against h0. A (1 ) con dence interval for 1 is b 1 tn 2(1 /2) se(b 1) . With similar steps we get that a 100(1 )% con dence interval for 0 is. Theory con dence interval for b 0 where the standard error of b 0 is given by b 0 tn 2(1 /2)se(b 0),