note6d - y i- y i ) 2 ] ( n-p ) Weighted Least Square...

Info iconThis preview shows pages 1–9. Sign up to view the full content.

View Full Document Right Arrow Icon
STAT5044: Regression and Anova Inyoung Kim
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Outline 1 Weighted Least Square
Background image of page 2
Weighted Least Square Only problem: nonconstant variance Linear model with heterogeneity: Y i = β 0 + β 1 x i + σδ i ε i Y i δ i = β 0 δ i + β 1 δ i x i + σε We want to minimize this form. i { y i δ i - ( β 0 δ i + β 1 δ i x i ) } 2
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Weighted Least Square min β 0 , β 1 i { Y i δ i - ( β 0 δ i + β 1 δ i x i ) } 2 = i = 1 [ δ - 2 i { y i - ( β 0 + β 1 x i ) } 2 ] = i = 1 w i { y i - ( β 0 + β 1 x i ) } 2 , where w i = c × δ - 2 i and c is the positive constant. The “best” weight is proportion to the inverse of the variance. This is so called “weighted least square” criterion.
Background image of page 4
Weighted Least Square New Y new Y = w 1 / 2 1 ··· w 1 / 2 2 ··· ··· w 1 / 2 n Y 1 Y 2 . . . Y n = W 1 / 2 Y New mean: w 1 / 2 1 x 1 1 x 2 . . . . . . 1 x n ± β 0 β 1 ² = W 1 / 2 X β To minimize | w 1 / 2 y - w 1 / 2 X β | 2
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Weighted Least Square ˆ β = (( w 1 / 2 x ) t ( w 1 / 2 X )) - 1 ( w 1 / 2 x ) t w 1 / 2 y = ( x t ( w 1 / 2 ) t ( w 1 / 2 ) x ) - 1 X t w 1 / 2 w 1 / 2 y = ( x t wx ) - 1 x t wy This is a linear form of y N ( X β , σ 2 w - 1 ) ˆ β N ( β , ( X t WX ) - 1 σ 2 )
Background image of page 6
Weighted Least Square ˆ σ 2 = [ n i δ - 2 i ( y i - ˆ y i ) 2 ] ( n - p ) = [ n i w i (
Background image of page 7

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 8
Background image of page 9
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: y i- y i ) 2 ] ( n-p ) Weighted Least Square Recall: Y i = + 1 x i + i i , i w-1 / 2 i r i = y i-( + 1 x i ) behave like i i r 2 i behave like 2 2 i 2 i r 2 i 2 i behave like 2 2 i Weighted Least Square Remark 1: Use WLS when the unequal variance is only problem. (heteroscedasticity: different variance) Remark 2: Give larger weight to the point with smaller variance As a result, one gets better efciency in and better power is testing. Remark3: Diagnostic plot should be performed on the weighted residual, w 1 / 2 i r i ....
View Full Document

Page1 / 9

note6d - y i- y i ) 2 ] ( n-p ) Weighted Least Square...

This preview shows document pages 1 - 9. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online