This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Handout 3 Analysis of variance approach to regression Consider the identity Y i Y = ( Y i Y ) + ( Y i Y i ) (1) Since Y i = Y i , we conclude that Y is the average of the fitted values { Y i } . So Y i Y is the deviation of Y i from the mean and Y i Y i is the residual. Square both sides of (1) and sum over i . Then it will turn out that X ( Y i Y ) 2 = X ( Y i Y ) 2 + X ( Y i Y i ) 2 + 2 X ( Y i Y )( Y i Y i ) . It can be shown that the cross product term ( Y i Y )( Y i Y i ) equals zero. So we have X ( Y i Y ) 2 = X ( Y i Y ) 2 + X ( Y i Y i ) 2 . This is an important identity. Here are some notations. SSTO = ( Y i Y ) 2 [ SSTO stands for total sum of squares], SSR = ( Y i Y ) 2 [ SSR stands for regression sum of squares], and SSE = ( Y i Y i ) 2 . So we have the following important identity SSTO = SSR + SSE. Degrees of freedom : Degrees of freedom for SSTO is n 1. Degrees of freedom for SSR is equal to # of beta parameters estimated  1 . Since we have estimated two beta parameters, df ( SSR ) = 1. Degrees of freedom for SSE is equal to n  #of beta parameters estimated . So, df ( SSE ) = n 2. Note that we have df ( SSTO ) = df ( SSR ) + df ( SSE ) . Mean squares: For any sum of squares, the mean square is defined to be the sum of squares divided by its degrees of freedom. So MSE = SSE df ( SSE ) = SSE n 2 , MSR = SSR df ( SSR ) = SSR, MSTO = SSTO df ( SSTO ) = SSTO n 1 ....
View Full
Document
 Summer '09
 Jiang

Click to edit the document details