Rajarshi Guha <[EMAIL PROTECTED]> wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > Hi, > I'm trying to implement some regression diagnostics. One of the > diagnostics (leverage) requires the hat matrix. > > I'm a little confused regarding the nature of this matrix. Is it supposed > to be derived from the augmented matrix (ie the design matrix with a first > column of ones) or is it derived from the design matrix?
If your model has an intercept, then you use the column of ones. If it doesn't, you don't. > Now we know that leverage values are the diagonal elements of the hat > matrix, so if the hat matrix is derived from the augmented matrix then > what does item does H(0,0) give the leverage for? (Since in this case > H(1,1) would give me the leverage for Observation 1, H(2,2) for > Observation 2 and so on) You're certainly confused. There are n diagonal elements of the hat matrix, no matter whether you have p or p+1 parameters. Glen . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
