timothyfoley_NENSP@yahoo. science forum beginner
Joined: 23 Apr 2006
Posts: 2

Posted: Sun Apr 23, 2006 10:09 pm Post subject:
Algebraic computation of the Hat diagonal



I would like to compute studentized residuals for predicted
concentrations in assays that use first order and second order
polynomial regression, as well as nonlinear 4parameter logistic
models.
Does anyone have an algebraic solution that can be used in Excel to
compute hii in these applications. I would like to use studentized
residuals to assess the prediction quality of calibration models used
in biomedical assays.
Basically, I would like to compute the following:
t = Cal (actual)  Cal (estimated)/RMSE * SQRT (1 hii).
RMSE is computed as SQRT [Sum (cal actual  cal estimated)^2/np]
The Hii statistic would adjust the ratio in the preceding equation for
leverage. Calibration points with t values that are greater than 2.5 or
3 would be candidates for elimination from the the current calibration.
I was initially going to ignore leverage issues, but since many of the
calibrations I deal with have a wide dynamic range, I thought I would
include hii in the estimate of t.
The main issue is that I need to find an algebraic approximation for
the matrix operations for ease of use. Can anyone supply formulas or
better yet a macro? Advice from the chemometrically inclined would be
most welcome.
Thanks,
Tim 
