C:\Mis Documentos

C:Mis DocumentosWcalib1.pdf wcalib1.pdf (24 k)

muy bien - looks good - I'll read it tonight - muchas gracias - Uwe

I forgot to write that RDS=constant yields variance=constant*Y(x)^2 = constant*m^2*X^2 (because Y=mX), which would give a weighing factor proportional to 1/X^2.

Marcelo

Marcelo: Thank you, this was nice. I have been looking exactly for this kind of explanation. It is not voodoo, and a chemist can still understand it. The next time we have a discussion on this I know how to deal with it. - Uwe

A spreadsheet with a custom function to do weighted least squares linear fits in Excel. Using the routines in this spreadsheet, you can do fits to data with uncertainties, with a call to one function!

Sorry the file was too large and I couldnt upload the spreadshet. If you want it it is here: http://www.arsdigita.com/wp/display/33220/34136.wimpy

Let me recommend a short but very nice book
Statistics for Analytical Chemistry
by J.C. Miller, J.N. Miller

Others more advanced but not too difficult:
Applied Regression Analysis and Other Multivariable Methods
D.G. Kleinbaum - Lawrence L. Kupper

And finally a classic, a well known book that can be useful:
Understanding Robust and Exploratory Data Analysis
by Tuckey
This last is about robust statistical methods and data transforming.

There are some alternative methods for curve fitting based on robust methods which don't assume any distribution for errors: multiresponse factors, regression based on medians, etc. I've never tried them on routine analysis, it would be very nice to know if someone have used them.
I usually need log-log transforming for GC-MS calibration curves over broad analytical ranges (LOQ to 100-300 x LOQ) due to the non linear response of MSD.
Regards

Marcelo