Method Validation - Linearity

[HW Mueller]

Whew...everybody probably got a headache by now. So statistics is for converting potentially depressing data analysis into a headache.
If Rutherford were still among us he would probably say

KISS

[bartjoosen]

HW Mueller:

bert,
sample data are in table 4, page 148, the explanation is in the text on p. 147 section (d)

This dataset is the one where everybody is talking about, but which I don't have

[HW Mueller]

As stasted above, this is just an example of a very obvious systematic error at low concentration levels of a standard curve, which the statistics didn´t "see".

(If you can´t get at J Chrom I will e-mail you the data, or whatever, if you want to. Contacts: Hans.W.Mueller@strz.uni-giessen.de or muhawyc@yahoo.com)

[bartjoosen]

Hans,

thanks for the article, now I can finally talk with you guys

So here we go:
It seems to me that table 4 contains 2 datasets of 2 different compounds.
So if I'm right, you can't calculate some ANOVA results with this kind of data. When you want to perform a lack-of-fit test and ANOVA, you should have repeated measures. When these repeated measures consequently indicate an higher measures value than the calculated value, ANOVA should indicate this.

Regards

Bart

[HW Mueller]

Sorry, right now I don´t have the time to learn to post the data or do a OCR and then present the data, but it´s not really neccessary. As stated before, a carryover caused the lowest concentration of a dilution/standard curve to be about double of what was in the standard, etc. The math can not recognize that as being a systematic error. The more linear curve (visibly, the corr. coeffs are all very high) was obtained after cleaning the Rheodyne so that no carryover occurred. (The stasndard injected where identical)