HPLC calibration curve - in general

[Norbert]

Hi,

I was wondering about a quite fundamental problem, but it is right now a discussion in our lab. What is in your opinion the bottom threshold for a acceptable calibration curve (R square)? 0.999 / 0.99x or 0.98. Thanks for your help in this.

Cheers Norbert

[aceto_81]

For our validation, we don't even use Rsquared to accept linearity, we use a lack of fit test.

Rsquared doesn't say something about the fit, only a bit about the spreadness, so it depends on which spread you want to allow.

Other method of calculation we sometimes use:
%RSD of Area/conc over the whole calibration range.

Ace

[JGK]

Our lab uses r > 0.995 which would equate to R2 > 0.997

[DR]

While there are several mentions of the benefits of using a weighted least squares fit (search those words & Tom for more), we also tend to use r²>0.99 as a suitability specification in our labs whenever a curve is used.

[Alfred88]

Dear Norbert:
For routine tests, we require R>= 0.98.
For validations, we normally require R>=0.99; or 0.995 (depends). We may look at the residuals.
Alfred

[Noser222]

A reference I have says 0.98 for impurities and related substances and 0.99 for the active.

[mbicking]

1. In my understanding of statistics, one should only use r^2, not r. I believe that is the correct form, based on the equations involved.

2. As others have noted, r^2 is not a perfect indicator of calibration quality. "It is a good place to start, but a bad place to stop." I can show you curves with r^2 > 0.999 where the analyzed concentration for some of the standards is off by more than 50%.

My preference for evaluation:
a. r^2
b. residuals plot
c. zero intercept test

The fact that regulatory bodies only specify r^2 as a criterion reflects the statistical ignorance of the individual(s) who wrote the document. Actually, it reflects the statistical ignorance of our society in general, but that's a little off topic.

[Norbert]

By far too late!! But thanks for your posts!!

/Norbert