F-Test

[dresdentl]

Cmax/Cmin= 250 as I mentioned.

s^2(of Cmax)/s^2(of Cmin) = 34911,11.

Cmax (15 points) and C min (11 points). Therefore, according to F-test the Ratio s^2(of Cmax)/s^2(of Cmin) shouldn't be higher then about 4,7 .

Therefore you have an extreme inhomogenity of variances (s^2) and weighting with s^2 is necessery.
If you reduce the range of analyses to Cmax/Cmin= 32, then s^2(of Cmax)/s^2(of Cmin) = 406. The ratio then shoudn't be higher then 4,85 and therefore weighting is still necessery.

The data in my last posting are derived from simfit which I used for calculationg weighted linear regression.

[tom jupille]

Theoretical value: In my case I have nine calibration points: 1,2,4,8,16,32,64,128,256 ng/ml. The 9 calibration points concentrations represent the theoretical value for each point. If you calculate the concentrations using the linear equation derived from linear regression (least squares) reversely, then you get your analyses concentrations based on your calibration curve. The difference between C(theoretical) and C(linear equation) is the error. The % error should be ≤ 20% at the lowest calibration point and ≤ 15 % at the other calibration points. At least 2/3 of the data should fulfill these requirements.

The simfit datas which I posted before use residuals. The residuals are between 10,56 % and 1,11%. ,
The % error calculated as described before is maximal 10% (average 4,6 %). These is very similiar to the residuals mentioned in the simfit data.

On the other hand, I read there is some difference between % error and residuals. Does someone no the exact difference?

I can also give further details if someone is interested.

.

[tom jupille]

There is a great series of articles on Statistics in Analytical Chemistry that runs every couple of months in American Laboratory. Archived articles are available for download a pdf files, but you do have to register with them in order to download. I'd recommend that you check out the following as relevant to your problem:

Least squares:
http://tinyurl.com/2e9fxu7

Calibration diagnostics (series of three articles):
http://tinyurl.com/24bt6z7
http://tinyurl.com/29x9k5w
http://tinyurl.com/249xc6c