Reproducibility and repeatability how to evaluate

[Cornelis]

Dear all,

SAS offers PROC VARCOMP where you can use it for calculation of repeatability and reproducibility using variance components.

I have downloaded a paper from Caroli that demonstrates the coefficient of repeatability/reproducibility calculations.

http://www.sascommunity.org/seugi/SEUGI1998/caroli_posters.pdf

The coefficient of reproducibility is defined as the probability for an analytical measure to be produced …

R = Var(Sample) / (Var(Sample) + Var(Error))   .. in %

Var(sample) = variation of sample, var(Error) = variation of error calculated by REML.

So far as I can see, coefficient of reproducibility represents the ratio between variation sample and total variation. A value of R = 93 % has been found in the paper and the questions are:

• What is acceptable criterion of coefficient of reproducibility? > 90 %?
• Is there any reference that will tell more about coefficient of reproducibility?
• How do we calculate the reproducibility, i.e. the error “± value” with 95 % confidence limit? Usually, the reproducibility is frequently defined as √2 * 1.96 * SD, where SD = standard deviation. This value is different from the coefficient of reproducibility and what can we do with both calculations?

Your contribution is highly appreciated.

Best regards,

Cornelis.

[PaigeMiller]

• What is acceptable criterion of coefficient of reproducibility? > 90 %?

I'm pretty sure this would be user judgment, based upon the understanding of the application being analyzed.

• Is there any reference that will tell more about coefficient of reproducibility?

Have you tried using a search engine?

• How do we calculate the reproducibility, i.e. the error “± value” with 95 % confidence limit? Usually, the reproducibility is frequently defined as √2 * 1.96 * SD, where SD = standard deviation. This value is different from the coefficient of reproducibility and what can we do with both calculations?

While I don't know the answer, it is unlikely that the reproducibility is normally distributed, so I doubt that the standard formulas apply here. But I don't really understand what question you are asking, specifically, you want a ± confidence interval for what quantity?