% difference of peak area or % difference of SQRT peak area?

[eva1983]

Hi,

I have 2 or 3 GC replicates of the same sample (so, 2 or 3 injections of the same sample also called instrumental replicates) and I want to use only the data of the replicates that are similar enough. In order to do that, Im about to calculate the % of difference between the replicates and choose only those ones w <10% different. my question is: Do I have to calculate the % difference with the peak area or the SQRT of the peak area?

My calibration curve is calculated with the SQRT of the peak area because the peak area is non linear, is exponential. Therefore if I calculate % difference of the peak area, the difference that I will accept is going to depend on the centration

[tkubowicz]

Hello

As I understand you want to check injection precision. So just calculate RSD for number of injections you have. Area or SQRT do not matter.

Regards

Tomasz Kubowicz

[dap]

Hello.
First of all median instead of average could be used.
Another way is to use https://en.wikipedia.org/wiki/Dixon's_Q_test f.e. Maximal Q calculated would be related to the most outlier value.
If you have several injections from several samples https://en.wikipedia.org/wiki/F-test and https://en.wikipedia.org/wiki/Student's_t-test could be used to merge (or not to merge) all of them.
PS And also https://en.wikipedia.org/wiki/Measurement_uncertainty could be calculated and compared with % difference.

[Peter Apps]

What is the rationale for your approach versus taking the mean of all three results ?

Peter

[GOM]

"and choose only those ones w <10% different. "

This comment really concerns me - only choosing those data that fit with what you want

Ralph