Stats question!
Aug. 8th, 2014 08:22 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
kaberettThere is a standard ("Aldrich") that I measure ~10 times per mass spectrometry session, to make sure I'm getting approximately the right numbers out. I currently have ~104 measurements of this standard, with a standard deviation of +- 0.45 units.
I also measure samples, with number of (acceptable) measurements per sample between 2 and 13 inclusive.
Where the standard deviation between measurements for a sample is greater than my long-term stdev for the Aldrich, I use the standard deviation between measurements for the sample.
Where the standard deviation between measurements for a sample is less than that for Aldrich, I can either use the Aldrich stdev or I can use the measured stdev for the specific sample. How many measurements of a specific sample do I have to make before it becomes legit to trust the measured stdev for the sample rather than my long-term average for the Aldrich standard? Any pointers on how to work this out would be greatly appreciated - my supervisor is handwaving as "maybe ten?" but I'd like to have some kind of defensible reason for doing the thing...
I also measure samples, with number of (acceptable) measurements per sample between 2 and 13 inclusive.
Where the standard deviation between measurements for a sample is greater than my long-term stdev for the Aldrich, I use the standard deviation between measurements for the sample.
Where the standard deviation between measurements for a sample is less than that for Aldrich, I can either use the Aldrich stdev or I can use the measured stdev for the specific sample. How many measurements of a specific sample do I have to make before it becomes legit to trust the measured stdev for the sample rather than my long-term average for the Aldrich standard? Any pointers on how to work this out would be greatly appreciated - my supervisor is handwaving as "maybe ten?" but I'd like to have some kind of defensible reason for doing the thing...
(no subject)
Date: 2014-08-08 08:41 pm (UTC)(no subject)
Date: 2014-08-08 08:54 pm (UTC)(no subject)
Date: 2014-08-24 11:17 am (UTC)'about ten' is probably not good enough, because people tend to make up numbers like 10. 10 is a nice, friendly number, so people will pick it a lot.
However, there is a way of working out what 'about ten' should really be. This is stuff from process control (business stuff). Hopefully more information on this will be forthcoming soon, but this might be enough to start some googling.
(no subject)
Date: 2014-09-14 05:56 pm (UTC)