That's just step 1. If the results are much higher than the preset target, it understates the higher value and overstates the variability if we plot the data on a normal scale. But, if the scale were a log scale, the numbers would be much closer together and the max values would be lower. Or, better, we might use the Poisson.
Not mentioned yet is how we would regularly verify the calibration.
Step 2 is the gathering of the samples--this is a real quagmire, since we're not sure what or where to sample.
Etc. etc. etc. all this contributes to error, knocking down those higher numbers.
I really like what Onlvee found--that's why it needs really fine toothed scrutiny.
Edited 1 time(s). Last edit at 19-Jun-16 23:33 by drrayeye.