So, this time, I asked Pat about an instrument calibration. I was hoping he would give me a few industry agreed-upon limits for things, but I got so much more...
- 1. Error 1 --> There is no such thing as a "linear instrument"
a. 99% of instruments are on an "S-curve"
b. 1% are on "exponential" or at least locally exponential
A. You can never find 'zero'
For those who don't know math, zero is a placeholder. Analytically speaking, you can't have a count of zero. Zero is no count at all. Someone smart may argue with you and say, "Zero means something in binary notation." Zero still means nothing. Stop overthinking man!
2. What does this mean in the real-world of binomial distribution?
a. A calibration curve cannot go through the origin.
b. You cannot have negative values for the instrumentation
A. In geometry or theoretical physics you can have negative values
1. This is not geometry or theoretical physics.
a. Get over yourselves.
3. The top of the S-curve (see 1.a.) is where you'll break the darn detector
a. Pushing to show how high your peaks can be = burning out the detector
A. Choose your dynamic range to be linear (i.e. on the linear part of the S-curve)
b. So, we operate the instrument within our error range
A. The error range comes from some type of regulation
4. Some people like to have a high point and a low point and claim linearity.
a. This is like the old explorers finding the East Coast of Canada, then finding the West Coast of Canada and assuming all of Canada in between is flat. You run into problems.
b. Claiming linearity with too few points is problematic for calibration curves. This is why you want ≥ 5 point calibration for whatever curve or dynamic range you have.
--
Assuming I have fallen asleep on him, Pat reiterates, "SO ZERO IS NOT A POINT ON A CALIBRATION CURVE".
I wake up.
Pat continues.
--
5. One of the first things in calibration is to show linearity. The lowest point should be
a. below your detection level requirement for the method or,
b. maximum 2x the lowest detection level requirement.
A. Why the option for 2x DL? Because we assume it is not linear down there! = S-curve!
1. I'm now sounding like my priest on Sunday trying to teach a point to the congregation...
6. So, the error of your instrument can only be stated by the error of your calibration.
a. Therefore, if you're working toward 5% error, there is +/- 5% error on your measurements.
A. The laws of statistics don't let you around this unless you make tighter controls on your cal curve, but good luck with that. Sh-t happens.
B. This will be a significant factor when we talk about the horrors of significant figures.
1. Spoiler alert: How do you trust 1 decimal place with 5% error?
7. We always check if the calibration is controlled by running
a. a mid-level calibration standard
b. a low level calibration standard to ensure detection limit
--
This should be clear, even to Smiley.
~ Filtered by Dave Hope ~
