posted 12 June 2001 04:20 PM              

They just showed me what they were doing to calculate uncertainty -- it looked close and might be OK -- but I thought I'd check to see what more experienced people thought.

They calculated uncertainty based on the uncertainty of the lab standard weight and the standard deviation of three reps of ABBA sequences (A= standard ; B= unknown). THey calculated a standard deviation based on six differences (A's - the adjacent B's).

posted 13 June 2001 08:57 AM              

Calculations at this point are based on three repeats of six comparisons of working standard to lab standard. Here's an example of the data

1g
A 0.99997
B 1.00015
B 1.00021
A 0.99998
A 0.99995
B 1.00020
B 1.00019
A 0.99987
A 0.99993
B 1.00019
B 1.00014
A 0.99991
A&B DIFF 0.00024

MEAN A 0.99994
MEAN B 1.00018

Uncertainty was calculated by determining the standard deviation of the lab standard's weight (not the uncertainty of the differences between the lab standard and the reference) and combining that with the standard deviation of the reference.

I see referenced in EA-4/02 that the difference within each ABBA grouping should be considered a single observation. If I directed the lab in that direction, there would only be three observations. Since they calibrate many similar sets of weights to the same reference, can I pool the standard deviations from similar comparisons (e.g., all similar 1 gram weight references)?

posted 14 June 2001 10:27 AM               

Subject unit is a class F 1 gram mass.

Measurement parameter is 1 gram reference mass calibrated with a stated uncertainty of 0.01832 mg from the calibration source.

five readings on the class F mass yield standard deviation of .005 mg at 95.0 confidence level.

next I use uncertainty=(xmass*(1-(air density/ref mass density)))/(1-(air density/mass density\))-.001

combining these values gives uncertainty of .018mg with infinite degrees of freedom.

root sum square of the component uncertainties give combined uncertainty of .0259mg at 95% confidience level and 348 degrees of freedom.

Hope this helps

posted 14 June 2001 11:26 AM               

Hope this helps,

posted 14 June 2001 03:09 PM              

quote:

---

Originally posted by tomvehoski:
Try downloading the freeware proficiency calculator from http://www.proficiency.org. It is in the downloads section. There are examples for mass and scale calibration, along with many others. It was very useful on a recent 17025 audit for a scale calibration company I consulted for.

Hope this helps,

Tom

---

Tom,

I downloaded the software -- looks real good.
I have a few things to figure out -- (I'm referring to the UC mass 10 kg nominal.unc file. For example 'comparator linearity'.

This example and the replay from Greg have helped me a lot. We were initially going to ignore boyancy -- but I can see that I need to take that into account.

I'm sure we will be using this software. I like that examples are included.

Debbie

posted 14 June 2001 03:13 PM              

quote:

---

Originally posted by DICKIE:
Debbie,
I'll give you some of the details of a 1 gram analysis I just did.

Subject unit is a class F 1 gram mass.

Measurement parameter is 1 gram reference mass calibrated with a stated uncertainty of 0.01832 mg from the calibration source.

Greg,

Thanks for your help. I've looked up a lot of information on boyancy -- I see that I can't ignore it. I'll have to go down to the lab and check out conditions to figure out the air density. I already know temperature and humidity.

I really appreciate you taking the time to review the information in my post and reply.

Thanks again.

Debbie

five readings on the class F mass yield standard deviation of .005 mg at 95.0 confidence level.

next I use uncertainty=(xmass*(1-(air density/ref mass density)))/(1-(air density/mass density\))-.001

combining these values gives uncertainty of .018mg with infinite degrees of freedom.

root sum square of the component uncertainties give combined uncertainty of .0259mg at 95% confidience level and 348 degrees of freedom.

Hope this helps

Greg

---