Subject: Help with uncertainty calculations
From: "Paul Chamberlain"
To: "ISO Guide 25 Discussion (E-mail)"
Date: Wed, 19 Apr 2000 17:31:50 -0600
We are a small laboratory preparing for our assessment to Guide 25. We
are in the process of reevaluating and recalculating our uncertainties.
Whenever we calibrate a test item we take several measurements (from 3
to 10). In our case it takes the majority of the time to setup the
calibration and it is relatively easy to take several readings. Each
measurement is in fact a separate calibration repeated under the same
setup under the "same" conditions. Obviously there is some variation in
the data and we calculate and report the average. We also calculate the
standard deviation and the data must fall within "limits" which are a %
of the standard deviation compared to the average.
What is the best/recommended way to use this data (short term noise) to
estimate and report a standard uncertainty that will then be combined
with other standard uncertainty estimates? We are already estimating
several Type B uncertainties that can be theoritically correlated to
environmental conditions, uncertainties of reference standards and
measurement equipment. In addition, we estimate an uncertainty that
accounts for long term drift which is obtained using control charts.
Thanks for your intput.
Paul Chamberlain LACO Technologies, Inc. 1-800-465-1004
paulc@lacotech.com www.lacotech.com
-----snippo-------------
Date: Thu, 20 Apr 2000 16:17:18 -0700 From: "Dr. Howard Castrup"
To: Greg Gogates Subject: RE:
Help with uncertainty calculations
Raul,
It sounds like you're doing a very thorough job in estimating
uncertainties. Ordinarily,the variation in the data you mentioned is
referred to as "random error." With regard to how to handle the random
error standard deviation computed from a sample of data, it depends on
whether you will be reporting the mean value obtained from the sample or
are attempting to "characterize" the random uncertainty for individual
values obtained with your measurement process.
If you are reporting the mean value, you equate the random error
standard uncertainty with the random error standard deviation divided by
the square root of the sample size. If you are characterizing your
process, you equate the standard uncertainty with the standard deviation
as computed from your data.
It is not likely that the random error in your process is correlated
with other sources of error (e.g., measurment system bias, operator
bias, environmental bias, etc.). If this is so, you can simply combine
the random error standard uncertainty in rss with other uncertainties.
If the random error in your process is correlated with one or more other
measurement errors, you need to take the correlations into account.
I hope this helps.
Howard Castrup
President, Integrated Sciences Group