ISO 17025 FAQ
Questions, and PQA's Answer

Submit a Question for ISO 17025 FAQ. Send your
question by email to dwhitred@pqa.net
Q1: InterLab Comparisons
Q2: What does
it take for ISO 17025 Certification
Q3: Measurement Accuracy & Units
Conversion 
I have some queries
regarding Inter lab comparison.
Firstly, when I'm the
originator and participating lab at the same time, how can I ensure that the
results obtained are correct? Is it necessary that I should know the actual
results before analysis? I mean if I'm getting 15% result of x sample and
the other lab is getting 14.8% result, whereas the actual value should be
around 20%, how would I interpret the results? Because I would judge the
both results, and if they are close to each other, I will consider it as the
final result. Therefore according to me, in this scenario, 14.9% would be
the final value.
Also what is the outcome
of this activity? I mean after analysis and gathering results from all labs,
what should be my next step?
Waiting for your reply
Regards,
Amna Malik
Quality Assurance Officer
PQA's Answer
For interlab comparison, you are attempting to determine:
 the magnitude and trend over time in any errors;
 factors that are affecting the results (ie. age of reagents, lab
temperature, lab technician, etc.)
 the interlab calibration factors for mean and variance, and
comparison to the true value
Before starting your analysis, lets assume your lab technician knows that
the real sample value (ie. as tested by a National Laboratory is 16.285),
and all other labs have already tested & submitted their results as 16.284,
16.286, 16.285, and 16.287 Do you think the lab tech will report some
drastically different value? I don't think so. If they get a very
different value, they will search forever to find their error, or secretly
retest to come up with the same answer as everybody else.
Will they do the same for an unknown customer's sample? I don't think so.
This prior knowledge defeats the purpose of the interlab tests. The lab
technician must not know the true sample value before analyzing the
sample, nor the results of others. This ensures their independence. Get
someone else trained (other than the person testing your sample) to prepare
your samples, collect the data from other labs, and secretly label the
samples so no one (except the person holding the secret key) can tell them
apart.
When you get the results from everybody, are some labs consistently higher,
others consistently low? If a lab tests the same sample divided into 10
equal parts, do they get the same answer for all 10 samples?
If you take 10 different samples randomly spread over the normal operating
range of the customer's samples, divide those samples into two (making 20
samples, 10 pairs), when labeled randomly so tester cannot tell which sample
is which (but you secretly know), can the tester tell which of the 20
samples are pairs? This experiment has replication, so you can estimate the
error of the testing method.
The data must be analyzed statistically to determine the error due to the
lab technician, error due to the test method itself, and random error.
Using DOE (Design of Experiments) will maximize the knowledge gained with
the minimum number of samples tested. If your test costs (eg. lab tech
salary, reagents, sample preparation time, sample cost, etc.) are
significant/expensive, you economically MUST use DOE. If you don't have the
skills, learn them, or get somebody hired/contracted who does.
PQA can perform this statistical analysis for you, or you can buy the
software, take a course, and do it for yourself. Otherwise, you can hire
someone locally. But somebody must do it.
One simple way to see what you have is to plot true sample value on the Y
axis, the lab's result on the X axis, and a different data point style (ie.
circle, square, etc.) for each lab. Of course, all data points should lie
on a 45 degree line, but this rarely occurs due to intra and interlab
errors.
If everything is fine today, how does it change over time? Summer
vs. winter? New lab tech vs. experienced veteran? The new test
machine vs. the old reliable machine?
Once you have the statistical analysis done, you can calculate the
calibration factor for each participating lab, lab technician, analysis
machine, etc. One way of doing it (it may not be best for you) is to report
the true value, as calculated by the calibration factors, so that no matter
how & where the test is done, you know the true value, and this true value
is what is reported to the client. This takes a lot of data, tests, and
effort on an ongoing basis, to ensure the calibration stays consistent.
Glenn Black P.Eng. CQE CQA

Good evening, I am
interested in finding out what it takes to become an ISO 17025 Certified
Laboratory.
Allen Hardison, Los Angeles California, USA
PQA's Answer
I quickly toured your website. I assume your inquiry is related to
_________, and you are planning a lab, or getting certified for an existing
lab.
Our role is to help you get ready for 17025 certification as quickly and
cheaply as possible. Most people waste precious resources in the
certification as it's their first time. After you have done as many as us,
you get good and fast, without cutting corners. If fact, you get optimum
implementation that helps your operation improve its competitiveness, rather
than barely meeting the minimum requirements. Find someone locally
with similar outlook and experience, or hire PQA.
We would start by doing an onsite audit, learn what you need, then design a
custom implementation strategy. We then help you get started, getting you
over all the hard items before we leave. This would take at least 2 days on
site, to as much as 5 days, depending on your circumstances. After that,
you get ongoing support for whatever you need via phone, fax, email, and
Internet until you get your certification.
We can help you do an Internal Auditing training course while we are onsite,
or you can do it later online for our website. Either way, you can take
care of your system as you design and implement it.
You can get certified in 4 weeks to 6 months, depending on where you're
starting from.
To discuss further, give me a call
Glenn Black P.Eng. CQE CQA

Don't
know if you can help me out here, but here goes...
I am working on my Master's
thesis, and I am confused as to how to proceed here. I have read a number of
citations on rounding error/sig. digits, but I am still at a loss at to how
to proceed. Perhaps you could point me towards some reference I could use.
Here is my problem. I have measured the inside bend radius of my test
samples using a set of radius gauges. The gauges are incremented in 1/64"
steps, so it's a case of 'best fit' to a particular gauge. All measurements
were either 7/64", 1/8", or 9/64". Now, I have to present these data in
decimal inches and in SI, so 1/8" would become 0.125". But, converting 7/64
or 9/64 using the same conversion factor gives 6 decimal places before
resolving to zeros, and keeping that many figures seems wrong (One
publication I read stated that unit conversions should retain all figures,
or something to that effect). Also, I have to convert to SI, so again, I
have a question as to how many digits to retain. A more fundamental issue (I
think) is that I can only resolve 1/64", so wouldn't my measurement for 1/8"
be 1/8" plus or minus 1/128"? And again, how would one properly represent
these measurements in SI. It's kind of frustrating, because 1), I feel I
should know this, and 2) I feel I should be able to find some reference that
discussed these issues. So, I'm hoping you have the time to point me in the
right direction. Maybe I'm not framing the question properly in the search
engines I am using.
Many thanks,
Vic Lewis
PQA's Answer
They are certainly good questions, and many people get stumped on these
same issues.
In conversion, there are a few schools of thought.
 There is enough error in the original measurement. You don't
need to introduce additional error by units conversion.
 To be reversible, you need a sufficient number of digits. For example
to go from Imperial to SI and back again to Imperial, can you regain the
same original value? How many significant figures do you have to
carry to retain the accuracy of the original measurement in this cyclic
conversion?
 As your gauge is in ranges, the extreme measurements included in this
bracket can be converted to SI with maximum digits retained, then take the
median, then round off to the level of the original Imperial measurement.
If the measurement gauge increment is 1/64", you can interpolate to +1/128"
(half an increment, or + 0.007813) as you stated. Which means you
really know the true usually within 1/3 to 1/10 when we allow for all the
other sources of error (need R&R study to know for sure), so your true
sample size is + 0.023438 or as close as +0.07813. The
number of significant digits should be able to resolve to this level.
If all samples measured from 7/64" to 9/64" you have a range of 1/32", so
you should have a measurement gauge good to 1/320" (ie. can repeatedly
differentiate + 1/320" with a 95% probability, including the person
eyeballing the samples, so as to get an Error of Measurement < 10%
of the process variation. Good luck on that. To me, this is the largest
problem, and everything else is a moot point, because garbage in, garbage
out.

Submit a Question for ISO 17025 FAQ. Send your
question by email to dwhitred@pqa.net 
