A2la G104 - 2014 PDF

A2LA
Document Revised:
G104 – Guide for Estimation of Measurement December 4, 2014
Uncertainty in Testing
Page 1 of 32
G104 - Guide for

Estimation of Measurement Uncertainty
In Testing
December 2014
2014 by A2LA
All rights reserved. No part of this document may be reproduced in any form or by any means without
the prior written permission of A2LA.
L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

A2LA
Document Revised:
Page 2 of 32
Table of Contents
I. INTRODUCTION ................................................................................................................................. 3
II. DEFINITIONS ..................................................................................................................................... 4
III. TENSILE STRENGTH EXAMPLE .......................................................................................................... 8
IV. INTRODUCTION TO THE GUM METHOD ......................................................................................... 10
a. GUM Terminology .................................................................................................................. 10
b. GUM measurement description............................................................................................... 11
c. Type A evaluation of standard uncertainty ............................................................................. 12
d. Type B evaluation of standard uncertainty .............................................................................. 12
e. Distribution.............................................................................................................................. 13
f. Type B summary ..................................................................................................................... 16
g. Uncertainty budget.................................................................................................................. 16
h. Reporting uncertainty .............................................................................................................. 24
APPENDIX A ........................................................................................................................................... 25
REFERENCES .............................................................................................................................................. 32

A2LA
Document Revised:
Page 3 of 32
I. Introduction
“Testing” is a term that covers a huge range of activities. Not every test is a measurement. However, for
those tests that are measurements and those that include measurements, uncertainty of measurement is an
important topic. The following is a simple introduction to the estimation of measurement uncertainty that
is applicable to testing in general1.
The purpose of measurement is to determine the value for a quantity of interest. Examples include the
concentration of alcohol in a blood sample, boiling point of water at 1 atmosphere of pressure, the
Rockwell hardness of a metal specimen, the tensile strength of a plastic compound, and the length of a
metal specimen at 20°C.
Calibrations are tests that compare indicated values to input quantities. A calibration is a measurement.
Looking something up in a reference book is not measurement. Nominal quantities such as hot, cold, or
pretty are not measurements. Measurements are processes that determine quantity values.
Before a measurement can be made, we have to know what we are to measure (the “measurand”), the
method and procedure to be used, the test conditions, the measurement devices and systems to be used,
and other relevant factors. (See VIM 2.1, GUM 3.1)
One of those relevant factors is the measurement uncertainty required. For example, lumber for a dog
house does not have to be measured as accurately as piston rods for automobile engines. When we report
the results of a measurement, it is important that we report the value and the uncertainty so that they are
understandable and relevant to the user.
Every measurement has uncertainty associated with it. Measurement devices, calibration standards,
reagents, and tools are not perfect. Environmental conditions, processes, procedures, and people are also
imperfect and variable.
In order for two measurements to be compared, both must trace back to a common reference. In order for
two measurement uncertainty statements to be compared, they must also both trace back to a common
reference. The appropriate method of measurement uncertainty calculation depends upon the nature of the
test and may be as simple or complicated as necessary to meet requirements.
Measurement uncertainty is important not only for calibrations but in any test that involves
measurements.
This guide is an introduction to test measurement uncertainty using the method of estimation described
in the JCGM 100:2008 Guide to the Expression of Uncertainty in Measurement (GUM)2.
1
The application of control charts for estimating measurement uncertainty is covered in Appendix A.
2
The type of analysis required for actual test measurement uncertainty estimation depends upon the nature of the
test. (See A2LA R205).

A2LA
Document Revised:
Page 4 of 32
A mechanical testing example is used for illustration in this guide but the methods presented are
applicable to many test situations.
II. Definitions3
Measurement uncertainty4: a non-negative parameter characterizing the dispersion of the quantity

values being attributed to a measurand, based on the information used.
Think of it as a parameter associated with the result of a measurement that characterizes the dispersion of
the quantity values that is attributed to the measurand.
Uncertainty in a measurement quantity is a result both of our incomplete knowledge of the value of the
measured quantity and of the factors influencing it. There are many possible sources of uncertainty in
measurement including5:
1) incomplete definition of the quantity being measured;
2) imperfect realization of the definition of the quantity being measured;
3) non-representative sampling;
4) inadequate knowledge of the effects of environmental conditions on the measurement or imperfect

measurement of environmental conditions;
5) personal bias in reading analog instruments, including the effects of parallax;
6) finite resolution or discrimination threshold;
7) inexact values of measurement standards and reference materials;
8) inexact values of constants and other parameters obtained from external sources and used in the data-
reduction algorithm;
9) approximations and assumptions incorporated in the measurement method and procedure;
10) variations in repeated observations of the measurand under apparently identical conditions.
These sources of uncertainty are not necessarily independent and some or all can contribute to the
variations in repeated observations. Not only can uncertainties be introduced by measurement equipment
and test methods, but also by the person performing the test, data analysis, the environment, and a host of
other factors.
3
International Vocabulary of Metrology – Basic and General Concepts and Associated Terms”,
JCM200:2008, BIPM
4
VIM 2.26
5
GUM 3.3.2.

A2LA
Document Revised:
Page 5 of 32
Measurement uncertainty in testing: Tests are performed in accordance with test procedures. The use
of recognized standard procedures (e.g., ASTM D638) eliminates many potential sources of measurement
uncertainty. Definitions, calculations, and other information necessary to evaluate the test data are
contained in such test procedures. The procedure addresses test measurement statistics and uncertainty at
the level necessary to meet test requirements. Some of the most commonly used terms and concepts
follow.
Standard deviation Repeated

measurements from a controlled process are
described by the Normal (or “Standard”)
probability distribution that yields an
average and standard deviation for the set.
The average value is usually taken as the
best estimate of the measured quantity. This
average is obtained from a number, n, of
test results according to the formula below.
If we designate each of the test results by
Figure 1
the symbol xi, the following equation gives
the average, x , of n test results:
1 n
 xi
1)
x
n i 1
The standard deviation, s, characterizes the variability, or spread, in the observed values xi. It is given by
n
 x  x 
2
i
s i 1 2)
n 1
Precision: is closeness of agreement between indications or measured quantity values obtained by

repeated measurements on the same or similar objects under specified conditions. Measurement
precision is usually expressed as standard deviation, variance, or some other measure of imprecision.
Measurement precision is used to define repeatability, reproducibility, and other statistics.
To evaluate test measurement precision, many test procedures require the testing organization to perform
a number of repeat measurements and compare the repeatability standard deviation or some other statistic
to values specified by the test method.
Control charts (Appendix A) provide important tools for controlling test process precision and bias.

A2LA
Document Revised:
Page 6 of 32
Repeatability6: is a condition of measurement, out of a set of conditions that includes the same
measurement procedure, same operators, same measuring system, same operating conditions and same
location, and replicate measurements on the same or similar objects over a short period of time.
This involves precision under repeatability conditions, i.e. conditions where test results are obtained with
the same method on the same or similar test items in the same laboratory by the same operator using the
same equipment at the same location within short intervals of time. Repeatability may be expressed in
terms of multiples of the standard deviation.
Repeatability standard deviation is the standard deviation of test results obtained under repeatability
conditions.
Reproducibility7: is precision under reproducibility conditions, i.e. conditions where test results are
obtained with the same method on the same or similar test items in different laboratories, by different
operators, using different equipment, in different locations, or on different days. Reproducibility may be
expressed in terms of multiples of the standard deviation.
Reproducibility standard deviation is the standard deviation of test results obtained under reproducibility
conditions. The conditions under which reproducibility is determined should be clearly specified.
Bias8: the estimate of a systematic measurement error.
Think of it as the difference between the test results and an accepted reference value. Known biases can
be corrected. Bias is often called “systematic error”, but corrected biases are not errors. There may be one
or more error components, known or unknown, contributing to the bias.
Many test procedures require laboratories to demonstrate that their measurement bias is within prescribed
limits by one of the following methods:
1) Reference material: Using an appropriate reference standard or material, the laboratory should
perform replicate measurements to form an estimate of its bias, which is the difference between the
mean of its test results and the certified value of the standard or material. If the absolute value of this
bias is less than twice the reproducibility standard deviation given in the precision statement in the test
method, then the laboratory may consider that its bias is under control.
2) Interlaboratory comparison: Laboratories participating in proficiency testing schemes will have

available to them data from a large number of laboratories which they can use to estimate the bias of
their measurement results. Comparison of the lab mean to the grand mean or other assigned value in
such programs, for example, will allow them to demonstrate that their bias is under adequate control.
For many test procedures, bias has not or cannot be evaluated due to the lack of appropriate reference
material. In such cases, this fact should be clearly documented.
6
VIM 2.20
7
VIM 2.2.5
8
VIM 2.18

A2LA
Document Revised:
Page 7 of 32
Trueness (Measurement Accuracy9): the closeness of agreement between a measured quantity value and
a true quantity value of a measurand.
Think of it as the closeness of agreement between the average value obtained from a large set of test results
and an accepted reference value. The measure of trueness is normally expressed in terms of bias.
Measurement method10: a generic description of a logical organization of operations used in a

measurement.
It refers to the general description of the measurement such as comparison, substitution, etc.
Measurement procedure11: a detailed description of a measurement according to one or more

measurement principles and to a given measurement method, based on a measurement model and
including any calculation to obtain a measurement result.
Think of it as a detailed description of the measurement process. If the laboratory is applying a standard,
validated test method, the test method may include definitions for statistical quantities and estimates of
precision and bias obtained by interlaboratory comparison during the course of method validation. For
example, ASTM test methods are often accompanied by tables of values determined from round-robin
reproducibility tests including multiple laboratories and data from repeatability studies performed in a
single laboratory.
Measurand: refers to the particular quantity to be measured in a test. Any uncertainty analysis must
begin with a clear understanding of the quantity to be measured, the measurand.
Systematic errors12: a component of measurement error that in replicate measurements remains

constant or varies in a predictable manner.
Think of these as biases that cause a measurement result to differ from the true value. Taking repeated
measurements and averaging them does not improve systematic error. Known systematic errors can be
corrected.
Random errors13: a component of measurement error that in replicate measurements varies in an

unpredictable manner.
These result from variations in the values of repeated measurements. Taking more repeated
measurements generally reduces the random error.
Corrected value: is the measurement result after systematic effects (biases) are removed.
9
VIM 2.13
10
VIM 2.5
11
VIM 2.6
12
VIM 2.17
13
VIM 2.19

A2LA
Document Revised:
Page 8 of 32
Intermediate precision condition of measurement14: a condition of measurement, out of a set of

conditions that includes the same measurement procedure, same location, and replicate measurements on
the same or similar objects over an extended period of time, but may include other conditions involving
changes.
III. Tensile strength example
The ASTM D638 tensile strength at break test can be used to illustrate the application of test measurement
uncertainty principles. This mechanical test will also be used to illustrate the GUM method15.
Tensile strength at break, “S”, is defined16 as the force (called “load” in the ASTM document), F, divided
by the cross sectional area, A, of the test specimen. The design and dimensions of the test specimen,
accuracy of the force indicator, and other factors related to the test are clearly identified in the standard
and are required to comply with specified tolerances and to be performed in a specified manner. The
standard also gives rules and tolerances for other important aspects of the test, including specimen
preparation, mounting, conditioning, and speed of testing.
S = F/A 3)
The cross-sectional area, A, is defined as the thickness, T, of the specimen multiplied times the width, W,
when these measurements are made as specified in the standard.
A=TW 4)
The tensile testing machine stretches the test specimen, continuously measuring the load until the specimen
breaks. According to the method, dimensional measurements must be made with an uncertainty of ±0.001
in. or less.
The table below summarizes ASTM D638 test specifications appropriate to this example. The standard
addresses a wide range of tests and specimens not addressed here.
Test Specifications
Load (Force) Indicator Thickness Width
±1% of indicated value 0.13 ±0.02 in. 0.50 ±0.02 in.
Table 1
Let us assume that we perform this test in accordance with the standard. Five specimens are measured for
thickness and width, and then tested in the machine. We record the dimensional measurements and the
load at break for each (see table below). Then, in accordance with D638, we calculate tensile strength, S,
14
VIM 2.23
15
ASTM D638 does not require the test performer to calculate test measurement uncertainty.
16
For clarity and consistency in this document, the symbols used in ASTM D638 are not used here. See ASTM
D638 A2.24.

A2LA
Document Revised:
Page 9 of 32
for the test by dividing the maximum indicated load by the average specimen cross-sectional area. S and
standard deviation, Sr, are also calculated. Sr is the “in-laboratory standard deviation”.
Test Measured Measured Area,

Measured Load, F Calculated S
Specimen Thickness, T Width, W TW
1 0.124 in 0.499 in 0.0619 830 lb 13409 PSI
2 0.126 in 0.501 in 0.0631 900 lb (Maximum) 14263 PSI
3 0.125 in 0.500 in 0.0625 810 lb 12960 PSI

4 0.126 in 0.500 in 0.063 870 lb 13810 PSI
5 0.124 in 0.499 in 0.0619 850 lb 13732 PSI
Average: 0.125 in 0.500 in 0.0625 852.0 lb 13635 PSI
Std. Deviation, Sr: 485 PSI
ASTM D638 Tensile Strength = 900/0.0625 = 14,400 PSI

Table 2.
The standard method defines repeatability as Ir=2.83Sr, and reproducibility17 is defined as IR=2.83SR in the
standard18. We calculated Sr for the measurements we made. Tables of Sr and SR based on round robin
studies are provided in the ASTM document for reference.
The standard states that judgments made in accordance with these definitions of repeatability and
reproducibility “will have an approximate 95% (0.95) probability of being correct.
The standard also explains that bias has not been established for this test method because no recognized
standards exist.
As illustrated by this example, even though measurement uncertainty may not be calculated for a particular
test, standard metrological and statistical definitions and methods apply and that even though systematic
errors (bias) remain indeterminate, reliable and consistent test results can be produced and compared
among performing organizations.
The same definitions and methods are used in the GUM, along with a few more that were developed
specifically for its purposes.
17
The standard uses “in-laboratory” and “between-laboratory” to distinguish repeatability from reproducibility. See
also ISO 21748: Guidance for the use of repeatability, reproducibility and trueness estimates in measurement
uncertainty estimation.
18
2.83 is the Student’s T value for 5 repeated measurements (4 degrees of freedom) at the two sigma level.

A2LA
Document Revised:
Page 10 of 32
IV. Introduction to the GUM method
The GUM method is not magic. Its application will not produce accurate estimates of measurement
uncertainty from bad tests or poor research. What the GUM does provide is a consistent method for
estimating measurement uncertainties. These words from that document summarize the situation well:
Although this Guide provides a framework for assessing uncertainty, it cannot substitute for
critical thinking, intellectual honesty and professional skill. The evaluation of uncertainty is
neither a routine task nor a purely mathematical one; it depends on detailed knowledge of the
nature of the measurand and of the measurement. The quality and utility of the uncertainty quoted
for the result of a measurement therefore ultimately depend on the understanding, critical
analysis, and integrity of those who contribute to the assignment of its value. (GUM 3.4.8)
The GUM method is an eight step process:
1. Describe the measured value in terms of your measurement process. (Model the measurement.)
2. List the input quantities
3. Determine the uncertainty for each input quantity
4. Evaluate any covariances/correlations in input quantities
5. Calculate the measured value to report
6. Correctly combine the uncertainty components
7. Multiply the combined uncertainty by a coverage factor
8. Report the result in the proper format
a. GUM Terminology
In addition to standard measurement, mathematical, and statistical terms and methods, the GUM uses
terminology specifically developed for its method. The following are key concepts and terms.
Standard uncertainty: In the GUM, all sources of measurement uncertainty are treated as if they are
standard deviations of probability distributions. Standard uncertainty is the uncertainty of a measurement
expressed as a standard deviation.
Input quantities: are the quantities that determine the The area of a square piece of material is calculated
measured value, sometimes called the “output quantity”. from two input quantities, length and width. These
quantities may be affected by influence quantities
Influence quantities: are parameters that affect the input such as temperature and the resolution of the
quantities and, through them, the measurement result. measuring instrument.
Type A evaluation (of uncertainty): is an evaluation of uncertainty by the statistical analysis of a series
of observations. (Type A uncertainties are not random errors.)
Type B evaluation (of uncertainty): is an evaluation of uncertainty by means other than the statistical
analysis of series of observations. (Type B uncertainties are not systematic errors.)

A2LA
Document Revised:
Page 11 of 32
Combined standard uncertainty: is calculated by squaring all the significant Type A and Type B
uncertainties, adding them together, and then taking the square root of the sum19. This is sometimes called
the “root-sum method”.
Expanded uncertainty: is the combined standard uncertainty multiplied by a coverage factor, k. The
expanded uncertainty defines an interval around the measured value. The value of the measurand is
expected to be within this interval to an established confidence level, usually 95%.
Each of these terms will be illustrated and described more thoroughly later in this guide.
b. GUM measurement description
A measurement is considered to be a function of the all the input quantities that affect the measurement.
Sometimes, as in the tensile strength example, the function is a known equation. For many tests, however,
the “function” is not well defined. The GUM assumes the measurement result, y, is caused by one or more
input quantities, which are designated x1, x2,.., xn acting through some functional relationship, f:
y  f ( x1 , x 2 ,..., x n ) . 5)
Returning to the tensile strength at break example, the measurement model is equation 3, which could be
written to explicitly show the input quantities as
S = f(F,T,W) = F/TW
6)
The input quantities are force (load), thickness, and width. In the sections following, we will use the GUM
process to evaluate the influence quantities that affect them, the uncertainty contributions that result, and
the method by which they should be combined.
The GUM assumes that the uncertainty in the measurement result can be calculated by combining the
uncertainties of the input quantities. Uncertainties in input quantities may result from more than one source
or influence. Among the influence quantities that may affect measurement uncertainty are the following:20
– Repeatability
– Resolution
– Reproducibility
– Reference Standard Uncertainty
– Reference Standard Stability
– Environmental Factors
– Measurement specific contributors
• Alignment, scale, evaporation, mismatch, etc.
– Contributions required by method
• ASTM, ISO/IEC, Military Procedure, etc.
– Accreditation requirements
If practical, input quantities should be varied to determine their effects on the measurement result. An
uncertainty estimate should be based, as much as possible, on experimental data. If available, check
19
This is a simplified definition that applies in many test situations. In cases where influence quantities are
correlated the calculation is more complicated and the GUM should be consulted.
20
A2LA R205

A2LA
Document Revised:
Page 12 of 32
standards, control charts, or other measurement assurance methods should be used to establish that a
measurement system is in statistical control.
One of the characteristic features of the GUM is its designation of all uncertainty contributors as Type A
or Type B. There are no other categories. Type A uncertainty estimates are derived from the statistical
analysis of test data21. Any uncertainty contributor that is not derived from statistical analysis of test data
is a Type B uncertainty contributor22. Type A and Type B uncertainty contributions, once determined,
are both “standard uncertainties”.
c. Type A evaluation of standard uncertainty
Type A uncertainties are based upon repeated measurements from a controlled process and are described
by the familiar Normal (or “Standard”) probability distribution that yields an average and standard
deviation for the set.
The formulas for average and standard deviation have already been listed (equations 1 and 2).
The GUM uses the term standard uncertainty, s, for the standard deviation of measurement results. The
same statistic is also frequently called “experimental standard uncertainty”.
n
7)
2)
 x  x 
2
i
s i 1
.
n 1
Taking more samples generally improves estimates of the average and standard uncertainty. The statistic,
degrees of freedom, is calculated from the number of measurements. For simple averages,
DOF= n-1. 8)
The standard deviation for the average of a set of 30 values has 29 degrees of freedom23.
It seems obvious that a standard uncertainty estimated based on 30 repeated measurements is likely to be
better than one based on 5, such as in the tensile test example. However, large numbers of repeated tests
may be expensive or otherwise impractical.
It is best to use a calculator or spreadsheet program for statistical calculations. The functions AVERAGE
and STDEV in Excel can be used to find the average and standard deviation of test results quickly and
easily. In Excel the standard deviation of the average can be calculated as STDEV/SQRT(COUNT).
d. Type B evaluation of standard uncertainty
21
Sr in the tensile test example is a Type A uncertainty estimate.
22
An uncertainty contributor based upon statistical analysis of another organization’s measurement data is Type B.
It is only Type A if the contributor is based on our own data.
23
Degrees of freedom can be calculated for curve-fits and other cases beyond the scope of this paper.

A2LA
Document Revised:
Page 13 of 32
Type A uncertainty estimates apply standard statistical methods to test data based on the assumption of a
normal probability distribution.
However, many factors that are not based on repeated measurements may contribute to uncertainty of
measurement. Values from reference books, manufacturer’s specifications, ASTM standards, experience,
and many other sources of uncertainty are included in this category that the GUM calls “Type B”.
The question of how to combine statistical and non-statistical uncertainty contributors had vexed the
measurement community for many years. The GUM provided a practical and creative answer: treat non-
statistical uncertainties as if they were statistical uncertainties with standard deviations. To do this,
probability distribution functions were developed for common non-statistical uncertainty distributions,
and the necessary formulas were developed for calculating standard uncertainties for each distribution24.
Type B uncertainties are assumed to have infinite degrees of freedom25 because they are not improved by
additional repeated measurements.
The most common distributions and the formulas for their standard uncertainties follow.
e. Distribution
It might seem surprising that the normal distribution would be a Type B non-statistical uncertainty
distribution. However, its use is very common and in calibration reports, national and international
standards, test procedures, manufacturer’s manuals, and many other technical documents.
Uncertainties based on normal distributions are Type B if they are not the result of our own measurement
data. (In the example, the Sr calculated from the five tensile tests is Type A. The SR values listed in
ASTM D638 tables are Type B.)
24
The methods used to determine Type B distribution functions are contained in the GUM but are beyond the
scope of this introduction.
From M3003 --
25
For this reason, it is very important to set the interval large enough to be certain that there is a negligible
probability that the uncertainty could be greater.

A2LA
Document Revised:
Page 14 of 32
For a standard uncertainty or standard deviation value of ±a reported at a coverage or multiplication

factor of k,26 the standard uncertainty is
uN  a / k .
9)
Returning to the tensile test example, the ASTM standard requires that the load indicator have an accuracy
of 1% or better. Assume we have a calibration certificate that shows an expanded uncertainty at the time
of calibration of 1% of indication, evaluated with a coverage factor of k=2, at a 95% confidence level.
For our average measured force value of 852 lb, ±1% is ±8.5 lb, which, by the equation above, gives us a
standard uncertainty of
8.5lb
uN   4.25lb. 10)
2
Rectangular distribution
a a The rectangular, or “continuous”, is another very
common distribution for Type B uncertainties. It
applies to tolerances, specifications, and reference book
Probability ‐>
values, among many other parameters. It is also the

default distribution to be used whenever the actual
distribution is not known.
Xi‐a xi Xi+a The equation for the standard uncertainty of a

continuous distribution of equal values between the
Xi, the expected value, lies in the
limits +a to –a, the standard uncertainty is
middle of an interval of range 2a.
Figure 2
a
uR  . 11)
3
The force indicator on our tensile test stand has a resolution of 1.0 lb. This resolution can be evaluated as
a rectangular distribution with containment limits of 0.5 lb. This gives a standard uncertainty of
0.5lb 12)
u R1   0.29lb.
3
In order to verify that the specimen thickness and width complied with the standard requirement of ±0.02
in, we measured and recorded the dimensions of the test specimens to the nearest 0.001 in. This uncertainty
is described by another rectangular distribution:
0.0005in 13)
uR2   0.00028in .
3
This uncertainty applies to both Width and Thickness input quantities.
26
A coverage factor of 1 corresponds to 68% containment, k=2 corresponds to 95%, and so on. These are the same
statistics as for standard deviations of a normal population.

A2LA
Document Revised:
Page 15 of 32
Triangular distribution
It may be the case that we know that there is a tendency for the values of an uncertainty contributor to be
near the center of an interval. For example, imagine two test specimens lying side by side on an aluminum
plate. Assume the room temperature is controlled with a control limit of ±a. After the specimens have
reached thermal equilibrium, the most likely value for the difference in temperature between them is zero.
-a xi +a
Figure 3. The triangular distribution is used to model cases where containment limits are
known and values are more likely to be near the mean than at the extremes.
The triangular distribution may be used in such a case. A rectangular distribution may be used instead,
but a slightly larger estimated uncertainty will result.
a 14)
uT  .
6
This distribution does not apply to the tensile strength example.
U distribution
This distribution models situations where the most likely value of a measurand is at or near the containment
limits. For example, because of the way thermostats work, room temperature tends to be near the
maximum allowed deviation from the set point, i.e., the room temperature is most likely to be too hot or
too cold relative to the set point. Applications of the U-distribution are also common in microwave and
RF testing. The equation for the standard uncertainty for this distribution is
a 15)
12) uU  .
2
This distribution does not apply to the tensile strength example.

A2LA
Document Revised:
Page 16 of 32
f(x)
-a 0 a
Figure 4. The U distribution models cases where the value of a measurand is likely
to be near the containment limits.
f. Type B summary
For containment limits a, the standard uncertainty estimates associated with the various Type B
probability distributions described in this guide are as follows:
a 16)
Rectangular:  0.5774a
3
17)
a
 0.4082a
Triangular: 6
a 18)
 0.7071a
U-Shaped: 2
19)
Normal u  a / k , when k = coverage factor
g. Uncertainty budget
A table listing the sources and values of uncertainty components is an uncertainty budget. It is a useful
tool but there is no mandated format and many forms are used. It is quite possible to evaluate and combine
uncertainty contributions without using a budget. Along with the budget table or other listing of the
constituent uncertainties, it is important that a well-documented narrative be available for every
uncertainty analysis. An independent, detailed exposition is not needed every time an analysis is
undertaken, however. If the conditions and assumptions used to estimate an uncertainty are the same in
one case as they were in a past case, then the narrative developed for a past case is applicable to the present

A2LA
Document Revised:
Page 17 of 32
case and need not be duplicated. However, it may be necessary to update the values of specific uncertainty
contributors if new information becomes available.
For illustration, the tensile test example uncertainties calculated thus far are listed in the table below.
Uncertainty Source Standard Type Units Distribution

Uncertainty
Repeatability (Sr) 482.8 Type A PSI Normal
Resolution (Thickness) 289 Type B µin Rectangular
Resolution (Width) 289 Type B µin Rectangular
Reference Standard (Force 0.29 Type B lb Rectangular
Gage) Resolution
Ref. Standard Calibration 4.25 Type B lb Normal
Certificate
Tensile Strength Test Uncertainty Budget
Table 3.
Correlated input quantities
Once the uncertainty contributions from all significant influence quantities have been determined, the next
step in the GUM 8-step process is to identify covariances and correlations, if any, in the input quantities.
Correlation occurs when the values of input quantities are not independent. For example, in the tensile
strength measurement we’ve been examining, the measurements of the thickness and width of the test
specimen would be correlated if both quantities were measured with the same device. Correlated quantities
do not combine through least square summation.
Example: Four 100 pound weights are used together to calibrate a load cell at 400 pounds. These
weights were all calibrated at the same laboratory on the same scale. The uncertainty of each
weight is said to be 100 mg. At least some of this uncertainty will be a bias. For the sake of
example, assume the entire 100 mg is a positive bias from the expected value. The resultant bias
caused by using the four weights together is 400 mg. However, if the method of least square
summation were used the result would only 200 mg27, which is only half the true uncertainty.
Correlated input quantities are common in testing and a simple method for addressing them is to add the
correlated uncertainties together and use that sum in the combined uncertainty calculation. This method
is conservative but may result in a much larger total uncertainty estimate than would be obtained by a more
rigorous approach. In the table below, resolution uncertainties for thickness and width are added together
because they are correlated.
For more information, consult the GUM on the topic.
27
√100 100 100 100 200

A2LA
Document Revised:
Page 18 of 32
Uncertainty Source Standard Type Units Distribution

Uncertainty
Repeatability (Sr) 482.8 Type A lb Normal
Thickness & Width 578 Type B µin Rectangular
Correlated (289+289)
Reference Standard (Force 0.29 Type B lb Rectangular

Gage) Resolution
Ref. Standard Calibration 4.25 Type B lb Normal
Certificate
Table 4.
Calculating the measured value to report
The next step of the GUM 8-step process is to calculate the quantity value to be reported for the
measurement, usually an average measured or calculated value. The ASTM D638 method requires a
calculation of maximum tensile strength which we have already done. Here we will also calculate the
average tensile strength value.
Using the tensile strength formula with average values for force, thickness, and width, we calculate
the average value:
F 852lb
S   13632 lb / in2
TW (0.125in  0.500in)
Sensitivity coefficients
Before the uncertainty contributions from the input quantities can be combined, they must all be in the
same units. (You cannot add apples and oranges, or inches and millimeters.) Returning to our tensile
strength example, notice that the uncertainty for Thickness and Width is in units of “microinches” and
the uncertainty for Force is in “pounds”, but Tensile Strength is given in PSI28. Before the uncertainties
in all the input quantities can be combined, they must be converted into the units associated with the
measured value for the test, tensile strength at break. The applicable units are PSI.
Sensitivity coefficients is the GUM term for conversion factors that convert from input quantity units into
units of the measurand. These conversions may be made at any time in the uncertainty estimation process,
but they must be performed before the uncertainties are combined. The conversion is often quite simple.
For example, multiplying micro-inch values by the sensitivity coefficient 1,000,000 converts them into
inches. Unfortunately, determining sensitivity coefficient values is sometimes a difficult process.
28
lb/in2

A2LA
Document Revised:
Page 19 of 32
When the model function is known, sensitivity coefficients can be readily determined using calculus. (An
approximation method that does not require calculus is given in Appendix C.)
Mathematically, sensitivity coefficients are partial derivatives of the model function f with respect to the
input quantities. In particular, the sensitivity coefficient ci of the input quantity xi is given by
f 20)
ci 
xi
Example: sensitivity coefficients using calculus

The model function we’re using for the tensile strength determination is
F
S
TW
where S is the tensile strength, F is the load needed to break a test specimen, and T and W are the thickness
and width respectively of the test specimen. We obtain the sensitivity coefficients as follows:
S 1 S S F S S F S
cF    ; cT   2  ; cW   2
 .
F TW F T T W T W TW W
The average thickness of the test specimen is 0.125 in, the average width is 0.500 in, the average force
needed to break the specimens is 852.0 lb and the average tensile strength is 13632 lb/in2. With these
values we can determine the values of each of the sensitivity coefficients:
lb
S 13632 in 2
cF    16.0in  2 ;
F 852lb
 S  13632 in 2
lb
lb
cT    109056 3 ;
T 0.125in in
 S  13632 in 2
lb
lb
cW    27264 3 .
W 0.500in in
Determinations of sensitivity coefficients must take place at or very close to the input quantity values
actually measured in a test.
In the table below, the uncertainty contributors are listed in base units, Type A or Type B is indicated,
along with the distribution assumed (N for normal, R for rectangular). Multiplication by the Sensitivity

A2LA
Document Revised:
Page 20 of 32
Coefficient gives the uncertainty in measurement units of PSI. The table also lists degrees of freedom for
each standard uncertainty.
Uncertainty Source Standard Units Type, Sensitivity Standard DOF

Uncertainty Dist Coeff Uncertainty
(PSI)
Repeatability (Sr) 482.8 PSI A, N 1 483 4
Micrometer Resolution 0.0000289 Inch B, R 109056 ∞
31.5
(Thickness)
Micrometer Resolution 0.0000289 Inch B, R 27264 ∞
7.88
(Width)
Reference Standard (Force 0.289 Lb B, R 16 ∞
4.62
Gage) Resolution force
Ref. Standard Calibration 4.25 Lb B, N 16 ∞
68.0
Certificate force
Table 5.
Combining the contributors

Once all of the values of the standard uncertainty contributors ui have been estimated and the sensitivity
coefficients ci have been determined and applied, they are combined by “root-sum-square”, i.e., taking
the square root of the sum of the squares of the uncertainty estimates in order to determine the combined
standard uncertainty.
uc  c u
i
2 2
i i .
Because the uncertainties are combined by root sum square, it is common practice to list them in absolute
value rather than show negative signs in the budget table.

A2LA
Document Revised:
Page 21 of 32
Index Uncertainty Source Standard Units Type Sensitivit Standard DOF

Uncertainty , Dist y Coeff Uncertainty
(PSI)
1 Repeatability (Sr) 482.8 PSI A, N 1 483 4
Resolution (Thickness) 0.0000289 Inch B, R 109056 ∞
2 31.5+7.88=
Resolution (Width) 0.0000289 Inch B, R 27264 39.4 ∞
Reference Standard 0.289 Lb B, R 16 ∞
3 4.62
(Force Gage) Resolution force
4 Ref. Standard Calibration 4.25 Lb B, N 16 ∞
68.0
Certificate force
uc, Combined Uncertainty (RSS): 489 PSI

Table 6.
It is not necessary to use an uncertainty budget for these calculations and there is no mandatory format
for an uncertainty budget, if one is used. Most people find the uncertainty budget an efficient and
practical way to manage the calculations.
Notice the term near the middle where 31.5 and 7.88 PSI terms are combined prior to squaring. This is a
conservative method to account for correlation. Correlation effects can be quite large in some situations.
In this example, the effect is really negligible and only included for educational purposes. The reader is
referred to the GUM where this matter is addressed in depth.
Calculating the expanded uncertainty
The expanded uncertainty U is obtained by multiplying the combined standard uncertainty by coverage
factor k:
U  ku c .
The procedure for determining the coverage factor is presented below. The reader is urged to consult the
GUM for more information and the rationale behind the procedure.
Estimating the coverage factor
Multiples of the standard deviation of a population characterized by a normal probability distribution

provide the probabilities that a value lies within a specified range. In the same way, the coverage factor
provides the multiplier to be applied to the combined standard uncertainty to ensure that the measured
value lies within the uncertainty range to some specified confidence level. K=1 provides 68% confidence,
K=2 provides 95% confidence, K=3 provides 99% confidence, and so on. The coverage factor is a function
of the effective degrees of freedom for the combined uncertainty.

A2LA
Document Revised:
Page 22 of 32
K=2 is commonly used for calibration and test reports. This value is appropriate when Type B uncertainty
components dominate the uncertainty budget or when Type A components have been established with 30
repeated measurements or more.
For other situations, such as the tensile strength test example, where the Type A uncertainties dominate
and where there are fewer than 30 degrees of freedom, a larger value for k is required29. The appropriate
k value is based on the Student’s T distribution (Appendix 2) and the effective degrees of freedom for the
measurement. The reader is encouraged to refer to the GUM sections dealing with coverage factor and
degrees of freedom.
Degrees of freedom can be conservatively estimated by assuming infinite degrees of freedom whenever
Type B components comprise more than half the combined uncertainty and using the actual degrees of
freedom from the Type A portion of the budget otherwise. In the case of tensile testing, degrees of
freedom=4.
A rigorous estimate can be made by following the procedure below.
1) Obtain the estimate of the measurand y and the estimate of the combined standard uncertainty uc(y).
2) Estimate the effective degrees of freedom eff using the Welch-Satterthwaite formula:
uc4 ( y )
 eff 
ci4u 4 ( xi )
n

i 1
i
where i is the degrees of freedom of the estimate30 of the magnitude of the uncertainty contributor xi.
This equation is complicated, but can be readily performed with a spreadsheet.
We can calculate the effective degrees of freedom eff for the tensile test example as follows:
Because we have already applied the sensitivity coefficients, the terms ciu(xi) have already been calculated
in PSI and are u1=483 with 1=4, u2=39.3 with 2=∞, u3=4.62 with 3=∞, and u4=68.0 with 4=∞. uc= 489
PSI.
29
It should be noted that the ASTM standard provided the appropriate K factor for the uncertainty due to repeatability:
2.83.
30
The degrees of freedom of a Type A evaluation based on n repeated measurements is simply  = n – 1. If n
independent observations are used to determine both the slope and intercept of a straight line by the least squares
method, then the degrees of freedom of their respective standard uncertainties is  = n – 2. In general, for a least-
squares fit of m parameters to n data points the degrees of freedom of the standard uncertainty of each parameter is
 = n – m.

A2LA
Document Revised:
Page 23 of 32
 eff 
4894  4.
4834  39.34  4.624  68.04
4   
Consulting the Student’s t-table, we find that the value of t corresponding to 4 degrees of freedom at the
95% level of confidence is t = 2.8. So the expanded uncertainty of the tensile strength result is
lb
U  kuc  ( 2.8)(489 )  1,369PSI .
in 2
This is about 10% of our test result of 13600 psi. We round the final uncertainty estimate to no more than
two significant figures (U=1,400 PSI) so as not to convey the impression of greater accuracy than is
warranted. Standard rounding practice such as the one found in section 6.4 of ASTM E29 should be
followed, although it is common practice always to round uncertainty estimates to one or two significant
figures. It is best to do these uncertainty calculations with a spreadsheet so that intermediate-rounding
errors can be avoided. In this case, rounding errors were negligible.
If we had decided, because Type A dominates the uncertainty budget, to just use the k value associated
with the Type A degrees of freedom for the expanded uncertainty, we would have gotten the same result
without making complicated calculations.
In some tests there may be multiple, significant Type A uncertainties with differing degrees of freedom.
For such a case, the Welch-Satterthwaite is useful.
Reasonability
In the end, every uncertainty estimate should be subjected to a reasonability check. The analyst should
ask questions such as “Is this estimate in line with what I know about the nature of the measurement and
of the material?” “Can this estimate be supported with proficiency testing data, or data accumulated as
part of a measurement assurance program?” Uncertainty estimates that look strange -- either too big or
too small -- should be re-evaluated, looking first for mathematical blunders, second for uncertainty
contributors whose magnitudes may have been poorly estimated or neglected. Finally, it may be necessary
to revise the mathematical model.
Human judgment based on sound technical experience and professional integrity is of paramount
importance in evaluating uncertainty.

A2LA
Document Revised:
Page 24 of 32
h. Reporting uncertainty
When reporting the result of a measurement, at a minimum one should provide the following:
1) A full description of how the measurand Y is defined;
2) The result of the measurement as Y = y  U and give the units of y and U;
3) The value of the coverage factor k used to obtain U;
4) The approximate level of confidence associated with the interval y  U and state how it was
determined.
The numerical values of the estimates of the measurand and expanded uncertainty should not be given
with an excessive number of significant digits. Uncertainty estimates should be quoted to no more than
two significant figures.
When stating measurement results and uncertainty estimates it is always advisable to err on the side of
providing too much information rather than too little and this information must be stated as clearly as
possible.
The statement of our tensile strength result might take the following form:
“The maximum tensile strength at break is determined to be 14,400 PSI based on tests of five specimens
conducted in accordance with ASTM D638.
“For these test specimens, the average tensile strength at break was also determined. Average tensile
strength at break = 13,636 PSI  1,400 PSI. The uncertainty listed is the expanded uncertainty based on
a coverage factor of 2.8 (95% confidence) calculated in accordance with the GUM.”
It should be noted that we did not improve the ASTM D638 test by performing uncertainty calculations.
We also did not eliminate the unknown bias mentioned earlier that is associated with this measurement.
We did provide additional information that might be of value and, in the process, demonstrated the GUM
method with an actual test example.
END

A2LA
Document Revised:
Page 25 of 32
APPENDIX A
Control Charts
A control chart31 (also called “Shewhart Chart”) is a plot of some characteristic statistic for test data with
a center line that is the mean value, and upper and lower lines at established control limits. There may
also be upper and lower warning limits shown on the chart.
In the figure below, the characteristic statistic is a measured value.
103
102.5
102 Upper Control Limit

Upper Warning Limit
Measured Value
101.5
Mean Value
101
Lower Warning Limit
100.5 Lower Control Limit
100
99.5
99
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Sample
Measured Value Sample Mean UCL UWL LCL LWL
A Control Chart
It is common practice to set upper and lower control limits on a control chart at three times the standard
deviation (“3 ”) to ensure that approximately 99.7% of measurements are within the limits for a process
in statistical control.
It is sometimes useful to plot 2  upper and lower warning limits which contain 95.0% of the measurements
for a measurement process in statistical control.
Control charts can be used to estimate measurement uncertainty if the following conditions are met:
1) The control test sample has a certified or otherwise known or accepted value. Then, bias in the
measurement process may be identified and corrected in the calculation of measurement results. There
will be some uncertainty associated with bias corrections, so it may be necessary to identify and
quantify this uncertainty and root-sum-square it with the standard deviation associated with the control
limits.
31
See also ASTM E2554: Standard Practice for Estimating and Monitoring the Uncertainty of Test Results of a
Test Method in a Single Laboratory Using a Control Sample Program and ISO 21748: Guidance for the use of
repeatability, reproducibility and trueness estimates in measurement uncertain estimation.

A2LA
Document Revised:
Page 26 of 32
2) The value of the measurand represented by the control sample should be close to the value of the
measurand actually obtained during routine testing since, in general, the uncertainty of a measurement
will be some function of the “level of the test”, or value of the measurand. Consequently, it may be
necessary to track several control samples at different measurement levels to properly assess the
measurement uncertainty across the range of the measurand encountered in the testing laboratory.
3) The measurement process for control samples should be the same as for routine samples, including
subsampling and sample preparation. If it is not, then additional uncertainty components may have to
be considered.
4) The measurement process must be in statistical control as demonstrated by the control chart. This
means that a sufficient number of data points must be collected to establish that the process in in
control and to ensure that the estimate of the population standard deviation is reasonably accurate.
There are no universally applicable rules, but 20 – 25 subgroups of 4 or 5 are generally considered
adequate for providing preliminary estimates. Measurement processes that are not in statistical control
must be brought into control before the control chart can be properly constructed.
Control charting is often a reliable, simple tool for estimating measurement uncertainty for testing.
However, control charts are not practical in all testing situations. Tests that are conducted infrequently, for
which multiple repeated measurements are not practical, or for which a reference material is not available,
do not lend themselves to control charting.

A2LA
Document Revised:
Page 27 of 32
APPENDIX B
Student’s t-Table
Degrees Degrees
of One Two Three of One Two Three
Freedo Sigma, Sigma, Sigma, Freedo Sigma, Sigma, Sigma,
m 68% 95% 99.7% m 68% 95% 99.7%
1.81899 1.01384 2.05552 3.27361
1 3 12.7062 212.205 26 3 9 1
1.31157 4.30265 18.2163 1.01311 2.05183 3.26129
2 8 3 1 27 2 1 4
1.18892 3.18244 8.89145 1.01243 2.04840 3.24992
3 9 6 6 28 4 7 9
1.13439 2.77644 6.43484 1.01180
4 7 5 8 29 4 2.04523 3.23941
1.10366 2.57058 5.37602 1.01121 2.04227 3.22964
5 8 2 5 30 6 2 6
1.08397 2.44691 4.80024 1.01066 2.03951 3.22055
6 6 2 3 31 7 3 9
1.07028 2.36462 4.44212 1.01015 2.03693
7 7 4 5 32 2 3 3.21208
1.06022 2.30600 4.19914 1.00966 2.03451 3.20415
8 4 4 9 33 9 5 1
1.05251 2.26215 4.02398 1.00921 2.03224
9 5 7 7 34 5 5 3.19672
1.04642 2.22813 3.89195 1.00878 2.03010 3.18974
10 3 9 5 35 8 8 1
1.04148 2.20098 3.78898 1.00838 2.02809 3.18317
11 6 5 2 36 4 4 5
1.03740 2.17881 3.70648 1.00800 2.02619 3.17698
12 5 3 7 37 3 2 6
1.03397 2.16036 3.63894 1.00764 2.02439 3.17114
13 6 9 7 38 2 4 2
1.03105 2.14478 3.58265 1.00729 2.02269 3.16561
14 3 7 3 39 9 1 5
1.02853 3.53502 1.00697 2.02107 3.16038
15 3 2.13145 5 40 4 5 1
1.02633 2.11990 3.49421 1.00444 2.00855 3.12007
16 7 5 2 50 6 9 6
1.02440 2.10981 3.45885 1.00276 2.00029 3.09371
17 7 6 4 60 8 8 3
1.02269 2.10092 1.00157 1.99443 3.07512
18 8 2 3.42793 70 2 7 7
1.02117 2.09302 3.40065 1.00067 1.99006 3.06131
19 2 4 8 80 7 3 9

A2LA
Document Revised:
Page 28 of 32
1.01980 2.08596 3.37642 0.99998 1.98667 3.05065

20 4 3 8 90 3 5 7
1.01856 2.07961 3.35475 0.99942 1.98397 3.04217
21 8 4 9 100 7 2 5
1.01744 2.07387 3.33526 0.99693 1.97189 3.00453
22 7 3 7 200 6 6 5
1.01642 2.06865 3.31763 0.99610 1.96790 2.99217
23 6 8 9 300 9 3 7
1.01549 2.06389 3.30162 0.99569 1.96591 2.98603
24 2 9 2 400 6 2 2
1.01463 2.05953 3.28700 0.99544 2.98235
25 4 9 5 500 8 1.96472 7
0.99445 1.95996 2.96773
∞ 8 4 9

A2LA
Document Revised:
Page 29 of 32
APPENDIX C
Case 1: estimating sensitivity coefficients mathematically
The model function we’re using for the tensile strength determination is
F
S
TW
The average thickness of the test specimen is 0.125 in, the average width is 0.500 in, the average force
needed to break the specimens is 852.0 lb and the average tensile strength is 13632 lb/in2. With these
values we can estimate the values of each of the sensitivity coefficients.
The method is a mathematical approximation based on the model function. It consists of making a
calculation using actual data, changing one of the quantities of interest, Force, for example, by a small and
arbitrary amount and then recalculating the output quantity. From the change in output quantity value
caused by a small change in one input quantity, the sensitivity coefficient can be directly calculated.
Below, this method will be used to estimate each sensitivity coefficient for our tensile strength example.
One parameter will be changed for each calculation while the others will be held constant.
Using the measured load of 852 lbs, thickness of 0.125 in, and width of 0.500 in in the formula above
gives a value of S = 13632 PSI.
If we hold everything else constant and increase F by about 1% to 860 lbs, which we call F’, we can
calculate what would be the changed tensile strength at break, S’=860/(0.125×0.500) = 13,760 PSI.
The difference in tensile strength is S’-S= S = 13760-13632 = 128 PSI
The difference in force is F’-F= F= 8 lbs
With this information, we calculate the sensitivity coefficient for force, cF S/F = 128/8 = 16 PSI/lb
The amount of the assumed change is not important, but it must be small.
Similarly, if we make an imaginary increase in T of about 1% to 0.130 in., the tensile strength at break
would be S’=852/(0.130×0.500) = 13,108 PSI
The difference S = 13108-13632 = 524 PSI

The difference T= 0.005 in. cT c
cT S/T = 524/0.005 = 104,800 PSI/in
If we make an imaginary increase in W of about 1% to 0.505 in., the tensile strength at break would be
S’=852/(0.125×0.505) = 13,497 PSI
The difference S = 13497-13632 = 135 PSI

A2LA
Document Revised:
Page 30 of 32
The difference W= 0.005 in.

CW S/W = 524/0.005 = 27,000 PSI/in
Comparison to the sensitivity coefficients calculated using calculus will show close but not perfect
agreement. These estimates are close enough, however, to be used.
Case 2: estimating sensitivity coefficients experimentally

Experimental determination of sensitivity coefficients is preferable to approximation, if practical32.
Sensitivity coefficients may be determined experimentally through a process very similar to that described
above for mathematical estimation. The principal difference is that, in this case, the small changes in input
quantities used to calculate sensitivity are real, based on actual data from tests.
For this case, we will use the same tensile strength example and model function:
F
S
TW
Let us identify the set of test specimens used in our examples up to now as Set 1. We select another set of
5 test specimens, Set 2, that are slightly thicker33 and repeat the test. The results for the Set 1 and Set 2
tests are shown in the table below.
Set Average Average Average Average Measured

Measured Measured Area, Load, F Average
Thickness, Width, W TW Calculated S
T
1 0.125 in 0.500 in 0.0625 852.0 lb 13632 PSI
2 0.128 in 0.500 in 0.0640 890.6 lb 13916 PSI
The difference in S between the two sets is S = 13916-13632 = 284 PSI.

The difference in thickness between the two sets is T= 0.128 – 0.125 = 0.003 in.
The sensitivity coefficient for thickness, cT S/T = 284/0.003 = 94,667 PSI/in.
The same process could be used again to estimate the coefficient for width, if an appropriate set of test
specimens could be located. This set of specimens would have the same average thickness but different
average width than the specimens in the previous set.
Unfortunately, experimental determination of sensitivity coefficients is not always practical and

sometimes not possible. Multiple sets of test specimens with the necessary dimensions might not be
32
The model function may not include all influence quantities; it may be an approximation of some more complex
function, it may suffer from other limitations. The response of the model function is necessarily hypothetical.
Experimental data, on the contrary, reveals the actual performance of the test system when parameters are varied.
33
The new specimens must remain within the ASTM D638 tolerances.

A2LA
Document Revised:
Page 31 of 32
available. The time and expense to perform multiple tests for the purpose of estimating sensitivity
coefficients might be unacceptable expenditures. The precision of the indicator might not be adequate to
reveal the effect of small changes in input quantities. There might not be enough tolerance allowable in
the input quantity to perform experimental determinations for sensitivity.
Sometimes, a sensitivity coefficient can be determined by an experiment as simple as changing room

temperature by one degree and observing the corresponding change in some quantity.

A2LA
Document Revised:
Page 32 of 32
References
“Introducing the Concept of Uncertainty of Measurement in Testing in Association with the Application
of the Standard ISO/IEC 17025”, ILAC-G17
“Standard Test Method for Tensile Properties of Plastics”, ASTM 638

BIPM JCGM 100:2008, Evaluation of measurement data – Guide to the expression of uncertainty in
measurement (GUM 1995 with minor corrections).
BIPM JCGM 200:2012, International vocabulary of metrology - Basic and general concepts and
associated (VIM) 3rd edition (2008 version with minor corrections).
“P103A2LA Policy on Measurement Uncertainty for Testing Laboratories”, American Association for
Laboratory Accreditation, 2013
Document Revision History
Date Description
July 2002  Initial publication
December 4, 2014  Complete re-write of the document

A2la G104 - 2014 PDF

Uploaded by

Copyright:

Available Formats

A2la G104 - 2014 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A2la G104 - 2014 PDF

Uploaded by

Copyright:

Available Formats

A2LA

G104 - Guide for

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Measurement uncertainty4: a non-negative parameter characterizing the dispersion of the quantity

1) incomplete definition of the quantity being measured;

2) imperfect realization of the definition of the quantity being measured;

4) inadequate knowledge of the effects of environmental conditions on the measurement or imperfect

5) personal bias in reading analog instruments, including the effects of parallax;

6) finite resolution or discrimination threshold;

7) inexact values of measurement standards and reference materials;

9) approximations and assumptions incorporated in the measurement method and procedure;

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Standard deviation Repeated

Precision: is closeness of agreement between indications or measured quantity values obtained by

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Bias8: the estimate of a systematic measurement error.

2) Interlaboratory comparison: Laboratories participating in proficiency testing schemes will have

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Measurement method10: a generic description of a logical organization of operations used in a

Measurement procedure11: a detailed description of a measurement according to one or more

Systematic errors12: a component of measurement error that in replicate measurements remains

Random errors13: a component of measurement error that in replicate measurements varies in an

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Intermediate precision condition of measurement14: a condition of measurement, out of a set of

III. Tensile strength example

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Test Measured Measured Area,

1 0.124 in 0.499 in 0.0619 830 lb 13409 PSI

2 0.126 in 0.501 in 0.0631 900 lb (Maximum) 14263 PSI

3 0.125 in 0.500 in 0.0625 810 lb 12960 PSI

ASTM D638 Tensile Strength = 900/0.0625 = 14,400 PSI

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

IV. Introduction to the GUM method

The GUM method is an eight step process:

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

b. GUM measurement description

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

c. Type A evaluation of standard uncertainty

d. Type B evaluation of standard uncertainty

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

For a standard uncertainty or standard deviation value of ±a reported at a coverage or multiplication

values, among many other parameters. It is also the

Xi‐a xi Xi+a The equation for the standard uncertainty of a

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Uncertainty Source Standard Type Units Distribution

Correlated input quantities

For more information, consult the GUM on the topic.

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Uncertainty Source Standard Type Units Distribution

Reference Standard (Force 0.29 Type B lb Rectangular

Calculating the measured value to report

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing

Example: sensitivity coefficients using calculus

L:\Guidance\G104 – Guide for Estimation of Uncertainty in Testing