Unc Sidekick User Manual

User Manual
Uncertainty Sidekick
Integrated Sciences Group

Copyrights and Trademarks
Uncertainty Sidekick 1.0
© 2005-2007, Integrated Sciences Group
14608 Casitas Canyon Road
Bakersfield, CA 93306
All rights reserved.
Other product names mentioned in this user manual may be trademarks or registered trademarks of their
respective companies and are hereby acknowledged.
Proprietary Notice
Neither this manual, nor Uncertainty Sidekick software, or portions thereof may be resold.
Technical Support
Inquiries concerning Uncertainty Sidekick software and associated documentation should be directed to
Integrated Sciences Group, 14608 Casitas Canyon Road, Bakersfield, CA, 93306, USA.
Tech Support: 1-661-872-1683
Sales: 1-800-400-7866
Fax: 1-661-872-3669
Email: tech@isgmax.com
Service Updates
Minor updates and corrections will be issued from time to time in the form of Service Updates.
Check our website at http://www.isgmax.com for free downloads.
ii
Table of Contents
Foreword v
Chapter 1 Installation and Startup 1

1.1 Installation Procedure 1
1.2 Video and Color Recommendations 2
1.3 System Requirements 2
1.4 Release Notes 3
1.5 Uncertainty Sidekick Files 3
1.6 Program Startup 3
Chapter 2 Overview of Features 4

2.1 Step-by-step Analysis Procedures 4
2.2 Analysis Screens 4
2.3 Automatic Uncertainty Combination 4
2.4 Bayesian Analysis (SMPC) 4
2.5 Charts and Plots 4
2.6 Analysis Notes 5
2.7 Saving and Opening Files 5
2.8 Reporting Analysis Results 5
2.9 Measurement Units Database 5
2.10 Type B Degrees of Freedom Calculator 6
2.11 Access to External Applications 6
2.12 On-Screen Help 6
Chapter 3 The Main Screen 7

3.1 Main Screen Menu 7
3.2 The Analysis Path 8
3.3 Analysis Details 9
3.4 Error Component Table 9
3.5 Include in Analysis 10
3.6 Analysis Results 10
3.7 Combined Error Distribution 10
3.8 Parameter Bias Estimates 10
Chapter 4 Example 1 – DC Voltmeter Calibration 11

4.1 Measurement Process Overview 11
4.2 Uncertainty Analysis Procedure 12
Chapter 5 Example 2 – Micrometer Calibration 24

Chapter 6 Example 3 – Thermometer Calibration 39

iii
Chapter 7 Printing Analysis Results 52
7.1 Report Options Screen 52
7.2 Selecting Report Fonts 52
7.3 Setting Reports Nomenclature 52
7.4 Previewing and Printing Reports 53
Chapter 8 Uncertainty Analysis Basics 54

8.1 Define the Measurement Process 54
8.2 Identify Error Sources and Distributions 54
8.3 Estimate Uncertainties 56
8.4 Combine Uncertainties 60
8.5 Report the Analysis Results 63
Chapter 9 SMPC Methodology 64

9.1 Basic Principles 64
9.2 Equal Accuracies 64
9.3 Unequal Accuracies 64
9.4 Single Sample Case (Example 1 – DC Voltmeter Calibration) 65
9.5 Two Sample Case (Example 3 – Thermometer Calibration) 65
9.6 Cautions and Conditions 66
Nomenclature and Terminology 67
Suggested Reading 75
Subject Index 76
iv
Foreword
Welcome to Uncertainty Sidekick, a free software tool that contains state-of-the-art analytical methods and techniques
designed to guide you through the analysis process. Uncertainty Sidekick incorporates specially designed screens and
worksheets that help you identify sources of error common to most direct measurement processes and to estimate the
uncertainties in these errors.
Things to Keep in Mind

This freeware program can be used to develop uncertainty estimates for direct measurement scenarios. Direct
measurements basically consist of a measuring parameter "X" that directly measures the value of subject parameter "Y"
(i.e., X measures Y). In other words, the value of the subject parameter is obtained directly by measurement and is not
determined indirectly by computing its value from the measurement of other variables or quantities.
Examples of direct measurements include, but are not limited to the following:
• Measuring the length of an object with a ruler, micrometer, or other such device.
• Measuring a 1 gram mass with a scale, precision balance, or other device.
• Measuring the temperature of a substance using a thermometer, thermocouple or other temperature
sensing device.
It is important to note that direct measurements can consist of directly measuring two quantities of interest. This occurs
when one is interested in ensuring that parts fit together properly. For example, direct measurements of a pin length or
diameter may be made, as well as the length or diameter of the pin slot or receptacle to verify that an adequate fit will be
achieved. In this case, we are interested in the uncertainties in the measurements of the pin and receiving slot
dimensions and, ultimately, would like to ensure that the specified slot dimension exceeds the associated pin dimension
by some specified confidence level (e.g., 99%).
Although Uncertainty Sidekick is limited to the analysis of direct measurements, the statistical methods and techniques
incorporated in this program are rigorous. We have not attempted to “dumb down” the analysis process to accommodate
those in the measurement science community that espouse simplistic uncertainty estimates over realistic estimates.
What You Need to Do

To familiarize yourself with the principles of uncertainty analysis, we recommend that you read Chapter 8 Uncertainty
Analysis Basics. You should also review the file entitled Intro to Uncertainty Analysis.pdf. This file contains a
presentation of the basic statistical methods and concepts that are key to understanding uncertainty analysis.
We also recommend that you work through the analysis examples included with this program. In working through these
examples, you will discover that there are a few simple steps that are common to most analyses. These steps include
1. Define the Measurement Process

2. Identify the Error Sources
3. Estimate and Combine the Uncertainties
4. Report the Analysis Results
The key to developing realistic uncertainty estimates is to use your own technical knowledge about the measurement
process coupled with your common sense.
How to Use this Manual

The best way to use this manual is review the program features presented in Chapter 2; and the layout and functions of
the Main screen presented in Chapter 3. Then, you should work through the analysis examples presented in Chapters 4
through 6. Each example is designed to uncover different aspects or features of Uncertainty Sidekick and to reinforce
the basics steps in conducting an uncertainty analysis.
Note on Reporting Results

The term “significant digits” will appear from time to time in discussions on uncertainty analysis. In Uncertainty
Sidekick, it does not necessarily mean the number of digits past the decimal that you are justified in using to express an
v
uncertainty estimate. More often than not, it refers to how many places past the decimal you want to use to view a
number when comparing the effect of different factors on an uncertainty estimate.
The task of determining the number of significant digits for reporting your uncertainty estimates is up to you.
Uncertainty analysis results should be presented in such a way that they convey information without being misleading.
If an estimate is presented to a level of precision that is not justified on the grounds of the precision of the information
that was used to obtain it, a note should be included advising that is so, giving the reason for the precision in the reported
value.
The terminology used in Uncertainty Sidekick has been developed to avoid specialized language that may be confusing
to the general technical community. However, if you wish to report analysis results using terms that are unique to your
technical specialty or organization, we have included a feature that allows you to enter preferred nomenclature for use in
the analysis reports.
Note to Statisticians
Uncertainty Sidekick, like our other uncertainty analysis programs, is built on a methodological foundation that
incorporates the concepts, methods and techniques of the ISO Guide to the Expression of Uncertainty in
Measurement (the “GUM”). We have also developed and incorporated additional techniques and tools to handle
important uncertainty analysis concepts that are not currently included in the GUM.
vi
Chapter 1 Installation and Startup
1.1 Installation Procedure
To install Uncertainty Sidekick, double-click the setup.exe file.
This will display a message saying the Wise Installation System is
being activated. Once this completes, you will see the Welcome
screen.
To continue the installation, click the Next button. To exit the

installation, click the Cancel button. Clicking the Next button
closes the Welcome screen and displays the User Identification
Screen. Enter your Name and Organization and then click the
Next button to continue the installation.
The User Confirmation screen will be displayed next. If the

information is not correct, click the Back button to return to the
User Identification screen. Click Next to continue the installation
or click Cancel to exit the installation.
Clicking the Next button closes the User Confirmation screen and
displays the Destination Location screen which allows you to
select the directory folder where you will install Uncertainty
Sidekick.
The default
location is an Uncertainty Sidekick folder created on your
computer’s C:/ drive. You can select another directory location for
the Uncertainty Sidekick files by clicking the Browse button. Once
you’ve selected the Destination Directory, click the Next button to
continue the installation or click the Cancel button to exit the
installation.
Clicking the Next button, closes the Destination Location screen

and displays the Backup Replaced Files screen.
If you want to backup files that the installation may replace on

your computer, select the Yes option. If not, select the No option.
If you select the Yes option, the backed up files will be saved in
the Backup File Destination Directory. The default location is a
Backup folder that is created within the Uncertainty Sidekick
destination folder. You can select another directory location for
the Backup files by clicking the Browse button. Click the Next
button to continue the installation or click the Cancel button to exit
the installation.
1
Clicking the Next button closes the Backup Replaced Files screen
and displays the Select Program Manager Group screen. Select
one of the displayed groups listed or enter a group name of your
choice. This is the Uncertainty Sidekick Program Group that will
be displayed on your Start menu Programs list. Click the Next
button to continue the installation or click the Cancel button to
exit the installation.
Clicking the Next button closes the Select Program Manager

Group screen and displays the Start Installation screen. Click the
Next button to continue the installation or click the Cancel button
to exit the installation.
Clicking the Next button closes the Start Installation screen and
displays the Setup Progress screen. This screen displays the files
that are being installed on your computer, along with descriptions
of related setup activity. The completion percentage for the
installation is also shown.
Some of the files that have been installed will need to be registered
in your Windows Registry. This activity commences when the
Setup Progress screen closes and a screen appears informing you
that your computer’s configuration is being updated.
Once the configuration update is completed, a screen will appear

informing you that that Uncertainty Sidekick has been successfully
installed. Clicking the Finish button closes the installation.
On occasion, the setup program detects whether you need to restart

your computer to complete the installation process. If so, a
message will be displayed offering you the choice of restarting the
computer or leaving the installation without restarting.
1.2 Video and Color Recommendations

For optimum display of the Uncertainty Sidekick program screens
and worksheets, we recommend the following settings:
Video Resolution 1024 x 768 or better with Large Fonts

Color 16 bit (65,536) or better color
For Windows XP, you may also have to change your DPI setting to 120 (Large Size). Uncertainty Sidekick supports 800
x 600 resolution, but not all screen features and functions can be displayed properly at this or lower resolutions.
1.3 System Requirements

It is recommended that you run Uncertainty Sidekick on a Pentium II equivalent computer or better. Uncertainty
Sidekick will run on Windows 95, NT 4, 2000, ME, or XP operating systems. Hard disk space of 10 Mb is required to
install the Uncertainty Sidekick files. Uncertainty Sidekick will run with 32 megabytes of RAM, but higher RAM is
preferable. The minimum required RAM depends on your operating system. If you are running Windows 95 or NT 4,
then 64 Mb should be adequate. If you are running Windows 2000 or XP you may find that you get better performance
with 128 Mb of RAM.
As you use Uncertainty Sidekick, you will save your work in a native file format with an .sdk extension. Each analysis
file requires 10 to 20 Kb of disk space. These files can be opened within Uncertainty Sidekick for review, modification,
or report generation.
2
1.4 Release Notes
You are encouraged to read the file Release Notes.pdf installed in the Readme folder of your Uncertainty Sidekick
application directory. This file, if installed, includes information about fixes, features and functions that were
implemented since the initial release of the software and User Manual. The Release Notes.pdf file will not be installed
as part of the setup program for the initial software release. All subsequent software updates will be accompanied by a
release notes file.
1.5 Uncertainty Sidekick Files

The following files will be installed in the program location directory specified during the program setup:
Uncertainty Sidekick.exe - The program executable file.

Measurement Units.mdb - The measurement units database.
Unwise.exe - An executable for uninstalling the program.
Install.log - A file that contains the history of files that have been installed and actions that
have been taken during the installation process.
Additional files will be installed in subfolders within the program location directory. These include the following folders
and files:
Examples Folder
Three analysis files are included in this folder to illustrate program features and functions and provide basic guidance in
conducting uncertainty analyses for direct measurements.
Example 1 - DC Voltmeter Calibration.sdk

Example 2 - Micrometer Calibration.sdk
Example 3 - Thermometer Calibration.sdk
Help Folder
Two files that provide on-screen Uncertainty Sidekick Help are installed in this folder.
Sidekick.hlp - The Uncertainty Sidekick Help file.

Sidekick.cnt - The Table of Contents for the Uncertainty Sidekick Help.
Uncertainty Analysis Overview Folder

This folder contains a file that covers the basic statistical methods and concepts that are key tounderstanding uncertainty
analysis.
User Manual Folder

This folder contains the Uncertainty Sidekick User Manual file in Adobe Acrobat format.
1.6 Program Startup

Uncertainty Sidekick can be activated by double-clicking the Uncertainty Sidekick.exe file in the program directory,
selecting it from the Start menu Programs list, or double-clicking the program icon on the computer desktop.
3
Chapter 2 Overview of Features
Uncertainty Sidekick is a Windows application that provides a comprehensive approach to estimating uncertainties for
direct measurements. Step-by-step procedures and user-interactive screens assist you in identifying the technical
information needed to estimate uncertainties common to most direct measurement processes. To use Uncertainty
Sidekick, you only need to know information that falls within your technical specialty. All the necessary statistical
methods and techniques for computing and combining uncertainties are built into Uncertainty Sidekick.
2.1 Step-by-step Analysis Procedures

Uncertainty Sidekick provides an interactive step-by-step analysis pathway that helps ensure that all aspects of the
process are included in your analysis. Just select the desired step from the Analysis Path on the left side of the Main
Screen and the appropriate worksheet or screen will pop-up prompting you to select an option or input the required
information. Enter the selection(s) and/or input information, press OK and return to the Main Screen. The
corresponding analysis procedure step has now been changed to signify its completion.
The Subject Parameter and the Measuring Parameter Uncertainty worksheets also contain an interactive, step-by-step
Definition Path to assist you in identifying and entering all pertinent information for estimating parameter bias and other
measurement process uncertainties.
2.2 Analysis Screens

Uncertainty Sidekick taps into your expertise through specially designed worksheets and screens that serve as a user
interface with a three-fold purpose. First, they help you identify the technical information needed for uncertainty
analysis. Second, they capture this information and pass it on the "internal expert" for analysis. Third, you can select
options for displaying and reporting analysis results.
These user-interactive, drill-down worksheets and screens provide a structured way of estimating uncertainties for
various error sources such as measuring parameter bias, random error, resolution error, operator bias and environmental
factors error. For each worksheet or screen, the user supplies only technical information about the measurement. You
are not asked any statistical questions beyond estimating a few simple probabilities. First time users will find that
analyses take minutes, not days or weeks to perform.
2.3 Automatic Uncertainty Combination

Uncertainty Sidekick combines component uncertainties and measurement process uncertainties to provide an estimate
of total uncertainty in accordance with ISO and NIST guidelines. Uncertainty combination is automatic, but the user has
the option of selecting which error sources to include or exclude from the final total uncertainty estimate via the Main
Screen and other worksheets or screens. The Welch-Satterthwaite method is used to compute degrees of freedom for
combined uncertainties, in compliance with ISO and NIST guidelines.
2.4 Bayesian Analysis (SMPC)

SMPC stands for "Statistical Measurement Process Control." SMPC operates from subject parameter and measuring
parameter tolerance information and in-tolerance probabilities, in conjunction with sampled measurement data. It is a
new way of estimating the value of a subject parameter based on measurements made by a measuring parameter. SMPC
also provides an estimate of the value of the measuring parameter, based on the same measurements, and computes the
in-tolerance probabilities of both parameters.
In Uncertainty Sidekick, SMPC works for cases where the measuring parameter measures the subject parameter, where
the subject parameter measures the measuring parameter, or where both the subject parameter and measuring parameter
are used to sample a third artifact used as a reference.
2.5 Charts and Plots

Uncertainty Sidekick allows you to plot a pareto diagram displaying the relative contributions of error sources to
overall uncertainty. A pareto diagram is a bar chart that ranks the error source contributions from the most significant to
least significant as a percent of the total uncertainty. Because some error sources may be correlated to others, the percent
contribution is obtained by extracting the uncertainty for each error source in turn and computing the impact on the total.
Pareto diagrams are also available in the Main Screen, Subject Parameter Uncertainty, Measuring Parameter Uncertainty,
and Measuring Environment Uncertainty worksheets.
4
Uncertainty Sidekick also displays the statistical distributions (i.e., probability density functions) for the subject
parameter bias, measuring parameter bias, and combined measurement process error. The combined measurement
process error distribution is determined from the total standard uncertainty resulting from individual measurement
process errors (e.g., random error, resolution error, environmental error, operator bias, ...).
For process errors or parameter bias with in-tolerance probability less than 100%, Uncertainty Sidekick assumes a
normal distribution. A uniform or rectangular distribution is assumed for parameter bias with 100% in-tolerance
probability.
2.6 Analysis Notes

Uncertainty Sidekick provides notes pages for all
analysis worksheets and screens to store pertinent
information that may be useful if the analysis is
revisited at a later date. In addition to being saved with
each analysis, the content of a Notes page can be saved
to an external file in Rich Text File (RTF) format.
2.7 Saving and Opening Files

Analysis results can be saved to individual files for later
review or modification via the File menu on the Main
Screen Menu Bar. Files are saved in a native format,
with an .sdk extension, that is designed for efficient
data storage and retrieval. In general, it is a good
practice to periodically save your analysis work during
various stages of the development process.
Selecting New Analysis from the File menu causes current entries to be deleted and cleared. If changes have been made,
you will be asked if you want to save them. Selecting Open Analysis opens an Open Analysis File dialog for locating
and opening analysis files. Selecting Save Analysis As opens a Save Analysis As dialog with which you can save an
analysis file with a file name and selected location of your choice. If you are working with an opened file, selecting
Save Analysis updates and saves the open file. Otherwise, the Save Analysis function activates the Save Analysis As
function.
2.8 Reporting Analysis Results

Uncertainty Sidekick provides the capability to preview and print analysis results in various levels of detail. Reports can
be previewed and printed directly from nearly all of the analysis screens or worksheets via the Report Options Screen.
Preferred report fonts and nomenclature can be set by the user.
Reports can printed to any printer that has the appropriate Windows printer driver. If you are using a laptop, the analysis
report print preview option is only active if you have a printer driver installed. The analysis reports can also be copied
and pasted as images into Word, Excel, PowerPoint, or other such applications. Reporting analysis results is discussed in
more detail in Chapter 7.
2.9 Measurement Units Database

Uncertainty Sidekick links to a database
(Measurement Units.mdb) that includes over eighty
measurement areas and several hundred units in
compliance with NIST Special Publication 811, Guide
for the Use of the International System of Units (SI).
You can add new measurement areas and units to the

database via the Update Measurement Units Screen,
which is accessed by selecting Measurement Units
from the Edit menu on Uncertainty Sidekick’s Main
Screen. You can also add new units to an existing
measurement area or edit the measurement areas or
units.
5
2.10 Type B Degrees of Freedom Calculator
Uncertainty Sidekick assists the user in computing the degrees of
freedom for heuristically determined uncertainty estimates via the
Type B Degrees of Freedom Calculator. There are three formats in
which to input containment limits (±L) and containment probabilities
(p). The containment probability is p = C / 100 where C is the
percentage of values of y observed within ±L.
Formats:
1. Containment limits and containment probability.
Approximately C% (±∆c%) of observed values have been
found to lie within the limits ±L (±∆L).
2. Containment limits and history of observations.

Approximately n out of N observed values have been found
to lie within the limits ±L (±∆L).
3. Containment limits, probability and number of cases.

Approximately C% of N observed values have been found
to lie within the limits ±L (±∆L).
2.11 Access to External Applications

During an analysis, it may be useful to gather
information from an external source, perform off-line
calculations or document a procedure or analysis
result. External applications can be run from within
Uncertainty Sidekick using the Run External
Application screen.
External programs can be added to the Applications

List by entering the appropriate Application Data and
clicking Add on the menu. The applications that you
add to the list, and their associated paths, are stored
when you close the screen. This information is
retained for all future analyses, until you update or
delete them the list.
2.12 On-Screen Help

Uncertainty Sidekick offers on-screen help using the
Windows Help engine. The Help file provides a
comprehensive reference of technical topics for quick
guidance during an analysis.
The contents of the Help file can be accessed from all

analysis worksheets and screens.
Clicking Help on a worksheet or screen menu

activates the relevant on-screen Help topic. The Help
file also contains a tree-view table of contents,
indexed list and search capabilities for 200+ topics.
6
Chapter 3 The Main Screen
Every aspect of the uncertainty analysis process can be accessed from Uncertainty Sidekick’s Main Screen. Boxes with
a white background will accept data entered from the keyboard or pasted from the Windows clipboard. In this chapter,
we will discuss the various options and functions that can be accessed from the Main Screen.
3.1 Main Screen Menu

The Main Screen menu provides access to a wide variety of functions. We will describe the highlights of each.
File
The analyses you develop with Uncertainty Sidekick can be saved in native file format, which have a
.sdk extension. From the File menu, you can open an analysis, save an analysis or initiate a new one.
The File menu also provides options for previewing and printing analysis reports.
Edit
The Edit menu allows you to cut, copy, paste or delete entries in any of the
data entry fields on the Main Screen. The Edit menu also accesses the Main
Screen Notes and the Update Measurement Units screen.
Report
Analysis reports can be previewed and printed via the Report menu. The
Report Options screen can also be accessed for entering Administrative data
and selecting the reports you’d like to print. The report fonts
and nomenclature can also be set via the Report menu.
7
Setup
The Setup menu provides a list of settings that you can customize and us as defaults. Selecting
General opens the Options screen where you can change or update the User Identification
information. Selecting File Locations opens the Options screen where you can view and modify the
default locations for the analysis, program, units database and help files. Selecting Report Fonts
opens the Options screen where you can choose the font types and sizes for various sections of the
analysis reports and set them as your defaults. Selecting Reports Nomenclature opens the Options
screen where you can enter preferred nomenclature to be used in the analysis reports and pareto diagrams.
Measurement Configuration
Selecting Measurement Configuration opens a screen that allows you to specify the measurement configuration that best
describes the relationship between the subject parameter and the measuring parameter.
Plot Options
The Plot Options menu allows you to select the desired settings to be used for the Combined
Error Distribution plot that is displayed on the Main Screen.
Pareto Diagram
The Pareto Diagram menu allows you to view the Summary or Detailed Pareto Chart screen.
Selecting Summary displays the pareto chart ranking the error component uncertainties from
most significant to least significant as determined by percent contribution
to total uncertainty. Selecting Detailed displays the pareto chart ranking all
pertinent measurement process error uncertainties from most significant to
least significant as determined by percent contribution to total uncertainty.
Run
The Run menu allows you to open an External Application; or open a Spreadsheet
or Calculator program set by the user. Selecting Options opens the Options screen where you can
specify the preferred spreadsheet and calculator you’d like to access from the Run menu.
Help
The Help menu provides access to on-screen help for
Uncertainty Sidekick’s features and functions. This includes
access to the Table of Contents, Using Help, and About
Sidekick.
3.2 The Analysis Path

The interactive Analysis Path, located on the left side of the Main Screen, provides a structured
walk-through of the basic steps in analyzing uncertainty for typical direct measurement
situations.
• Setup the Analysis – This step involves selecting the appropriate measurement
configuration and entering a nominal or expected value and selecting the measurement area
and units.
• Describe What is Being Measured – This step involves entering information for estimating
uncertainties due to subject parameter bias, resolution, repeatability or other errors.
• Describe the Measurement Reference – This step involves entering information for
estimating uncertainties due to measuring parameter bias, resolution, repeatability or other
errors.
• Describe the Measuring Environment – This step involves entering information for
estimating uncertainty due to environmental or other ancillary error sources. Correlations
between environmental or ancillary error sources are also accounted for in this step.
• Describe Who is Making the Measurement – This step involves entering information for
estimating the uncertainty due to operator bias and accounting for correlations.
8
Selecting any of the steps launches the appropriate analysis worksheet or screen. When an analysis step has been
completed, the Analysis Path step changes from Red to Gold.
3.3 Analysis Details

The Analysis Details section of the Main Screen is where key information about the uncertainty analysis is entered or
selected. This includes the analysis title, measurement area, nominal units, tolerance units, and nominal value (if
applicable). The analysis option, selected from the Measurement Configuration screen is also displayed in this section.
3.4 Error Component Table

Uncertainty Sidekick assumes that, in every direct measurement, there are two primary parameters: the subject parameter
and the measuring parameter. The subject parameter is the attribute that we seek to gain information about from a
measurement or set of measurements. The measuring parameter is a function or attribute of a measuring device that is
used to obtain information that allows us to quantify the subject parameter.
The Main Screen Error Component Table provides a summary of uncertainty analysis results for the subject parameter,
measuring parameter, measuring environment and operator. Each error component consists of measurement process
errors that can be analyzed via drill-down analysis worksheets and screens. These worksheets and screens can be
accessed by clicking the desired error component in the table. The functionality and use of the drill-down worksheets
and screens are described in detail in Chapters 4 through 6. Data can also be entered directly into the Error Component
table if desired.
Depending upon how an uncertainty estimate is arrived at, it is classified as Type A, Type B, or Type A,B. Type A
estimates are those that are determined statistically from sampled data. Type B estimates are those that are determined
heuristically, i.e., in the absence of sampled data. Type A,B estimates are determined through a combination of heuristic
and statistical methods.
Error limits are the (±) limits that are expected to contain errors from a given source with some specified level of
confidence or probability. Uncertainty or standard uncertainty is the uncertainty in the error or errors associated with
that source. If you enter a value in the Error Limits box and the estimated probability of containment in the Confidence
box, then Uncertainty Sidekick will compute the standard uncertainty and display it in the Uncertainty box. Uncertainty
can also be estimated using the drill-down worksheets or screens.
If a standard uncertainty is arrived at statistically from a sample of data, it has associated with it a sample size. In this
case, the degrees of freedom is simply the sample size minus one. If the standard uncertainty is determined
heuristically, then Uncertainty Sidekick automatically sets the associated degrees of freedom to infinite and displays a
“∞” symbol, unless otherwise specified by you. In some cases, the degrees of freedom for a type B uncertainty estimate
can be determined. This is done by using the Type B Degrees of Freedom Calculator that is built into Uncertainty
Sidekick. A “∞” symbol will also appear for uncertainty estimates with degrees of freedom exceeding 5,000.
9
3.5 Include in Analysis
After uncertainty estimates have been made for all relevant errors, they can be selected to be included in the computation
of the total combined uncertainty. You do this by selecting Include to the right of the appropriate error component. A
check mark instructs Uncertainty Sidekick to include the selected error component uncertainty in the computation of
total uncertainty that is displayed in the results section of the Main Screen. The Include feature also allows you to see
the impact of omitting an error component uncertainty on the total uncertainty.
3.6 Analysis Results

The Main Screen Analysis Results are displayed below the Error
Component table. The Total Uncertainty is the total standard
uncertainty resulting from the root sum square (RSS) combination of
the component uncertainties. If you have a statistics background,
you know that this method only works if the error sources are
independent of one another. For the overwhelming majority of
direct measurements, the error components listed on the Main screen
will be independent.
The Expanded Uncertainty is computed by multiplying the total uncertainty by the coverage factor obtained from the
t-distribution table for the displayed degrees of freedom and a user specified Confidence level. The degrees of freedom
for the total uncertainty is computed using the Welch-Satterthwaite formula
uT4
ν= n
ui4
∑ν
i =1 i
where uT is the total uncertainty, and ui and vi are the uncertainties and degrees of freedom for the component errors.
3.7 Combined Error Distribution

The combined error distribution is also displayed on the
Main Screen below the Error Component table. A normal or
student’s t distribution is displayed, depending upon the
degrees of freedom for the total uncertainty.
The Plot Options menu allows you to display 2-sigma limits

(i.e., twice the total uncertainty), 3-sigma limits, subject
parameter tolerance limits, or measuring parameter tolerance
limits.
You can also choose to center the plot about the nominal
value (i.e., 0 Deviation from Nominal Value), the bias offset
of the measuring parameter (if it has been specified), the bias offset of the measuring parameter (if it has been specified),
the subject parameter measured mean (if data have been entered), the measuring parameter measured mean (if data have
been entered), or estimated true value of the subject parameter bias.
3.8 Parameter Bias Estimates

The Parameter Bias Estimates section of the Main Screen shows estimates of bias and bias uncertainty for the subject
parameter and measurement parameter. Also shown is an estimate of the subject parameter's true value. These estimates
are obtained using a Bayesian method of analysis referred to as SMPC, which is discussed in more detail in Chapter 8.
10
Chapter 4 Example 1 – DC Voltmeter Calibration
In this chapter, we will use a simple example to go through the basic steps of estimating measurement process
uncertainties, component uncertainties and total uncertainty using Uncertainty Sidekick. This example consists of
calibrating the voltmeter function of a digital multimeter using a voltage reference with a nominal output of 10 VDC.
The following Uncertainty Sidekick worksheets and screens will be used for this analysis:
Main Screen
Measurement Configuration Screen
Subject Parameter Uncertainty Worksheet
Subject Parameter Bias Uncertainty Worksheet
Measurement Data Entry Worksheet
Subject Parameter Resolution Error Worksheet
Measuring Parameter Uncertainty Worksheet
Measuring Parameter Bias Uncertainty Worksheet
Report Options Screen
4.1 Measurement Process Overview

An Agilent 973A Digital Multimeter (DMM) has been purchased and calibrated at 10V DC using a Fluke 732B DC
Voltage Reference. The 10V nominal output of the DC voltage reference was measured with the 973A DMM in a
laboratory with temperature and relative humidity conditions of 25 °C ± 2 °C and 20% ± 5%, respectively. A total of 15
repeat measurements were recorded, as shown in the Table 4.1.
Table 4.1 Repeat Measurements from 973A DMM Calibration

Measurement DC Voltage
1 10.005
2 10.002
3 9.994
4 10.002
5 9.995
6 10.002
7 9.998
8 10.004
9 10.005
10 10.003
11 9.998
12 9.997
13 10.002
14 9.998
15 9.997
We want to estimate and report the total uncertainty in the mean or average DC voltage computed from the above data
sample. We will compare our analysis results to manufacturer specifications to decide if we should purchase additional
973A DMMs.
Specifications for Fluke 732B DC Voltage Reference

Manufacturer specifications1 state the acceptable operating conditions for the Fluke 732B to be 15 °C to 35 °C and a
relative humidity of less than 95% for temperatures up to 30 °C. The laboratory conditions during the calibration
process fall within these recommended temperature and humidity ranges.
1
Fluke 734A Reference & Transfer Standards Specifications, www.fluke.com.
11
Fluke conducts tests and applies statistical methods to determine the output stability of individual 732Bs and their group
mean with time.2 The output of the 732B is compared to a 10V Josephson Array and a linear regression curve is
obtained, as shown in Figure 4.1.
-9.2
-9.3
-9.4
-9.5
Output -9.6
Difference
(ppm) -9.7
Linear
-9.8 Regression
Curve
-9.9
-10.0
-10.1
0 10 20 30 40 50 60 70 80 90 100 110 120
Elapsed Time (days)
Figure 4.1 Output of 732B minus Output of 10V Josephson Junction
The random variation about the regression line is used to determine the time-dependent stability of the 732B output.
Fluke states a guaranteed stability of the 732B 10 V DC output to be ± 0.3 ppm/month and ± 2.0 ppm/year.2 The 732B
used in our calibration was manufactured 3.5 years ago, so the long-term stability specification applies for this analysis.
The linear regression equation (4-1) is used to estimate the bias offset of the 732B at some future time. Therefore, this
equation can be used to estimate the output of the 732B at the time the calibration measurements were made.
Bias offset = - 9.3 ppm - 0.0.00667 ppm/day × elapsed time (days) (4-1)
Specifications for Agilent 973A Digital Multimeter

Manufacturer specifications for the 973A DMM state a digital resolution of 1 mV and accuracy of 0.1% of reading for a
0-50 V range setting3. The environmental operating conditions for the DMM are 23 °C ± 5 °C and less than 80% relative
humidity.
4.2 Uncertainty Analysis Procedure

We will now go through the basic analysis steps, including those for entering information into the relevant Uncertainty
Sidekick worksheets and screens to estimate uncertainties in the measurement process errors and component errors; and
compute the total uncertainty in the 973A DC voltage reading.
Step 1. Define the physical quantity whose value is estimated via measurement. This quantity is often referred to as the
Measurand or Subject Parameter. In this analysis, the Subject Parameter is the DC voltage reading from the 973A
DMM. The Measuring Parameter is the output of the 732B voltage reference.
As discussed previously in the measurement process overview, the nominal reading from the 973A is dependent upon the
expected nominal output of the voltage reference at the time of measurement. The expected output from the voltage
reference after 3.5 years can be computed off-line in a spreadsheet application. The calculation is done as follows:
Calibration: Philosophy in Practice, 2nd Edition, Copyright © 1994 Fluke Corporation.

2
Agilent 970 Series Hand Held Multimeters Data Sheet, Copyright © 1998, 2000 Agilent Technologies, www.agilent.com.
3
12
Voltage Reference Output = 10 V + (- 9.3 – 0.00667 × 3.5 × 365) × (10 / 1,000,000) V
= 10 V – 0.000178 V
= 9.9998 V
Therefore, we expect the 732 B voltage reference to have an adjusted output of 9.9998 VDC. We will use this as our
nominal value for the analysis.
Step 2. Identify the measurement process errors. These fundamental error sources are the basic elements of uncertainty
analysis. In the DC voltage measurement process, the following errors must be considered:
• Bias in the value of the 732B voltage reference.

• Error associated with the stability of the voltage reference.
• Error associated with the repeat measurements taken.
• Error associated with the digital resolution of the 973A DMM.
Since the measurements were obtained under acceptable environmental conditions, as specified by both manufacturers,
we will not consider errors due to environmental factors. Additionally, operator error or bias is not considered relevant
to this measurement process.
In this analysis, both the subject parameter and measuring parameter are treated as toleranced values. We will not include
the bias uncertainty for the 973A DMM, since this is essentially what we are trying to estimate from our uncertainty
analysis. We will compare our estimated uncertainty in the 10V DC measurement to that specified by the manufacturer
using a method called statistical measurement process control (SMPC).
Step 3. Estimate the uncertainties in the specified measurement process errors and compute the error component
uncertainties and the total uncertainty. Uncertainty Sidekick will be used to estimate the uncertainties using the data
entry procedures outlined below.
1. Open the Uncertainty Sidekick program by double-clicking the associated desktop icon or select it from the
Windows Programs list. Uncertainty Sidekick’s Main Screen should be displayed once the program has been
activated. Enter the Analysis Details in the top portion of the screen as follows:
Main Screen
a. In the Analysis Title box, enter ‘Calibration of 973A Digital Multimeter at 10V DC Nominal Reading.’
b. Select DC Voltage from the Measurement Area dropdown list.
c. Select V from the Nominal Units dropdown list and select mV from the Tolerance Units dropdown list.
d. Enter 9.9998 for the Nominal Value.
e. Select Save Analysis from the File menu. Select the location where you want the file to be saved and enter ‘DC
Voltmeter Calibration’ for the file name.
The left side of the main screen contains an Analysis

Path that lists steps that we will use to enter the
information needed to estimate the process uncertainties
and the total uncertainty.
2. Click the first button in the Analysis Path entitled Setup

the Analysis. This activates the Measurement
Configuration Screen that allows you to specify the
measurement configuration that best describes the
relationship between the subject parameter and the
measuring parameter.

There are four configurations to choose from:
13
1) Active Reference Model, 2) Passive Reference Model, 3) Indirect Model and 4) General Model.
In this case, the 973A DMM (Subject Parameter) is measuring the output from the 732B voltage reference
(Measuring Parameter).
c. Select the Passive Reference Model for the measurement configuration.

d. Click OK to return to Uncertainty Sidekick’s Main Screen.
e. Select Save Analysis from the File menu.
The Analysis Details section of the Main Screen should now have the following information:
3. Click the button in the Analysis Path entitled Describe What’s Being Measured. This opens the Subject Parameter
Uncertainty Worksheet were basic descriptive information and other relevant details about the 973A DMM are
entered as follows:
a. In the Subject Parameter Identification section of the Worksheet, enter ‘Agilent’ for the Manufacturer, enter
‘973A’ for the Model Number, enter ‘1xyz’ for the Serial Number, enter ‘Digital Multimeter’ for the
Description, and enter ‘10 Volt DC Nominal Reading’ for the Parameter Name.
b. Click the Parameter Bias button in the Error Source table or click the Estimate the Parameter Bias Uncertainty
button in the Definition Path located on the right side of the worksheet. Either action will open the Subject
Parameter Bias Uncertainty Worksheet.
c. Make sure that Tolerance Data is selected for the Compute Bias Uncertainty From option and that Tolerance
Units is selected for the Display Tolerance Data option.
d. In the Tolerance Data section of the Worksheet, make sure that the Two Sided and Symmetric Tolerance
options are selected for the Tolerance Type.
e. For the Parameter Specs enter 0.1 % of Nominal for the Upper Limit. The upper and lower tolerance limits
should now have values of 10 mV.
f. Since the 973A DMM is an off-the-shelf item, experience dictates that an in-tolerance probability of 95% is
appropriate. Enter a value of 95.0 for the % In-Tolerance at Time of Measurement and press the Tab or Enter
key.
14
Based on the selected tolerance options and data entered, Uncertainty Sidekick should select the normal
distribution for the subject parameter bias and compute an estimated uncertainty in the bias of 5.1 mV.
g. Click OK on the Worksheet menu bar to save the entries and return to the Subject Parameter Uncertainty
Worksheet. Although we will not be including the subject parameter bias uncertainty in the overall uncertainty
analysis, we will use this information for the SMPC analysis which will be discussed later.
h. Click the Random Error button in the Error Source table or click the Estimate Repeatability Uncertainty button
in the Definition Path to open the Measurement Data Entry Worksheet for the subject parameter.
i. Make sure that Enter Individual Values is selected for the Data Sample option and that Measured Values is
selected for the Data Entry option.
j. Enter the repeat measurements listed in Table 4.1 into the Measurement Value table. The key sample statistics
should have the following values:
Mean Value = 10.0001 V
Sample Size = 15
Std. Uncertainty = 3.6 mV
Since we will be reporting the sample mean value, we must check the Use uncertainty in the Mean box. A standard
uncertainty of 0.9 mV should now be displayed.in the Measurement Data Entry Worksheet.
15
k. Click OK on the menu bar to save the entries and return to the Subject Parameter Uncertainty Worksheet.
l. Make sure that the Include option is checked for the Random Error uncertainty estimate.
m. Click the Resolution Error button in the Error Source table or click the Estimate Resolution Uncertainty button
in the Definition Path to open the Subject Parameter Resolution Error Worksheet.
n. Select Digital as the Display Type. Because the
resolution uncertainty is expressed in mV, enter 0 for
the Displayed Decimal Precision and press the tab
key. The Computed Resolution Error Limits should
be 0.5 mV with 100% Confidence Level.
The Computed Standard Uncertainty for the digital
resolution error should be 0.29 mV.
o. Click OK on the menu bar to return to the Subject
Parameter Uncertainty Worksheet.
p. Make sure that the Include option is checked for the
Resolution Error uncertainty estimate.
We have input all relevant information and data
needed to estimate the measurement process
uncertainties for the Subject Parameter.
Remember that we are not including the subject parameter bias uncertainty, because this is what we are trying
to estimate from our analysis.
A Combined Uncertainty of 0.9452 mV and 17 Degrees of Freedom should be displayed for the 973A DMM
(Subject Parameter).
The completed Subject Parameter Uncertainty Worksheet is shown on the following page for reference.
q. Click OK on the menu bar to return to the Uncertainty Sidekick Main Screen.
Main Screen
r. Select Save Analysis from the File menu to save what we have done so far.
4. Click the button in the Analysis Path entitled Describe the Measurement Reference. This opens the Measuring
Parameter Uncertainty Worksheet were basic descriptive information and other relevant details about the 732B
Voltage Reference are entered.
16
a. In the Measuring Parameter Identification section of the Worksheet, enter ‘Fluke’ for the Manufacturer, enter
‘732B’ for the Model Number, enter ‘3abc’ for the Serial Number, enter ‘DC Voltage Reference’ for the
Description, and enter ‘10 Volt DC Output’ for the Parameter Name.
button in the Definition Path located on the right side of the worksheet. Either action will open the Measuring
17
The Fluke 732B has a time-dependent stability specification of ± 2.0 ppm/year. Therefore, we compute the
stability specification at 3.5 years and convert it to desired mV units.
Stability Spec = ± 2.0 × (10 ÷ 1,000,000) × 3.5 × 1,000 mV

= ± 0.07 mV
e. Enter a Fixed Limit of 0.07 for the Upper Limit. The upper and lower tolerance limits should now have values
of 0.07 mV. From experience, we know that the Fluke 732B tolerance specs routinely exceed a coverage factor
of 3. Enter a value of 99.0 for the % In-Tolerance at Time of Measurement and press the Tab or Enter key.
distribution for the measuring parameter bias and compute an estimated uncertainty in the bias of 0.027 mV.
f. Click OK on the Menu to save the entries and return to the Measuring Parameter Uncertainty Worksheet.
g. Make sure that the Include option is checked for the Parameter Bias uncertainty estimate.
18
A Combined Uncertainty of 0.027 mV and infinite Degrees of Freedom should be displayed for the 732B
voltage reference (Measuring Parameter).
h. Click OK on the menu bar to return to the Uncertainty Sidekick Main Screen.
Main Screen
i. Make sure that the Include option is selected for Subject Parameter and Measuring Parameter Error
Components.
j. Select Two-Sigma Limits and Plot About the SU Measured Mean from the Plot Options menu.
k. Select Save Analysis from the File menu.

On the Uncertainty Sidekick Main Screen, an SU Mean Value of 10.0001 V, Total Uncertainty of 0.950 mV and
Degrees of Freedom of 17 should be displayed below the Error Component table. The Error Distribution plot should be
displayed.
The Parameter Bias Estimates are displayed at the bottom of the Main screen. These values are estimated using SMPC
analysis, which is a new and powerful way to estimate subject parameter and measuring parameter biases and the in-
tolerance probabilities for these biases. For this analysis, the displayed results should be
19
Measuring Subject Estimated
Parameter Parameter True Value
Estimated Bias 0.0000 mV 0.3216 mV 9.9994784 V
Uncertainty 0.0272 mV 0.9575 mV 0.9575 mV
In-Tol Probability 99.00 % 100.00 %
If there were no bias in the DC voltage recorded with the 973A DMM, these measurements could be taken as faithful
representations of the voltage reference's true value. The estimated true value for the bias accounts for the fact that some
unknown bias may exist in the DMM voltage measurements. This bias estimate is based on the uncertainty in the bias of
the voltage reference, on the uncertainty in the bias of the DMM voltage readings, and on the uncertainty of other error
components in the measurement process.
Recall that we assumed a 95% a priori in-tolerance probability relative to the manufacturer's ± 10 mV tolerance limits.
By comparison, the SMPC analysis estimates an a posteriori estimated bias of 0.32 mV with an uncertainty of 0.96 mV
and an in-tolerance probability of 100%. Consequently, the results of our uncertainty/SMPC analysis indicate that the
973A DMM performed very well relative to the manufacturer specifications.
We have accounted for all measurement uncertainties that are relevant to the calibration process and are now ready to
print the results of our analysis. In calibration, we are typically concerned about uncertainty for two main reasons. First,
we want to know whether we have excessive uncertainties due to sources of error in our measurement process. Second,
we want to communicate measured parameter values and uncertainty limits associated with these measured values to
those using this device.
20
Step 4. Print the analysis results. The simplest way to review and print our analysis results is via the Report Options
screen, which is accessed from Uncertainty Sidekick’s Main Screen by selecting Options from the Report menu.

1. Enter the following information into the Administrative Data section of the screen:
Report Title: Calibration of 973A Digital Multimeter at 10V DC Nominal Reading

Company: ACME Industries Location: Bakersfield, CA.
Submitted By: enter your name Title: Metrology Engineer
Organization: Calibration Lab Date: today’s date
2. Check the Include buttons to include the above information in the analysis report. Also check the Include File
Name.
3. In the Report Options section of the Screen, select Summary Report and Include Pareto.
4. Select Preview from the Print menu to display the Summary Report in Print Preview mode prior to printing.
Prior to printing, you may also want to enter a preferred nomenclature for the report. This can be done via the Options
screen, which is accessed by selecting Report Nomenclature from the Options menu.
Options Screen
5. Enter the following preferred report nomenclature
Default Preferred
SU Measurand
MTE Reference
Confidence Limit Tolerance Limit
Parameter Name Measured Quantity
Subject Unit Unit Under Test
Subject Parameter Measurand
Measuring Unit Standard
Measuring Parameter Reference
Operator Bias Reproducibility
Random Repeatability
SMPC Bayesian Analysis
21
Items in the nomenclature list that you leave blank in the preferred column will automatically use the default term.
6. Click the OK button to accept these entries and return to the Main Screen.
You can also set the preferred nomenclature to be displayed on the Pareto Diagram by selecting the Use Reports
Nomenclature on the Pareto Chart screen, which is accessed by selecting Summary or Detailed from the Pareto
Diagram menu on the Main Screen.
Pareto Chart Screen

7. Select the Use Reports Nomenclature option at the bottom of the screen
8. Click OK on the menu to return to the Main Screen.
There are a few features for displaying the Pareto Chart that is worth discussing. The Diagram menu provides options
for displaying the chart in 3D mode, vertical or horizontal orientation, with a gray or clear background. The View menu
provide options for display the chart only, the table only, or both the table and chart as shown above. The Copy menu
has options for copying selected cells, all cells, the table, or the chart. Options for selecting the report fonts and
nomenclature, previewing and printing are listed in the Report menu. The Run menu provides options for running
external applications.
Main Screen
9. Select Print Preview from the Report menu to display the Summary Report in Print Preview mode prior to printing.
A copy of the Summary Report for this analysis is shown on the following page.
22
Uncertainty Analysis Report Measurement Uncertainty
Calibration of 973A Digital Multimeter at 10V DC Nominal Reading

ACME Industries - Bakersfield, CA
26-Jan-2008
Submitted:
Joe Black
Metrology Engineer, Calibration Lab
6/26/07
File Name: DC Voltmeter Calibration.sdk

Subject Unit
Manufacturer: Agilent
Model Number: 973A
Description: Digital Multimeter
Measured Quantity: 10 Volt DC Nominal Reading
Estimated Parameter Value: 10.0001 V
Measuring Unit
Manufacturer: Fluke
Model Number: 732B
Description: DC Voltage Reference
Measured Quantity: 10 Volt DC Output
Analysis Results
Standard Confidence
Uncertainty Confidence Deg. Limits
Uncertainty Component (mV) Level Freedom (mV) Type
Measurand 0.95 95.00 17 2.00 A,B
Reference 0.027 95.00 Infinite 0.053 B
Combined Uncertainty 0.950 mV 95.00 17 2.004 mV A,B
Pareto Diagram
Rank Error Component Type Weight (%)
1
1 Measurand Repeatability A 73.952
2 Measurand Resolution B 23.829 2
3 Reference Bias B 2.219 3
4
5
6
7
8
9
10
0 20 40 60 80
23
Chapter 5 Example 2 – Micrometer Calibration
In this chapter, we will conduct an analysis that uses a few more worksheets to illustrate the capabilities of Uncertainty
Sidekick to assist you in estimating measurement process uncertainties. This example consists of calibrating a
micrometer with a gage block reference. In this analysis, we will also need to account for correlations between error
sources.
Main Screen
Measuring Environment Uncertainty Worksheet
Correlation Analysis Worksheet
Operator Bias Worksheet
Operator Bias Uncertainty Worksheet

A Mahr 40A 0-25 mm analog micrometer is calibrated at 10 mm using a Stahl 402 10-30 mm Class 2 (Grade 2) gage
block set. Multiple readings of the 10 mm gage block length were taken with the Mahr 40A micrometer under
laboratory environmental conditions of 24 °C ± 1 °C. The repeat measurements are listed in Table 5.1.
Table 5.1 Calibration Readings for Mahr 40A Micrometer

Measured Deviation from
Reading Length Nominal
(mm) (µm)
1 9.98 -20
2 10.01 10
3 10.03 30
4 9.97 -30
5 10.01 10
6 9.99 -10
7 9.98 -20
8 10.02 20
In this example, both the subject parameter and measuring parameter are treated as toleranced values. We want to
estimate and report the total uncertainty in the average length computed from sampled data. We also want to determine
whether the micrometer is within manufacturer tolerances for the 10 mm readings.
Specifications for the Gage Block Reference

The tolerance specifications for the Class 2 gage block set are obtained from tabulated data published by NIST4. A
subset of the data are listed in Tables 5.2 and 5.3. Gage block length is defined at the following standard reference
conditions:
temperature = 20 °C (68 °F) barometric pressure = 101.325 KPa (14.7 psia)
CO2 content of air = 0.03% water vapor pressure = 1.33 KPa (10 mm of mercury)
4
The Gage Block Handbook, NIST Monograph 180, 1995.
24
Only temperature has a measurable effect on the physical length of the gage block as a result of thermal expansion or
contraction. Gage block steel has a thermal expansion coefficient of 11.5 × 10-6/°C.
Table 5.2 Tolerance Grades for Metric Gage Blocks (µm)

Nominal Grade .5 Grade 1 Grade 2 Grade3
< 10 mm 0.03 0.05 +0.10, -0.05 +0.20, -0.10
< 25 mm 0.03 0.05 +0.10, -0.05 +0.30, -0.15
< 50 mm 0.05 0.10 +0.20, -0.10 +0.40, -0.20
< 75 mm 0.08 0.13 +0.25, -0.13 +0.45, -0.23
< 100 mm 0.10 0.15 +0.30, -0.15 +0.60, -0.30
Table 5.3 Additional Tolerance for Length, Flatness, and Parallelism (µm)
Nominal Grade .5 Grade 1 Grade 2 Grade3
< 100 mm ± 0.03 ± 0.05 ± 0.08 ± 0.10
< 200 mm ± 0.08 ± 0.15 ± 0.20
< 300 mm ± 0.10 ± 0.20 ± 0.25
< 500 mm ± 0.13 ± 0.25 ± 0.30
Specifications for the Analog Micrometer

Manufacturer specifications5 for the Mahr 40A micrometer state an analog resolution of 0.01 mm (10 µm) and error
limits of ± 0.004 mm (4 µm). No limiting range of operating conditions are stated for the device. For this example, the
thermal expansion coefficient of the micrometer is taken to be 5.6 × 10-6/deg °C.

We will now go through the basic analysis steps, including those for entering information into the Uncertainty Sidekick
worksheets and screens to estimate uncertainties in the measurement process errors and component errors; and compute
the total uncertainty in the micrometer reading.
Step 1. Define the physical quantity whose value is estimated via measurement. In this analysis, the Subject Parameter
is the length reading from the Mahr 40A analog micrometer. The Measuring Parameter is the length of the Class 2 gauge
block, which has a nominal or stated value of 10 mm.
Step 2. Identify the measurement process errors. In the 10 mm length measurement process, we must account for the
following errors:
• Bias in the value of the 10 mm gage block length.

• Error associated with the repeat measurements taken.
• Error associated with the analog resolution of the Mahr 40A micrometer.
• Operator bias resulting from his/her perception of the analog readings.
• Environmental factors errors resulting from thermal expansion of the gage block and the micrometer.
In this analysis, we will not include the bias uncertainty for the Mahr 40A, since this is what we are trying to estimate
from our uncertainty analysis. We will compare our estimated uncertainty in the gage block length measurement to that
specified by the manufacturer using SMPC.
uncertainties and the total uncertainty. We will use Uncertainty Sidekick to estimate the uncertainties using the data entry
procedures outlined below.
5
Mahr 40A Micrometer Product Page, www.mahr.com/mahrfederal/content/product/handtools/micrometer/40a.html.
25
Main Screen
a. In the Analysis Title box, enter ‘Calibration of Mahr 40A Analog Micrometer at 10 mm Nominal Length.’
b. Select Length from the Measurement Area dropdown list. The Nominal Units and Tolerance Units will
automatically be set to m (meters).
c. Select mm (millimeters) from the Nominal Units dropdown list and select um (micrometers) from the Tolerance
Units dropdown list and enter a value of 10 for the Nominal Value.
d. Select Save Analysis from the File menu. Select the location where you want the file to be saved and enter
‘Analog Micrometer Calibration’ for the file name.
The left side of the main screen contains an Analysis Path that lists the steps that we will use to enter the
information needed to estimate the process uncertainties and the total uncertainty.
2. Click the Analysis Path button entitled Setup the Analysis

to open the Measurement Configuration screen.
The 40A (Subject Parameter) is measuring the length of
the gage block (Measuring Parameter).
a. Select the Passive Reference Model for the
measurement configuration.
b. Click OK to return the Main screen.
c. Select Save Analysis from the File menu.
Main Screen
The Analysis Details section of the Main Screen should
have the following information:
3. Click the Analysis Path button entitled Describe What’s Being Measured to open the Subject Parameter
Uncertainty Worksheet.
a. In the Subject Parameter Identification section of the Worksheet, enter ‘Mahr’ for the Manufacturer, ‘40A’ for
the Model Number, ‘am556123’ for the Serial Number, ‘0-25 mm Analog Micrometer’ for the Description, and
‘10 mm Micrometer Reading’ for the Parameter Name.
26
b. Click the Parameter Bias button in the Error Source table or click the Definition Path button entitled Estimate
the Parameter Bias Uncertainty to open the Subject Parameter Bias Uncertainty Worksheet.
e. For the Parameter Specs enter a Fixed value of 4 for the Upper Limit and press the Tab key. The upper and
lower tolerance limits should now have values of 4 um.
f. Since the Mahr 40A micrometer is an off-the-shelf item, experience dictates that an in-tolerance probability of
85% is appropriate. Enter a value of 85.0 for the % In-Tolerance at Time of Measurement and press the Tab or
Enter key.
distribution for the subject parameter bias and compute an estimated uncertainty in the bias of 2.8 um.
27
j. Enter the repeat measurements listed in Table 5.1 into the Measurement Value table. The key sample statistics
should have the following values:
Mean Value = 9.999 mm
Sample Size = 8
Std. Uncertainty = 21.7 um
Since we will be reporting the sample mean value, we must check the Use Uncertainty in the Mean box. A
standard uncertainty of 7.7 um should now be displayed.
m. Click the Resolution Error button in the Error Source
table or click the Definition Path button entitled
Estimate Resolution Uncertainty to open the Subject
Parameter Resolution Error Worksheet.
n. Select Analog as the Display Type. Enter a value of
10 um for the Smallest Increment of Resolution and
95.0 for the Confidence Level.
resolution error should be 2.6 um.
28
We have input all relevant information and data needed to estimate the measurement process uncertainties for
the Subject Parameter.
Remember that we are not including the subject parameter bias uncertainty, because this is what we are trying
to estimate from our analysis.
A Combined Uncertainty of 8.1 um and 9 Degrees of Freedom should be displayed for the 40A micrometer
reading (Subject Parameter).
q. Click OK on the menu bar to return to the Main Screen.

Main Screen
Parameter Uncertainty Worksheet were basic descriptive information and other relevant details about the Stahl gage
block reference are entered.
a. In the Measuring Parameter Identification section of the Worksheet, enter ‘Stahl’ for the Manufacturer, ‘402’
for the Model Number, ‘gb13578’ for the Serial Number, ’10-30 mm Class 2 Gage Block Set’ for the
Description, and ‘10 mm Gage Block Length’ for the Parameter Name.
29
d. In the Tolerance Data section of the Worksheet, make sure that the Two Sided option is selected and the
Symmetric Tolerance option is unchecked.
e. Enter a Fixed Limit of 0.18 for the Upper Limit and 0.13 for the Lower Limit. For this analysis, we will assume
that the gage block tolerance specs have a containment probability or confidence level of 90%. Enter a value of
90.0 for the % In-Tolerance at Time of Measurement and press the Tab or Enter key.
30
Based on the selected tolerance options and data entered, Uncertainty Sidekick should have computed an
estimated uncertainty in the bias of 0.091 um.
A Combined Uncertainty of 0.091 um and infinite Degrees of Freedom should be displayed for the gage block
length (Measuring Parameter).
h. Click OK on the menu bar to return to the Uncertainty Sidekick Main Screen.
Main Screen
i. Select Save Analysis from the File menu.
j. Make sure that the Include option is selected for Subject Parameter and Measuring Parameter Error
Components.
During the micrometer calibration, the lab temperature was determined to be 24 °C ± 1 °C with 95% confidence. We
need to correct both the micrometer reading and the gage block length to the standard reference temperature of 20 °C.
First, we need to do some off-line calculations to estimate the net length expansion resulting from the + 4 °C difference
between the lab and reference temperatures. We will use this to adjust or correct our length measurements to the standard
temperature of 20 °C.
Second, we need to estimate the uncertainty of the length expansion resulting from the ± 1 °C temperature variation. We
will use the data listed in Table 5.4 for our calculations.
31
Table 5.4 Thermal Expansion and Interaction Coefficients for Gage Block and Micrometer
Expansion Nominal Units Interaction
Coefficient Value Conversion Nominal Coefficient
Description ( / °C) (mm) (um/mm) Value (um) (um/°C)
Gage Block Length 1.15E-05 10 1000 10000 1.15E-01
Micrometer Reading 5.60E-06 10 1000 10000 5.60E-02
The net length expansion is computed using equation (5-1).
Net Expansion = Temp. Diff. × Nominal Value × (Gage Block Exp. Coeff. – Micr. Exp. Coeff) (5-1)
= 4 °C × 10 mm × (1.15E-05 – 5.6E-06) / °C
= 2.35 E-04 mm
Therefore, we need to adjust our length measurements by -2.35E-04 mm or - 0.235 um. We do this in the Measurement
Data Entry Worksheet, which is accessed from the Subject Parameter Uncertainty Worksheet.
5. Click the Subject Parameter button in the Error Component table in the Main Screen to open the Subject Parameter
a. Click the Random Error button in the Error Source table to open the Measurement Data Entry Worksheet.
b. Select Deviation from Nominal for the Data Entry option.
c. Enter - 0.235 for the Environmental Correction and check the Include box below. The Mean Deviation should
change from – 1.3 to - 1.5 um.
d. Select Measured Values for the Data Entry option. A Mean Value of 9.9985 should now be displayed.
e. Click OK on the menu to accept these changes and return to the Subject Parameter Uncertainty Worksheet.
f. Click OK on the menu to return to the Main Screen.
Main Screen
g. Select Save Analysis from the File menu to save what we have completed thus far.
We will now estimate the uncertainty in the net length expansion correction.
6. Click the button in the Analysis Path entitled Describe the Measuring Environment. This opens the Measuring
Environment Uncertainty Worksheet were we will estimate the uncertainty due to thermal expansion of the gage
block and the micrometer.
a. Enter ‘Amb Temp – Gage Block’ for the first Environment Parameter, select Temperature Interval from the
Measurement Area drop-down list, enter 1.0 for the Error Limits, and select deg C from the Units drop-down
list.
b. Enter 95.0 for the % In-Tolerance and 0.115 for the Interaction Coefficient and press the Tab key. A standard
uncertainty of 0.059 um should be displayed for the gage block expansion.
c. Enter ‘Amb Temp – Micrometer’ for the second Environment Parameter, select Temperature Interval from the
list.
d. Enter 95.0 for the % In-Tolerance and 0.056 for the Interaction Coefficient and press the Tab key. A standard
uncertainty of 0.0286 um should be displayed for the micrometer expansion.
A Combined Uncertainty of 0.065 um should now be displayed at the bottom of the worksheet. However, we know
that the temperature correction errors for the gage block and micrometer are not independent. In fact, they are
strongly correlated by the laboratory temperature error.
32
We will use the Correlation Coefficients Worksheet to correct our combined uncertainty estimate.
e. Select Correlations from the Measuring Environment Uncertainty Worksheet menu. This will open the
Correlation Coefficients Worksheet.
Correlation Coefficients Worksheet
f. Select Amb Temp – Gage Block from the Variable 1 list and select Amb Temp – Micrometer from the
Variable 2 list.
We know that both the gage block and micrometer length corrections will err in the same direction and by
proportional amounts as temperature changes. We also know that the thermal expansion in the gage block is
partially offset by the thermal expansion in the micrometer. Therefore, the temperature corrections are
considered to be compensating and a correlation coefficient of - 1 should apply.
This is best illustrated by considering the variance addition rule,6 which is the fundamental basis for
combining uncertainties. The combined uncertainty for two error sources, ε1 and ε2, is
uT = var(ε1 + ε 2 ) = u12 + u22 + ρ1,2 u1u2
where u1 and u2 are the uncertainties in ε1 and ε2, respectively, and ρ1,2 is the correlation coefficient for the two
error sources. The correlation coefficient can range in value from -1 to +1.
If the two error sources are statistically independent, then ρ1,2 = 0 and the uncertainties are combined in root-
sum-square (RSS). However, if the two error sources are strongly correlated, then ρ1,2 = - 1 or +1. If the errors
are positively correlated, then ρ1,2 = 1 and the uncertainties are combined linearly. Conversely, if the errors are
negatively correlated (compensating), then ρ1,2 = - 1 and the combined uncertainty is the absolute value of the
difference in the uncertainties.
ρ1,2 = 0 : uT = u12 + u22
ρ1,2 = +1 : uT = u12 + u22 + 2u1u2 = u1 + u2
ρ1,2 = -1 : uT = u12 + u22 − 2u1u2 = u1 − u2
g. Enter a value of - 1 for the Correlation Coefficient.
6
Castrup, H. T.: “Estimating and Combining Uncertainties,” presented at the 8th Annual ITEA Instrumentation Workshop, Lancaster,
CA, May 5, 2004.
33
h. Click OK to accept these entries and return to the Measuring Environment Uncertainty Worksheet.
A Combined Uncertainty of 0.030 um should now be displayed. This is the expected combined uncertainty we
get if we subtract the two uncertainties listed in the Environmental Factors table.
i. Click OK on the worksheet menu to save our calculations and return to the Main Screen.
Main Screen
j. Make sure that the Include option is selected for Measuring Environment Error Component.
k. Select Save Analysis from the File menu to save what we have done so far.
Since the micrometer provides an analog reading, the operator’s perception of this reading can vary with his or her
perspective relative to the indicated value. In addition, the operator may not consistently tighten the micrometer enough
to get a flush fit with the gage block.
Therefore, we need to factor in some uncertainty due to operator error or bias. In this example, let us suppose that 95%
of the time the operator bias will not exceed 50% of micrometer resolution.
7. Click the button in the Analysis Path entitled Describe Who is Making the Measurement. This opens the
Operator Uncertainty Worksheet.
Operator Uncertainty Worksheet
a. Click the Analyze button for the Subject Parameter to open the Operator Bias Uncertainty Worksheet.
Operator Bias Uncertainty Worksheet
b. Enter 50 for the % of Resolution and 95.0 for the Confidence Level. A value of 2.55 um should be computed
for the operator bias uncertainty.
c. Click OK to accept these entries and return to the Operator Uncertainty Worksheet.
34
Operator Uncertainty Worksheet
d. Click the Include button for the Subject Parameter to include the Operator Bias Uncertainty.
e. Click OK on the menu to return to the Main Screen.

Main Screen
f. Make sure that the Include option is selected for Operator Error Component.
g. Select Two-Sigma Limits and Plot About the SU Measured Mean from the Plot Options menu.
h. Select Save Analysis from the File menu to save our analysis.

On the Uncertainty Sidekick Main Screen, a SU Mean Value of 9.9985 mm, Total Uncertainty of 8.502 um and
Degrees of Freedom of 11 should be displayed below the Error Component table. The Error Distribution plot should
also be displayed.
analysis. In this analysis, the displayed results should be
35
Estimated Bias 0.0001 um -0.1215 um 10.000 mm
Uncertainty 0.0908 um 2.6402 um 2.6402 um
Recall that we assumed a 85% a priori in-tolerance probability relative to the manufacturer's ± 4 um tolerance limits. By
comparison, the SMPC analysis estimates an a posteriori estimated bias of - 0.1215 um with an uncertainty of 2.6402
um and an in-tolerance probability of 87%. Consequently, the results of our uncertainty/SMPC analysis indicate that the
Mahr 30A analog micrometer performed consistently relative to the manufacturer specifications.
print the results of our analysis.

Report Title: Calibration of Mahr 40A Analog Micrometer at 10 mm Nominal Length

Submitted By: enter your name Title: Lab Tech
36
Organization: Calibration Lab Date: today’s date
2. Check the Include buttons to include the above information in the output report. Also check the Include File Name.
3. In the Report Options section of the Screen, select Summary Report and Include Pareto options.
In the previous example, we entered and used preferred nomenclature. The Options screen should still indicate that we
are using the Preferred Nomenclature.
4. Select Preview from the menu options to display the Summary Report in Print Preview mode prior to printing.
A copy of the Summary Report for this analysis is shown on the following page.
37
Calibration of Mahr 40A Analog Micrometer at 10 mm Nominal Length

ACME Industries - Bakersfield, CA
26-Jan-2008
Submitted:
John Smith
Lab Tech, Cal Lab
6/15/07
File Name: Analog Micrometer Calibration.sdk

Subject Unit
Manufacturer: Mahr
Model Number: 40A
Description: 0-25 mm Analog Micrometer
Measured Quantity: 10 mm Micrometer Reading
Estimated Parameter Value: 9.9988 mm
Measuring Unit
Manufacturer: Stahl
Model Number: 402
Description: 10-30 mm Class 2 Gage Block Set
Measured Quantity: 10 mm Gage Block Length
Analysis Results
Standard Confidence
Uncertainty Component (um) Level Freedom (um) Type
Measurand 8.11 95.00 9 18.35 A,B
Reference 0.091 95.00 Infinite 0.178 B
Environment 0.030 95.00 Infinite 0.059 B
Operator Bias 2.551 95.00 Infinite 5.000 B
Combined Uncertainty 8.502 um 95.00 11 18.713 um A,B
Pareto Diagram
1
2 Operator Bias B 19.742 2
5 Environment B 0.233 5
6
7
8
9
10
0 10 20 30 40 50 60
38
Chapter 6 Example 3 – Thermometer Calibration
In this example, we will conduct an analysis that uses two measurement samples. One sample consists of measurements
obtained with the thermometer being calibrated and the second sample consists of measurements obtained from a
temperature reference. This example will illustrate the Indirect Measurement Configuration option and how SMPC
works when data are collected for the subject parameter and the measuring parameter.
Main Screen
Measuring Parameter Resolution Error Worksheet

An off-the-shelf digital thermometer is calibrated at 100 °C using an oil bath and a standard platinum resistance
thermometer (SPRT). The oil bath temperature is set using its internal temperature probe and the readings from the
digital thermometer and the SPRT are recorded. This process was repeated multiple times . The resulting data sample is
listed Table 6.1.
Table 6.1 Repeat Measurements Thermometer Calibration

Digital Reference
Thermometer SPRT
Measurement (°C) (°C)
1 100.01 100.005
2 100.00 100.002
3 99.98 99.994
4 99.99 100.002
5 100.01 99.995
6 100.02 100.002
7 99.98 99.998
8 99.97 100.004
9 100.01 100.005
10 99.99 100.003
In this example, the digital thermometer is a toleranced parameter and the SPRT is treated as a reference standard,
complete with a calibration certificate. We will also use the manufacturer specifications for the oil bath temperature in
this analysis.
Specifications for the Oil Bath

A Hart Scientific model 6330 constant temperature bath was used for the calibration. The tolerance specifications7 for
this oil bath indicate a stability of ± 0.005 °C and uniformity of ± 0.007 °C at 100 °C. The manufacturer specifications
7
www.hartscientific.com
39
also indicate a digital setting accuracy of ± 0.5 °C and repeatability of ± 0.01 °C. The oil bath has a digital display
resolution of 0.01 °C. In this analysis, we will assume that all specifications, except the digital resolution, represent 95%
containment or confidence limits.
The stability specification is an indication of how well the bath can maintain its control temperature. The uniformity (or
non-uniformity) is an indication of the temperature variation within the bath. Uniformity primarily depends on how the
bath fluid is mixed. Non-uniformity could result in different temperatures at two locations in the bath.
Specifications for the SPRT

A Hart Scientific model 5699 SPRT was used as the temperature reference. The SPRT calibration certificate reports an
expanded uncertainty of ± 0.004 °C with a coverage factor of 2. The digital resolution of the SPRT reading is 0.001 °C.
Specifications for the Digital Thermometer

Manufacturer specifications8 for the ThermoProbe model TL-1A thermometer state a digital resolution of 0.01°C and a
calibrated accuracy of ± 0.07 °C below 150 °C. We will assume the accuracy specification to be the 95% containment or
confidence limits.

We want to estimate and report the total uncertainty in the average temperature reading from the digital thermometer.
We also want to determine how well the thermometer performs relative the manufacturer specified accuracy. We will
now go through the basic analysis steps, including those for entering information into the relevant Uncertainty Sidekick
screens and worksheets.
Step 1. Define the physical quantity whose value is estimated via measurement. In this analysis, the Subject Parameter
is the temperature reading from the TL-1A digital thermometer. The Measuring Parameter is the temperature reading
from the SPRT.
Step 2. Identify the measurement process errors. In this temperature measurement process, we must account for the
following errors:
Bias in the reference SPRT.

Error associated with the repeat measurements taken with the SPRT.
Error associated with the resolution of the SPRT.
Error associated with the repeat measurements taken with the digital thermometer.
Error associated with the resolution of the digital thermometer.
Error associated with the oil bath temperature.
As in the other examples, we do not include the bias uncertainty for the digital thermometer. This is what we are trying
to determine from our analysis.
uncertainties and the total uncertainty. We will use Uncertainty Sidekick to estimate the uncertainties using the data
entry procedures outlined below.
Main Screen
a. In the Analysis Title box, enter ‘Calibration of ThermoProbe TL-1A Digital Thermometer at 100 deg C.’
b. Select Temperature Interval from the Measurement Area dropdown list. The Nominal Units and Tolerance
Units will automatically be set to K (Kelvin).
c. Select deg C from the Nominal Units and Tolerance Units dropdown lists.
d. Enter a value of 100.00 for the Nominal Value.
8
www.thermoprobe-inc.com/pdf/data_TL1A.pdf
40
e. Select Save Analysis from the File menu. Select the location where you want the file to be saved and enter
‘Digital Thermometer Calibration’ for the file name.
The left side of the main screen contains an Analysis Path that lists the steps that we will use to enter the
information needed to estimate the process uncertainties
and the total uncertainty.
2. Click the Analysis Path button entitled Setup the

Analysis to open the Measurement Configuration screen.
The digital thermometer (Subject Parameter) and the
reference SPRT (Measuring Parameter) are both
measuring the oil bath temperature. Therefore, the
Indirect Model should be used.
a. Select the Indirect Model for the measurement
configuration.
b. Click OK to return the Main Screen.
c. Select Save Analysis from the File menu.
Main Screen
The Analysis Details section of the Main Screen should have the following information:
3. Click the Analysis Path button entitled Describe What’s Being Measured to open the Subject Parameter
a. In the Subject Parameter Identification section of the Worksheet, enter ‘ThermoProbe, Inc.’ for the
Manufacturer, ‘TL-1A’ for the Model Number, ‘tp1352’ for the Serial Number, ‘-43 to 315 deg C Digital
Thermometer’ for the Description, and ‘Thermometer Temperature Reading’ for the Parameter Name.
b. Click the Parameter Bias button in the Error Source table or click the Definition Path button entitled Estimate
the Parameter Bias Uncertainty to open the Subject Parameter Bias Uncertainty Worksheet.
41
e. For the Parameter Specs enter a Fixed value of 0.07 for the Upper Limit and press the Tab key. The upper and
lower tolerance limits should now have values of 0.07 deg C.
f. Enter a value of 95.0 for the % In-Tolerance at Time of Measurement and press the Tab or Enter key.
distribution for the subject parameter bias and compute an estimated uncertainty in the bias of 0.036 deg C.
42
j. Enter the repeat measurements for the digital thermometer listed in Table 6.1 into the Measurement Value table.
Since we will be reporting the sample mean value, we must check the Use Uncertainty in the Mean box. The
key sample statistics should have the following values:
Mean Value = 99.996 deg C
Sample Size = 10
Std. Uncertainty = 0.005 deg C
m. Click the Resolution Error button in the Error Source table or click the Definition Path button entitled Estimate
Resolution Uncertainty to open the Subject Parameter Resolution Error Worksheet.
n. Select Digital as the Display Type. Enter 2 for the
Displayed Decimal Precision.
resolution error should be 0.0029 deg C.
We have input all relevant information and data
needed to estimate the measurement process
uncertainties for the Subject Parameter.
Remember that we are not including the subject
parameter bias uncertainty, because this is what we are trying to estimate from our analysis.
An Estimated Parameter Value of 100.00 deg C and a Combined Uncertainty of 0.0058 deg C with 16 Degrees
of Freedom should be displayed for the thermometer temperature reading (Subject Parameter).
q. Click OK on the menu bar to return to the Main Screen.
Main Screen
43
Parameter Uncertainty Worksheet were information about the SPRT is entered.
a. In the Measuring Parameter Identification section of the Worksheet, enter ‘Hart Scientific’ for the
Manufacturer, ‘5699’ for the Model Number, ‘sprt1598’ for the Serial Number, ’High-Temperature Metal-
Sheath SPRT’ for the Description, and ‘SPRT Temperature Reading’ for the Parameter Name.
44
c. Select Reference Data is selected for the Compute Bias Uncertainty From option and make sure that
Uncertainty Units is selected for the Display Reference Data option.
d. In the Reference Data section of the Worksheet, enter an Expanded Uncertainty of 0.004 and a Coverage
Factor of 2 press the Tab or Enter key.
e. Check Same as Above for the Confidence Level at Time of Measurement. This will display the bias
distribution plot and compute a Bias Uncertainty at Time of Measurement of 0.002 deg C.
h. Click the Random Error button to open the Measurement Data Entry Worksheet for the measuring parameter.
45
j. Enter the repeat measurements for the SPRT listed in Table 6.1 into the Measurement Value table. Since we
will be reporting the sample mean value, we must check the Use Uncertainty in the Mean box. The key sample
statistics should have the following values:
Mean Value = 100.0010 deg C
Sample Size = 10
Std. Uncertainty = 0.0013 deg C
k. Click OK on the menu bar to save the entries and return to the Measuring Parameter Uncertainty Worksheet.
m. Click the Resolution Error button in the Error Source table or click the Definition Path button entitled Estimate
Resolution Uncertainty to open the Measuring Parameter Resolution Error Worksheet.
Measuring Parameter Resolution Error Worksheet
n. Select Digital as the Display Type. Enter 3 for the
Displayed Decimal Precision.
resolution error should be 0.00029 deg C.
o. Click OK on the menu bar to return to the Measuring
An Estimated Parameter Value of 100.001 deg C and
Combined Uncertainty of 0.0024 deg with 105 Degrees of
Freedom should be computed for the reference SPRT.
q. Click the OK to return to the Main Screen.
Main Screen
r. Select Save Analysis from the File menu to save what we have done thus far.
We now have account for the error in the oil bath temperature. The display resolution the digital setting accuracy
and repeatability do not introduce additional errors in our measurement process, because we are using the SPRT as
our temperature reference. Only the bath stability and uniformity contribute measurement error.
46
5. Click the button in the Analysis Path entitled Describe the Measuring Environment. This opens the Measuring
Environment Uncertainty Worksheet were we will estimate the uncertainty due to the oil bath temperature stability
and uniformity.
a. Enter ‘Oil Bath Stability’ for the first Environment Parameter, select Temperature Interval from the
list.
b. Enter 95.0 for the % In-Tolerance and 1.0 for the Interaction Coefficient and press the Tab key. A standard
uncertainty of 0.0026 deg C should be displayed for the oil bath stability.
c. Enter ‘Oil Bath Uniformity’ for the second Environment Parameter, select Temperature Interval from the
list.
d. Enter 95.0 for the % In-Tolerance and 1.0 for the Interaction Coefficient and press the Tab key. A standard
uncertainty of 0.0036 deg C should be displayed for the oil bath uniformity. A Combined Uncertainty of 0.0044
deg C should now be displayed at the bottom of the worksheet.
e. Click OK to accept these entries and return to the Main Screen.
47
Main Screen
f. Make sure that the Include option is selected for the Measuring Environment Error Component.
g. Select Two-Sigma Limits and Plot About the SU Measured Mean from the Plot Options menu.
h. Select Save Analysis from the File menu.

On the Uncertainty Sidekick Main Screen, an SU Mean Value of 99.996 deg C and Total Uncertainty of 0.00767 deg C
with 49 Degrees of Freedom should be displayed.
48
analysis. In this analysis, the displayed results should be

Estimated Bias 0.000 deg C -0.0049 deg C 100 deg C
Uncertainty 0.0020 deg C 0.0057 deg C 0.0057 deg C
Recall that we assumed a 95% a priori in-tolerance probability for the manufacturer's ± 0.07 deg C tolerance limits. By
comparison, the SMPC analysis estimates an a posteriori estimated bias of –0.005 deg C with an uncertainty of 0.0057
deg C and an in-tolerance probability of 100%. Consequently, the results of our uncertainty/SMPC analysis indicate that
the ThermoProbe TL-1A performed very good compared to the manufacturer specifications.
print the results of our analysis.

Report Title: Calibration of ThermoProbe TL-1A Digital Thermometer at 100 deg C

49
Submitted By: enter your name Title: Engineer
Organization: Test & Evaluation Date: today’s date
2. Check the Include buttons to include the above information in the output report. Also check the Include File Name.
3. In the Report Options section of the Screen, select Summary Report and Include Pareto options.
We will use the preferred nomenclature that we entered previously for Example 1.
4. Select Preview from the menu options to display the Summary Report in Print Preview mode prior to printing.
A copy of the Summary Report for this analysis is shown below.
50
Calibration of ThermoProbe TL-1A Digital Thermometer at 100 deg C

ACME Industries - Bakersfield
26-Jan-2008
Submitted:
Bob King
Engineer, Test & Evaluation
6/1/07
File Name: Digital Thermometer Calibration.sdk

Subject Unit
Manufacturer: ThermoProbe, Inc.
Model Number: TL-1A
Description: -43 to 315 deg C Digital Thermometer
Measured Quantity: Thermometer Temperature Reading
Estimated Parameter Value: 99.996 deg C
Measuring Unit
Manufacturer: Hart Scientific
Model Number: 5699
Description: High-Temperature Metal-Sheath SPRT
Measured Quantity: SPRT Temperature Reading
Analysis Results
Standard Confidence
Uncertainty Component (deg C) Level Freedom (deg C) Type
Measurand 0.0058 95.00 16 0.0123 A,B
Reference 0.00240 95.00 105 0.00476 A,B
Environment 0.0044 95.00 Infinite 0.0086 B
Operator Bias 0.0 95.00
Combined Uncertainty 0.00767 deg 49 A,B

C
Pareto Diagram
1
2 Environment B 27.640 2
5 Reference Repeatability A 8.187 5
6 Reference Resolution B 1.826
6
7
8
9
10
0 10 20 30 40
51
Chapter 7 Printing Analysis Results
Uncertainty Sidekick provides the capability to preview and print analysis results in various levels of detail. Reports can
be previewed and printed directly from nearly all of the analysis screens or worksheets via the Report Options Screen.
Reports can be printed to any printer that has the appropriate Windows printer driver. If you are using a laptop, the
analysis report print preview option is only active if you have a printer driver installed.
7.1 Report Options Screen

The Report Options Screen is accessed from the Main Screen by
selecting Options from the Report menu. This screen provides the
flexibility to tailor the information you’d like to include in your
analysis report.
The Administrative Data section of the screen allows you to specify

and include the following report elements:
Report Title
Company Location
Submitted by Title Organization Date
Approved by Title Organization Date
You can store these report elements as your default for future reports
by checking the Set Default button. To recall default entries, check
the Use Default button. You can also include the analysis file name
in the report for reference. Including a company or organization logo completes the customization of your report.
The Report Options section of the screen allows you to select the analysis report you’d like to preview and print.
A report is selected by clicking its leftmost button. Once a report is selected, you can include the Notes and Pareto
Diagram, if available. To review or update the report notes, click the View Notes button.
7.2 Selecting Report Fonts

Report fonts can be selected from the Options screen, which is accessed by
selecting the Fonts option from the Reports menu on the Main screen or by
selecting Report Fonts from the Options menu on the Report Options screen.
You can select the desired font types, styles and sizes for the report title,
subheadings, dates, labels and text. A list of available fonts for each report
element can be accessed by clicking the right hand button.
If you want the displayed font properties to be the default for reports, check the
Set Parameters as Defaults box. To restore the current defaults, click the Use
Current Defaults button. To reset the default properties to Uncertainty
Sidekick’s default setting, click the Reset Defaults button. Click OK to
implement the font properties and return to the Report Options screen.
7.3 Setting Reports Nomenclature

The nomenclature used in Uncertainty Sidekick has been established to be understood within the general technical
community. However, if this nomenclature is not considered appropriate for your specific reporting purposes, it may be
superseded via the Options screen. This screen can be accessed by selecting the Nomenclature option from the
Reports menu on the Main screen or by selecting Report Nomenclature from the Options menu on the Report Options
Screen.
Enter the terminology you want for your reports in the Preferred column of the displayed table. If a term is left
blank in the Preferred column, the default term will be used. If you want this terminology in your analysis
reports, select the Use Preferred Nomenclature option. Your choices will be implemented and will be in place
for future reports. If you want to revert to Uncertainty Sidekick’s normal terminology, select the Use Default
Nomenclature option. Click OK to accept your entries and return to the Main screen or the Report Options
screen.
52
For example, suppose that you are calibrating a device in a standards laboratory. You might want to enter your preferred
terminology as follows:
Default Preferred
SU Measurand
MTE Reference
Confidence Limit Tolerance Limit
Deviation
Manufacturer
Model Number
Description
Serial Number
Control Number
Parameter Name Measured Quantity
Qualifier
Measurement Area
Subject Unit Unit Under Test
Subject Parameter Measurand
Measuring Unit Standard
Measuring Parameter Reference
Nominal Value
Measuring Environment
Operator Bias Reproducibility
Random Repeatability
Resolution
Other
SMPC Bayesian Analysis
You can also set the preferred nomenclature to be displayed

on the Pareto Diagrams by selecting the Use Reports Nomenclature on the Pareto Chart screen. The Pareto Chart
screen can be access from most of the analysis worksheets or screens.
7.4 Previewing and Printing Reports

Reports can be previewed in the Print Preview
Screen, which is accessed by selecting Print
Preview from the Main screen File menu or the
Reports menu. The Print Preview screen can also
be accessed from the Report Options screen by
selecting the Print Preview option from the Print
menu.
By clicking Copy on the menu, you can copy the

selected report to the Windows clipboard for
insertion into a Word document, Excel spreadsheet
or other external application. The report is copied
as a Windows Enhanced Metafile picture, which
has limited format options such as resizing
cropping.
You can also access an external Windows

application by selecting the External Application
option from the Run menu.
Reports can be printed using the Windows Print

dialog, which is accessed by selecting the Print
Analysis Report option from the Main screen File
menu or the Print option from the Reports menu. The Print dialog can also be accessed from the Report Options screen
by selecting the Print option from the Print menu.
53
Chapter 8 Uncertainty Analysis Basics
Measurement uncertainty analysis has been under investigation for several decades, formally commencing with the work
of Churchill Eisenhart, Harry Ku, Josepth Pontius and others at the U.S. National Bureau of Standards (now the National
Institute of Standards and Technology). The topic received considerable impetus during the early to mid 1990’s with the
effort undertaken by the ISO TAG 4 of WG 3 in developing the Guide to the Expression of Uncertainty in Measurement.
This guide, often referred to as the GUM, comprises the current principle reference for measurement uncertainty
analysis.
While the GUM provides general rules for analyzing and communicating measurement uncertainty, it does not focus on
providing step-by-step instructions for evaluating specific measurement processes. As a result, testing and calibration
personnel may find it difficult or confusing to apply these principles to specific measurement scenarios. In an effort to
alleviate the challenging task of evaluating measurement uncertainty, Integrated Sciences Group has adopted, extended
and refined many of the methods in the GUM. The results of our efforts have been published and incorporated in our
training courses and our Uncertainty Sidekick and UncertaintyAnalyzer software packages.
This chapter discusses key concepts and methods that provide a sound basis for evaluating measurement uncertainty.
The general uncertainty analysis procedure consists of the following steps:
Define the Measurement Process

Identify the Error Sources and Distributions
Estimate Uncertainties
Combine Uncertainties
Report the Analysis Results
8.1 Define the Measurement Process

The first step in any uncertainty analysis is to identify the physical quantity whose value is estimated via measurement.
This quantity may be a directly measured value, such as the weight of a 1 gm mass or the output of a voltage reference.
Alternatively, the quantity may be indirectly determined through the measurement of other variables, as in the case of
estimating the volume of a cylinder by measuring its length and diameter. The former type of measurements are called
“direct measurements,” while the latter are called “multivariate measurements.”
For multivariate measurements, it is important to develop an equation that defines the mathematical relationship between
the quantity of interest and the measured variables.
S = f ( x, y , z ) (8-1)
where
S = subject parameter or quantity of interest
f = mathematical function that relates S to measured quantities x, y, and z.
At this stage of the analysis, it is also useful to briefly describe the test setup, environmental conditions, technical
information about the instruments, reference standards, or other equipment used and the procedure for obtaining the
measurement(s). This information will help identify the measurement process errors.
8.2 Identify Error Sources and Distributions

In any given measurement scenario, each measured quantity is a potential source of error. The basic error model for a
measured quantity xmeas is
xmeas = xtrue + εx (8-2)
where εx is the measurement error. In turn, εx, is the sum of the errors encountered during the measurement process and
can be expressed as
εx = ε1 + ε2 + ... + εn (8-3)
where the numbered subscripts signify the different measurement process errors. The errors most often encountered in
54
making measurements include, but are not limited to the following:
Measurement Bias - The bias in the measuring device and/or the quantity being measured.
Random or Repeatability Error - The error associated with repeat measurements.
Resolution Error - The error resulting from the finite resolution of the measuring device and/or the
quantity being measured.
Digital Sampling Error - The error introduced by digitizing an analog signal.
Computation Error - The error due to round-off or computer truncation, numerical interpolation,
empirically determined equations, etc.
Operator Bias - The error introduced by the person making the measurements.
Environmental Factors Error - The error introduced by variations in environmental conditions or by
correcting for environmental conditions.
Stress Response Error - Additional error resulting from stresses incurred during shipping and handling
to the end-user after calibration.
Measurement process errors are the basic elements of uncertainty analysis. Once these fundamental error sources
have been identified, we can begin to develop uncertainty estimates.
Another important aspect of the uncertainty analysis process is the fact that measurement errors can be characterized by
statistical distributions. This is stated in Axiom 1.
Axiom 1 - Measurement errors are random variables that follow statistical distributions.
This statement basically indicates that errors can be described in such a way that their sign and magnitude have some
definable probability of occurrence.
Normal
σ Lognormal
m
Uniform
σ
σ
m = Mean M
M = Mode or Nominal
σ = Standard Deviation m
Figure 8-1 Error Distributions and Statistics
The statistical distribution for a type of measurement error is a mathematical description that relates the frequency of
occurrence of values with the values themselves. Error distributions include, but are not limited to Normal, Lognormal,
Uniform (rectangular), Triangular, Quadratic, Cosine, Exponential, U-Shaped, Trapezoidal and Student's t.
Each distribution is characterized by a set of statistics. The statistics most often used in uncertainty analysis are the
mean, or the mode, and the standard deviation. With the lognormal distribution, a limiting value and the median value
are also used.
In general, the three error distributions that have been found to be relevant to most real world measurement applications
are the normal, lognormal and Student's t distributions. Other distributions such as the uniform, triangular, quadratic,
trapezoidal and U-shaped have more specialized application. Recommendations for selecting the appropriate distribution
for a particular measurement error source have been developed and documented by Integrated Sciences Group.9
Uncertainty Sidekick employs the normal, Student’s t and uniform distributions. The normal distribution is applied as
9
H. Castrup: “Distributions for Uncertainty Analysis,” presented at the 2001 International Dimensional Workshop in Knoxville, TN
(available for download at www.isgmax.com).
55
the default distribution. The Student's t distribution is applied if the underlying distribution is normal, but the degrees of
freedom are not infinite. The uniform distribution is applied for digital resolution error.
8.3 Estimate Uncertainties

As previously stated, all measurements are accompanied by error. Our lack of knowledge about the sign and magnitude
of measurement error is called measurement uncertainty. With a basic understanding of error distributions and their
statistics, we can estimate uncertainties. We begin with the statistical quantity called the variance, which is defined as
the mean square dispersion of the distribution about its mean or mode value.
var(x) = Mean Square Dispersion in x
If a variable x follows a probability distribution, described by a probability density function f(x), then the mean square
dispersion or variance of the distribution is given by
2
∞
var( x ) = ∫−∞ ( x − µ x ) f ( x )dx (8-4)
where µx is the mean of x, sometimes referred to as the expectation value for x. In speaking of variations in x that are the
result of measurement error, we take µx to be the true value of the quantity being measured. From equation (8-2), we can
write εx = x – µx, and equation (8-4) can be expressed as
∞ 2
var( x ) = ∫−∞ ε x f [ x(ε x ) ] d ε x
(8-5)
= var(ε x )
where x(εx) = εx + µx.
Because of the form of this definition, the variance is also referred to as the mean square error.
Equation (8-5) shows that, if a quantity x is a random variable representing a population of measurements, then the
variance in x is just the variance in the error in x.
var(x) = var(xtrue + εx) = var(εx) (8-6)
At this point, we recognize that the spread in an error distribution is quantified by the distribution’s standard deviation
and that the standard deviation is the square root of the variance. This leads to Axiom 2.
Axiom 2 - The uncertainty, u, in a measurement is the square root of the variance in the measurement error.
Axiom 2 provides the crucial link between measurement error and measurement uncertainty. If x is a measured value,
then we can write
u x = var( x )
= var(ε x ) (8-7)
= uε x .
Equation (8-7) provides a third axiom that forms a solid and productive basis upon which uncertainties can be estimated.
Axiom 3 - The uncertainty in a measured value is equal to the uncertainty in the measurement error.
Axiom 3, together with Axioms 1 and 2, allow the computation and combination of measurement uncertainty to be
rigorously carried out by
drawing attention to what it is that we are uncertain of in making measurements, and

allowing for the development of measurement uncertainty models for measurement scenarios of any complexity.
56
There are two approaches to estimating the variance of error distributions and, thus, uncertainty. Type A estimates
involve data sampling and analysis. Type B estimates use engineering knowledge or recollected experience of
measurement processes.
8.3.1 Type A Estimates

A Type A uncertainty estimate is defined as an estimate obtained from a sample of data. Data sampling involves making
repeat measurements of the quantity of interest. It is important that each repeat measurement is independent,
representative and taken randomly.
Random sampling is a cornerstone for obtaining relevant statistical information. Thus, Type A estimates usually apply
to the uncertainty in repeatability or random error.
Because the data sample is drawn from a population of values, we make inferences about the population from certain
sample statistics and from assumptions about the way the population of values is distributed. A sample histogram can
aid in our attempt to picture the population distribution.
Probability
Density
Sample Histogram
Population Distribution
εx
Figure 8.2 - Random Error Distribution
The normal distribution is ordinarily assumed to be the underlying distribution for random errors. When samples are
taken, the sample mean and the sample standard deviation are computed and assumed to represent the mean and standard
deviation of the population distribution. However, this equivalence is only approximate. To account for this, the
Student's t distribution is used in place of the normal distribution in computing confidence limits around the sample
mean.
The sample mean can be thought of as an estimate of the value that we expect to get when we make a measurement.
This "expectation value" is called the population mean, which is expressed by the symbol µ. The sample mean, x , is
obtained by taking the average of the sampled values. The average value is computed by summing the values sampled
and dividing them by the sample size, n.
1 1 n
x= ( x1 + x2 + ... + xn ) = ∑ xi (8-8)
n n i =1
The sample standard deviation provides an estimate of how much the population is spread about the mean value. The
sample standard deviation, sx, is computed by taking the square root of the sum of the squares of sampled deviations
from the mean divided by the sample size minus one.
1 n
∑ ( xi − x )
2
sx = (8-9)
n − 1 i =1
The value n-1 is the degrees of freedom for the estimate, which signifies the number of independent pieces of
information that go into computing the estimate. The greater the degrees of freedom, the closer the sample estimate will
be to its population counterpart. The degrees of freedom for an uncertainty estimate is useful for establishing confidence
limits and other decision variables.
57
We have already stated that the sample standard deviation is an estimate of the uncertainty in a value drawn randomly
from its population. However, if the estimate is to represent the uncertainty in the mean value rather than the uncertainty
in a single measurement, then the uncertainty in the mean value should be used. The uncertainty in the mean value,
s x , is equal to the standard deviation divided by the square root of the sample size.
sx
sx = (8-10)
n
Once estimates of the sample mean and standard deviation have been obtained, and the degrees of freedom have been
noted, it becomes possible to compute limits that bound the sample mean with some specified level of confidence. These
limits are called confidence limits and the degree of confidence is called the confidence level.
Confidence limits can be expressed as multiples of the sample standard deviation. For normally distributed samples, this
multiple is called the t-statistic. The value of the t-statistic is determined by the desired percent confidence level, C, and
the degrees of freedom, ν, for the sample standard deviation. Confidence limits around the sample mean are given by
sx
x ± tα / 2,ν (8-11)
n
where α = (1 - C/100) and ν = n - 1.
8.3.2 Type B Estimates

In some cases, we must attempt to quantify the statistics of measurement error distributions by drawing on our
recollected experience concerning the values of measured quantities or on our knowledge of the errors in these
quantities. Estimates made in this manner are called heuristic or Type B estimates.
Conceivably, a Type B uncertainty estimate could be obtained by just "winging it." The problem is, that most of us do
not have a point of reference for abstract quantities such as standard deviations or uncertainties. At best, we have a range
of values that we have experienced or are able to surmise. The limiting values that bound these ranges are called
containment limits. These limits can be viewed as bounding either measured values or measurement errors.
x lies within ± L
In working with Type A estimates, we start with sample statistics and work toward developing confidence limits that
bound values of a population with a specified confidence level or probability. In making Type B estimates, we apply this
process in reverse. We begin with containment limits and a containment probability, estimate the degrees of freedom,
and use these quantities to estimate the standard deviation or uncertainty.10
x lies within ± L with C% confidence or probability
Containment limits may be estimated from experience or taken from some documented reference, such as manufacturer
tolerance limits, stated expanded uncertainties obtained from calibration records or certificates, or statistical process
control limits. Containment probability can be obtained from service history data, for example, as the number of
observed in-tolerances, nin-tol, divided by the number of calibrations, N.
nin-tol
C % = 100%
N
If a heuristic estimate is obtained solely from containment limits and containment probabilities, then the degrees of
freedom is usually taken to be infinite.
For example, if the measurement error is normally distributed, the uncertainty is computed from the containment limits,
± L, the inverse normal distribution function, Φ-1(.) , and the containment probability, p = C/100.
10
Containment limits and containment probability are analogous to confidence limits and confidence level determined from
Type A estimates.
58
L
u= (8-12)
⎛ 1+ p ⎞
Φ −1 ⎜ ⎟
⎝ 2 ⎠
The inverse normal distribution function can be found in statistics texts and in most spreadsheet programs.
If there is an uncertainty in the containment limits (e.g., ±L ± ∆L) or the containment probability (e.g., ±p ± ∆p), then it
becomes imperative to estimate the degrees of freedom. As with Type A uncertainty estimates, the degrees of freedom
quantifies the amount of information that goes into the Type B uncertainty estimate and is useful for establishing
confidence limits and other decision variables.
Annex G of the GUM provides a relationship for computing the degrees of freedom for a Type B uncertainty estimate
−2
1 u2 ( x) 1 ⎡ ∆u ( x ) ⎤
ν≈ ≈
2 σ 2 [u( x )] 2 ⎢⎣ u( x ) ⎥⎦
(8-13)
where σ2[u(x)] is the variance in the uncertainty estimate, u(x), and ∆u(x) is the relative uncertainty in the uncertainty
estimate.11 Hence, the degrees of freedom for a Type B estimate is inversely proportional to the square of the ratio of the
uncertainty in the uncertainty divided by the uncertainty.
While this approach is intuitively appealing, the GUM offers no advice about how to determine σ2[u(x)] or ∆u(x).
Fortunately, since the publication of the GUM, a methodology for determining σ2[u(x)] and computing the degrees of
freedom for Type B estimates has been developed.12
Once the containment limits, containment probability and the degrees of freedom have been established, we can estimate
the standard deviation or uncertainty of the distribution of interest. For instance, if the measurement errors are normally
distributed, we can construct a t-statistic based on the containment probability and degrees of freedom. The uncertainty
estimate is then obtained by dividing the containment limit by the t-statistic, according to equation (8-14).
L
u= (8-14)
tα / 2,ν
8.3.3 Estimating Measurement Process Uncertainties

Now that we are sufficiently versed in the statistics of error distributions and uncertainty estimates, we can discuss
methods for determining the uncertainties associated with the measurement process. As previously discussed,
measurement processes are characterized by errors from the following sources:
What is being measured – measurand (subject parameter) bias and resolution error
What it is being measured with – measurement reference (measuring parameter) bias and resolution error
Who is making the measurement - operator bias
How the measurement is made - digital sampling and computation errors
Where the measurement is made - random and environmental factors errors
What happens after the measurement - stress response error
Bias uncertainty is usually estimated heuristically via containment limits and containment probabilities. In some
instances, as with reference standards, the stated parameter value and its associated uncertainty estimate may be supplied
by a higher-level calibration laboratory. In this case, the bias uncertainty might have been obtained from statistical
analysis of the calibration data and associated measurement process errors.
11
This equation assumes that the underlying error distribution is normal.
12
Castrup, H.: "Estimating Category B Degrees of Freedom," presented at the 2000 Measurement Science Conference, January 21,
2001 in Anaheim, CA.
59
Random uncertainty is estimated from a statistical analysis of a sample of measurements. The estimated uncertainty is
equal to the sample standard deviation. While the sample standard deviation is an estimate of the uncertainty in a value
drawn randomly from its population, we often want to estimate and report the uncertainty in the mean value of a random
sample. The uncertainty in the mean value is the sample standard deviation divided by the square root of the sample
size.
Resolution uncertainty due to the precision of a reading or output is usually determined heuristically. The containment
limits are based on the resolution of the device and an estimated containment probability. The distribution for resolution
errors depends on whether the reading or output is expressed in analog or digital format. For analog displays, the
resolution error is assumed to have a normal distribution. The resolution uncertainty is computed by setting the
containment limits equal to a smallest increment of resolution and applying a containment probability that readings can
be discerned within these limits. For digital displays, the resolution error is assumed to have a uniform distribution.
The containment limits are ± half the smallest displayed digit and the containment probability is equal to 100%.
Digital sampling uncertainty estimation involves defining a representative "signal" to be sampled and specifying a
sampling rate, a sampling aperture time, a quantization precision (bits), an impulse response and hysteresis, a sampling
noise level, and a sensor bias uncertainty. It also involves deciding on a model or methodology for eventual conversion
of digitized data back to analog form. This multi-faceted analysis process is difficult without a structured template
approach.13
Computation uncertainty can result from round-off or computer truncation error, the linear interpolation of tabulated
values, or using curve fit equations. Computation uncertainty is usually determined heuristically. The uniform
distribution is applicable for round-off or computer truncation error.
Operator bias uncertainty resulting from error in the perception of a human operator may be determined statistically,
but is usually estimated heuristically. The usual way is to link it to resolution uncertainty or some other aspect of a
measurement that can influence operator perception. The underlying error distribution is typically normal, but in some
cases, the uniform distribution may apply.
Environmental factors uncertainty is typically estimated heuristically in a three-step process:

Estimate the uncertainty in each environmental/ancillary process error.
Multiply each environmental/ancillary uncertainty by its respective interaction coefficient.
Combine the environmental/ancillary uncertainties, accounting for all correlations.
The interaction coefficient relates an environmental/ancillary factor to an error source. For example, if the error source
under consideration is the measurement of length and the environmental factor is temperature, then the interaction
coefficient is the thermal expansion coefficient for length.
Stress response uncertainty resulting from shipping and handling is usually estimated heuristically. The method of
determination involves attempting to estimate the uncertainty in the stresses encountered and multiplying this uncertainty
by a stress response coefficient (i.e., the response of the parameter value to stress). Stress response uncertainty should be
considered, for example, when an item or device has been calibrated in an external laboratory to account for any
expected increase in the uncertainty reported on the calibration certificate due to stresses incurred during transport back
to the end-user. If a calibration laboratory already includes stress response uncertainty, it should be indicated in the
uncertainty budget (i.e., the itemized list of error sources).
8.4 Combine Uncertainties

Axiom 2 states that the uncertainty in the value of an error is equal to the square root of the variance of the error
distribution. As a consequence, we can apply the variance addition rule to obtain a method for correctly combining
uncertainties from different error sources.
8.4.1 Variance Addition Rule

For purposes of illustration, let us consider a measured quantity x.
13
Drill-down screens and worksheets for estimating measurement uncertainty due to analog-to-digital and/or digital-to-analog signal
conversion are available in UncertaintyAnalyzer.
60
x = xtrue + εx
We know that the total error, εx, consists of measurement process errors
εx = ε1 + ε2 + … + εk
where εk represents the kth error source.
Applying the variance addition rule yields
var(x) = var(εx)
= var(ε1 + ε2 + … + εk) (8-15)
= var(ε1) + var(ε2) + … + var(εk)
+ 2cov(ε1, ε2) + … + 2cov(εl, εk)
where l = k-1 and cov(εl, εk) is the covariance between measurement process errors.
If two variables ε1 and ε2 are described by a joint probability density function f(ε1, ε2), then the covariance of ε1 and ε2 is
given by
∞ ∞
cov(ε1 , ε 2 ) = ∫−∞ d ε1 ∫−∞ ε1ε 2 f (ε1, ε 2 )d ε 2 (8-16)
The covariance is rarely used explicitly. Instead, we use the correlation coefficient, ρkl, which is defined as
cov(ε k , ε l )
ρ kl = (8-17)
uk ul
where uk is the kth measurement process uncertainty.
If we recall Axiom 2, equation (8-15) can be expressed as
m m m
var(ε x ) = ∑ var(ε k ) + 2 ∑ ∑ ρ kl uk ul
k =1 k =1 l > k
(8-18)
m m m
= ∑ uk2 + 2 ∑ ∑ ρ kl uk ul
k =1 k =1 l > k
where m is the number of measurement process errors.
8.4.2 Total Uncertainty and Degrees of Freedom

With the variance addition rule, we have a logical approach for combining uncertainties that accounts for correlations
between error sources. For direct measurements, the total uncertainty, uT, can be generally expressed as
m m m
uT = ∑ uk + 2 ∑ ∑ ρ kl uk ul
2
(8-19)
k =1 k =1 l > k
When uncertainties are combined, we need to know the degrees of freedom for the total uncertainty. Generally speaking,
degrees of freedom signifies the amount information or knowledge that went into an uncertainty estimate. As might be
expected, the degrees of freedom for a combined uncertainty estimate is not a simple sum of the degrees of freedom for
each error source uncertainty.
61
The effective degrees of freedom, νeff, for the total uncertainty, uT, resulting from the combination of uncertainties ui and
associated degrees of freedom, νi, for n error sources can be estimated via the Welch-Satterthwaite formula14 given in
Annex G of the GUM.
uT4 *
ν eff = (8-20)
n ui4
∑
i νi
where
n
uT * = ∑ ui
2
(8-21)
i =1
8.4.3 Correlations Between Measurement Process Errors

To assess the impact of the correlated errors on the combined uncertainty, let us consider the measurement of a quantity
x that involves two error sources ε1 and ε2.
x = xtrue + ε1 + ε2
From Axioms 2 and 3 and the variance addition rule, the uncertainty in x is obtained from
u x = var( xtrue + ε1 + ε 2 ) = var(ε1 + ε 2 )
= u12 + u22 + 2 ρ1,2 u1u2
The correlation coefficient, ρ1,2, for two error sources can range in value from -1 to +1.
If the two error sources are statistically independent, then ρ1,2 = 0 and u x = u12 + u22 . Therefore, uncertainties of
statistically independent error sources are combined in a root-sum-square (RSS).
Conversely, if the two error sources are strongly correlated then ρ1,2 = 1 or -1. If ρ1,2 = 1, then
( u1 + u2 )
2
u x = u12 + u22 + 2u1u2 = = u1 + u2 . Therefore, the uncertainties are combined linearly.
When two error sources are strongly correlated and compensate for one another, then ρ1,2 = -1 and
( u1 − u2 )
2
u x = u12 + u22 − 2u1u2 = = u1 − u2 . Therefore, the combined uncertainty is the absolute value of the
difference between the individual uncertainties.
There typically aren't any correlations between measurement process errors for a given quantity. In general, it is safe to
assume that there are no correlations between the following measurement process errors.
Random Error and Parameter Bias (ρran,bias = 0)

Random Error and Operator Bias (ρran,oper = 0)
Parameter Bias and Resolution Error (ρbias,res = 0)
Parameter Bias and Operator Bias (ρbias,oper = 0)
Operator Bias and Environmental Factors Error (ρoper,env = 0)
Resolution Error and Environmental Factors Error (ρres, env = 0)
Digital Resolution Error and Operator Bias (ρdres,oper = 0)
14
This formula is based on the assumption that there are no correlations between error source uncertainties.
62
8.4.4 Confidence Limits and Expanded Uncertainty
As previously stated, the uncertainty, u, and degrees of freedom, ν, can be used to establish confidence limits. These are
the upper and lower limits that contain the true value, µ (estimated by the mean value x ), with some specified
confidence level or probability, p. Confidence limits are expressed as
x − tα / 2,ν u ≤ µ ≤ x + tα / 2,ν u (8-22)
where the multiplier is the t-statistic, tα/2ν, and α = 1- p.
The GUM defines the term expanded uncertainty as "the quantity defining an interval about the result of a
measurement that may be expected to encompass a large fraction of the distribution of values that could reasonably be
attributed to the measurand." In less obtuse language, the expanded uncertainty is basically defined as a set of limits (±
L) that are expected to contain the true value of the measurand. In this context, the expanded uncertainty, ku, is offered
as an approximate confidence limit, in which the coverage factor, k, is used in place of the t-statistic.
x − ku ≤ true value ≤ x + ku (8-23)
The introduction of the expanded uncertainty is confusing at best, since it is counter-intuitive to think of an uncertainty
as having a range. In actual practice, the term expanded uncertainty and uncertainty are used interchangeably. This, of
course, can lead to incorrect inferences and miscommunications. To mitigate this problem, the GUM also introduced the
term "standard uncertainty" to help distinguish uncertainty from expanded uncertainty. Unfortunately, confusion over
and misapplication of these terms persists.
8.5 Report the Analysis Results

When reporting the results of an uncertainty analysis, Section 7 of the GUM recommends that the following information
be included:
The estimated value of the quantity of interest (measurand) and its combined uncertainty and degrees of freedom.
The functional relationship between the quantity of interest and the measured components, along with the sensitivity
coefficients.
The value of each measurement component and its combined uncertainty and degrees of freedom
A list of the measurement process uncertainties and associated degrees of freedom for each component, along with a
description of how they were estimated.
A list of applicable correlation coefficients, including any cross-correlations between component uncertainties.
It is also a good practice to provide a brief description of the measurement process, including the procedures and
instrumentation used, and additional data, tables and plots that help clarify the analysis results.
63
Chapter 9 SMPC Methodology
SMPC stands for “Statistical Measurement Process Control.” It is a way of estimating the value of a subject parameter
based on measurements made by a measuring parameter. SMPC also provides an estimate of the value of the measuring
parameter, based on the same measurements. In doing so, SMPC formally recognizes that measuring parameters are not
perfect.
Uncertainty analysis in general acknowledges this fact by attempting to estimate the uncertainty associated with
measurements. A typical result of an uncertainty analysis is a statement of a measured value accompanied by an
estimate of the uncertainty in the value. While the uncertainty estimate stands as an admission that uncertainty exists in
a measurement, the measured value or a statistical mean of this value is usually taken at face value.
This apparent contradiction is especially interesting when the subject parameter is a toleranced quantity, as in the case of
calibrating and testing. To see this, imagine that we arbitrarily measure a very accurate subject parameter (i.e., one with
tight tolerances and a high in-tolerance probability) with a moderately accurate measuring parameter.
Obviously, no reasonable person would presume that the measured value or the mean of a sample of measured values
provides a good indication of the bias in the subject parameter. Instead, one would be inclined to turn things around and
use the measurement result as an indication of the bias in the measuring parameter. If we did this, we would be taking a
step toward understanding SMPC.
9.1 Basic Principles

The fundamental principle that is central to SMPC analysis states that: In measurement situations where we have a priori
knowledge of measuring parameter and subject parameter accuracies, the roles of measuring parameter and subject
parameter are reversible. This is called the Principle of Measurement Symmetry.
A prior knowledge is the information we have about the measuring and subject parameter before any measurements are
made. A posteriori estimates are estimates computed after taking measurements. As one would expect, SMPC analysis
is sensitive to all measurement error sources, not just parameter bias. Therefore, SMPC analysis is most effective when
all measurement errors and their associated uncertainties are also estimated. By including measurement uncertainty, you
are providing additional information from which to revise or refine the SMPC estimates.
9.2 Equal Accuracies

If we were to measure the value of a subject parameter with a measuring parameter of equal accuracy, we could easily
use the result of the measurement to estimate the “true” value of what was measured. For instance, suppose that the
measuring parameter is a spring-operated scale with tolerance limits of ± 0.5 gm and an in-tolerance probability of 90%.
Suppose also that the subject parameter is a calibrated weight with a nominal value of 5 gm, tolerance limits of ± 0.5 gm
and in-tolerance probability of 90%. Finally, suppose that the result of weighing the subject parameter (the value of the
weight) with the measuring parameter (the scale reading) is 4.6 gm.
While the scale says the weight is 0.4 gm low, the weight also says the scale is reading 0.4 low. We estimate the most
likely value for the weight by taking an average of what the weight says it weighs and what the scale says it weighs.
5.0 gm + 4.6 gm
Estimated True Weight =
2
= 4.8 gm
We can state that the most likely value of the weight is really 4.8 gm. Notice that by looking at the problem the way we
did, we have not only estimated a “best” value for the weight, but also arrived at a pair of correction factors. We can
apply a correction of + 0.2 gm to scale readings at 5 gm and a correction of – 0.2 gm to the value of the 5 gm weight.
9.3 Unequal Accuracies

Obviously, it doesn’t take a rocket scientist to deal with situations where measuring and subject parameters are of equal
accuracy. But what do we do when their accuracies are different? What if the weight in the above example had a
nominal value of 5.0 gm with ± 1.0 gm tolerance limits and 90% in-tolerance probability, or had ± 0.5 gm tolerance
limits with 85% in-tolerance probability?
64
Either of these conditions would lead us to believe that the scale is now more accurate than the weight. But how much
more accurate? Although we know that the scale is “more correct,” we also know that it is not perfect. What’s more,
now that we have gone through an exercise with equal accuracies, we are too sophisticated to simply regard the scale’s
reading as an estimate of the true value of the weight. We know that, the scale is providing information about the
weight, the weight is also providing information about the scale.
9.4 Single Sample Case (Example 1 – DC Voltmeter Calibration)

In Example 1, we estimated the measurement process uncertainties associated with the calibration of the DC voltage
function of a digital multimeter using a DC voltage reference. We had the following pre-analysis information (or a prior
knowledge) for the subject parameter (DC voltmeter) and the measuring parameter (voltage reference):
Tolerance In-Tolerance
Limit Probability
DC Voltmeter (subject parameter) ± 10 mV 95.00 %
Voltage Reference (measuring parameter) ± 0.07 mV 99.00 %
When we enter the repeat measurements for the DC voltmeter and estimated uncertainties for the other measurement
process errors, SMPC analysis gives the follow results:

Estimated Bias -0.0117 mV 0.3333 mV 9.9995 V
Uncertainty 0.0343 mV 0.9401 mV 0.9401 mV
Recall that the expected output from the voltage reference at the time of measurement was computed to be 9.9998 V. The
mean value for the 15 repeat voltage measurements was computed to be 10.0001 V with a standard deviation or
uncertainty 0f 0.93 mV. According to these measurements, the DC voltmeter is reading 0.3 mV high (SMPC says
0.3333 mV high).
The estimated true value is simply the nominal output from the voltage reference minus the subject parameter bias,
rounded to four significant digits. SMPC computes an 0.9401 mV uncertainty in this value that is slightly less than the
total uncertainty of 0.9456 mV. This refinement in the measurement uncertainty occurs because SMPC looks at the
measurement process in two ways, where the roles of the subject parameter and the measuring parameter are reversible.
An a priori bias uncertainty estimate for the DC voltmeter was 5.102 mV, based on the ± 10 mV tolerance limits and
95.00 % in-tolerance probability. After our uncertainty/SMPC analysis the bias offset of 0.3333 mV was determined
with an estimated uncertainty of 0.9401 mV. In comparison, it is easy to see why there is a 100% probability that the
DC voltmeter was operating within manufacturer tolerance limits at the time of our measurements.
It is also interesting to note that, give the SMPC estimated bias and uncertainty, there is only a 5.8% probability that the
voltage reference was within the manufacturer tolerance limits at the time of measurement. Fortunately, this was not a
serious concern, given the fact that the voltage reference was far more accurate than the DC voltmeter to begin with.
9.5 Two Sample Case (Example 3 – Thermometer Calibration)

In Example 3, we estimated the measurement process uncertainties for the calibration of a digital thermometer with an
SPRT temperature reference and an oil bath. We had the following pre-analysis information (or a prior knowledge) for
the subject parameter (digital thermometer) and the measuring parameter (SPRT temperature reference):
Tolerance Limit In-Tolerance

or Exp. Unc. Probability
Digital Thermometer ± 0.07 deg C 95.00 %
SPRT Temperature Reference ± 0.004 deg C 95.45 %
When we entered the repeat measurements for the thermometer and the SPRT, and estimated uncertainties for the other
measurement process errors, SMPC analysis gives the following results:
65
Estimated Bias 0.0001 deg C -0.0050 deg C 100.00 deg C
Uncertainty 0.0001 deg C 0.0056 deg C 0.0056 deg C
Recall that the nominal bath temperature at the time of measurement was 100.00 deg C. The mean value for the 10 repeat
thermometer measurements was computed to be 99.996 deg C with a standard deviation or uncertainty 0f 0.00521 deg C.
According to these measurements, the digital thermometer is reading 0.004 deg C low. However, the mean value for the
corresponding SPRT measurements was computed to be 100.001 deg C with an uncertainty of 0.00127 deg C.
Given that the SPRT is much more accurate than the digital thermometer, SMPC estimates the true value to be very close
to the 100.00 deg C nominal temperature. The estimated true value is rounded out to 2 significant digits so we cannot
see that it is slightly higher. Consequently, the estimated bias for the digital thermometer is refined to - 0.005 deg C.
SMPC also refined the uncertainty in the estimated true value to 0.0056 deg C, which is lower than the 0.00767 deg C
total uncertainty estimated without SMPC.
Although SMPC estimated a bias in the SPRT measurements of only 0.0001 deg C, the associated uncertainty is of equal
value. Therefore, there is only a 52% probability that the SPRT temperature reference was in-tolerance at the time we
made our measurements.
9.6 Cautions and Conditions

As with any analysis technique, the results are only as good as the information or data input into it. Or as the saying
goes: “garbage in, garbage out.” To avoid misusing SMPC, there are two provisions to keep in mind.
1. At its present stage of development, SMPC is strictly applicable when process errors and measurement biases are
normally distributed. For process errors or parameter bias with a containment probability (confidence level) less
than 100%, Uncertainty Sidekick assumes a normal distribution.
2. In using SMPC with more than one measurement sample, the samples must be based on measurements of the same
quantity. SMPC needs to “anchor” measurements to a common reference for comparison and evaluation.
If you are interested in the mathematics of SMPC, you are encouraged to read Appendix D of NASA Reference
Publication 1342, Metrology – Calibration and Measurement Process Guidelines.
66
Nomenclature and Terminology
Uncertainty Sidekick employs terms and definitions that are designed to be understood across a broad technology base.
Where appropriate, terms and definitions have been taken from internationally recognized standards and guidelines in the
fields of testing and calibration.
In some cases, these terms have been found to be somewhat misleading or confusing to many engaged in uncertainty
analysis. In these cases, Uncertainty Sidekick employs generalized terms and definitions. In all cases, we have
endeavored to ensure that terminology is used consistently throughout the program and associated documentation.
Name Definition
% Confidence See Confidence Level
% of Full Scale The contribution of a parameter's tolerance limit equal to a stated

percentage of the parameter's full scale.
% of Nominal The contribution of a parameter's tolerance limit equal to a stated

percentage of the parameter's nominal value.
% of Range The contribution of a parameter's tolerance limit equal to a stated

percentage of the parameter's range.
a posteriori value The value calculated after taking measurements.
a priori value The value indicated prior to taking measurements.
Accuracy In terms of instruments and scientific measuring systems, accuracy is

defined as the conformity of an indicated value to an accepted standard
value, or true value. (Taken from Van Nostrand's Scientific Encyclopedia,
Sixth Edition).
Active Reference Model A measurement configuration in which the measuring parameter measures
the value of the subject parameter directly.
Adjusted Mean The value of a parameter or error source obtained by applying a correction
factor to a nominal or mean value.
Analysis Type The method used to obtain estimates of uncertainty. Type A estimates
involve the analysis of data samples; whereas Type B estimates use
heuristic means.
Ancillary Parameters Parameters, other than the subject and measuring parameters, that
participate in a measurement or indirectly influence the subject parameter or
measuring parameter bias. An example of an ancillary parameter would be
an environmental factor such as ambient temperature that is measured and
used to apply an environmental correction to the subject or measuring
parameter values. Sometimes referred to as influence quantities.
Arithmetic Mean The sum of a set of values divided by the number of values in the set.
Artifact A measurable physical object or substance.
Asymmetric Distribution A statistical distribution of a quantity in which deviations in value of one

sign are more probable than deviations of the opposite sign. Examples of
asymmetric distributions are the lognormal and exponential distributions.
67
Asymmetric Tolerances Two-sided tolerance limits in which the magnitudes of the upper and lower
limits are unequal.
Attribute A characteristic, feature, or aspect of an object or substance.
Average See Arithmetic Mean.
Bias A systematic discrepancy between an indicated or declared value of an

artifact and its true value.
Bias Distribution See Error Distribution.
Bias Offset See Offset from Nominal.
Bias Uncertainty The uncertainty in the bias of a parameter or artifact.
Combined Uncertainty The uncertainty in the total error of a value of interest.
Compensating Biases Measuring parameter biases that offset or compensate one another. For
example, if the same measuring parameter is used to measure the inside
diameter of a sleeve and the outside diameter of a shaft that fits into the
sleeve, any error or bias in the two measurements will not affect the quality
of fit.
Component Error See error component.
Component Uncertainty The standard uncertainty for a component error.
Computation Error The error in a quantity obtained by computation. Normally due to machine
round-off error in values obtained by iteration. Sometimes applied to errors
in tabulated physical constants.
Confidence Level The probability that a set of error limits or containment limits will contain
errors for a given error source.
Confidence Limits Limits that bound errors for a given error source with a specified probability
or "confidence."
Containment Limits Limits that are specified to contain either a parameter value, deviations from
the nominal parameter value, or errors in the measurement of the parameter
value.
Containment Probability The probability that a parameter value or errors in the measurement of this
value will lie within specified containment limits.
Control Number A unique number identifying an item throughout its life cycle. Once
assigned, it should never be reassigned to another item even if the item to
which it was originally assigned is deleted from inventory.
Correlation Analysis An analysis that determines the extent to which two error sources influence
one another. Typically the analysis is based on ordered pairs of values of
the two error source variables.
Correlation Coefficient A measure of the extent to which two error sources are linearly related. A
function of the covariance between the two error sources. Correlation
coefficients range from minus one to plus one. A positive correlation
coefficient applies when increases in one source are accompanied by
increases in the other. A negative correlation coefficient applies when
68
increases in one source are accompanied by decreases in the other.
Covariance The covariance of two error sources is a measure of their statistical

dependence.
Coverage Factor A multiplier used to express an error limit or expanded uncertainty as a

multiple of the standard uncertainty.
Degrees of Freedom A statistical quantity that is related to the amount of information available
about an uncertainty estimate. The degrees of freedom signifies how
"good" the estimate is and serves as a useful statistic in determining
appropriate coverage factors and computing confidence limits and other
decision variables.
Deviation from Nominal The difference between a parameter's measured value and its nominal value.
Direct Measurements Measurements in which a measuring parameter X directly measures the

value of a subject parameter Y (i.e., X measures Y). In direct
measurements, the value of the subject parameter is obtained directly by
measurement and is not determined by computing its value from the
measurement of other variables or quantities.
Distribution Function See Cumulative Distribution Function.
Distribution Variance The mean square dispersion of a distribution about its mean or mode value.
See also Variance.
Effective Degrees of Freedom The degrees of freedom for combined uncertainties computed from the
Welch-Satterthwaite formula.
Error Distribution A statistical distribution that describes the relative frequency of occurrence
of values of a measurement error.
Error Component A component of error comprised of parameter bias, random error, operator
bias, resolution error and other measurement process errors.
Error Limits Bounding values that are expected to contain the error from a given source
with some specified level of probability or confidence.
Error Source A parameter, variable or constant that can contribute error to the
determination of the value of a subject parameter. Examples include:
measuring parameter bias, random error, resolution error, operator bias,
computation error and environmental factors error.
Error Source Correlation See Correlation Analysis
Error Source Uncertainty The uncertainty in the error of a given source.
Estimated True Value The value of a quantity obtained by SMPC (Bayesian) analysis.
Expanded Uncertainty A multiple of the standard uncertainty reflecting either a specified

confidence level or arbitrary coverage factor.
Fixed Tolerance Limit A tolerance limit for a parameter that does not depend on the nominal or
other values associated with the parameter.
Full Scale An identifying reference value for readings or outputs for a given
parameter. Often the highest value in a range.
69
Gaussian Distribution The normal distribution.
General Model A measurement configuration other than the Active Reference, Passive
Reference, or Indirect Model.
Heuristic Estimate An estimate resulting from accumulated experience and/or technical
knowledge concerning the uncertainty of an error source.
Hysteresis The resistance of a variable to a change in stimulus.
Hysteresis Error The residual signal in a sampling event left over from the previous sampling
event.
Independent Error Sources Error sources that are statistically independent. Two error sources are
statistically independent if one does not exert an influence over the other or
if both are not consistently influenced by a common agency. See also
Statistical Independence.
Indirect Model A measurement configuration in which both the Subject Parameter and
Measuring Parameter measure a common artifact. An example would be a
scale (subject parameter) and precision balance (measuring parameter) both
used to measure a reference mass.
Interaction Coefficient A number that converts the value of an environment or ancillary parameter
to a change exerted by the parameter on the bias of a subject parameter or
measuring parameter. In error analysis, the interaction coefficient is the
partial derivative of the measured variable with respect to the environment
parameter.
In-tolerance Probability The probability that a parameter value or the error in the value is contained
within its specified tolerance limits at the time of measurement.
Item A device or artifact with measurable attributes or parameters.
Level of Confidence See Confidence Level.
Lower Range Limit In a Parameter specification, the lower limit of a designated range of values
for which specified tolerances apply.
Lower Tolerance Limit The lower bounding limit that identifies acceptable versus unacceptable
parameter values.
Mean Deviation The difference between a sample mean value and the nominal value.
Mean Square Error See Variance.
Mean Value Sample Mean: The average value of a measurement sample. Population
Mean: The expectation value for measurements sampled from a population.
Measurand This is a term that has more than one interpretation. In section B.2.9 of
Annex B in the GUM, it is defined as "the particular quantity subject to
measurement." Another interpretation includes this definition but also
includes the specific conditions under which the measurement is made.
These conditions may include the identification of environmental factors,
ancillary equipment used, measuring personnel, etc.
To avoid confusion, the term "subject parameter" has been coined to

specifically represent the particular quantity subject to measurement.
70
Measurement Area A measurement discipline, such as DC Voltage, that applies to a subject or
measuring parameter.
Measurement Error The difference between the measured value of a parameter and its true
value.
Measurement Process Errors Measurement process errors refer to errors resulting from the measurement
process (e.g., measuring parameter bias, operator bias, environmental
factors, …). Measurement process errors are the basic elements of
uncertainty analysis. Once these fundamental error sources have been
identified, then we can begin to develop uncertainty estimates.
Measurement Process Uncertainties See Error Source Uncertainty
Measurement Uncertainty The uncertainty in a measurement or in the error in the measurement.
Measurement Units The units, such as volts, millivolts, etc., in which a measurement or
measurement error is expressed. Measurement units are categorized by
measurement area.
Measuring and Test Equipment The device or artifact featuring the Measuring Parameter.
Measuring Parameter Attribute of a measuring device that is used to obtain information that
quantifies the value of the subject parameter.
Mixed Estimates Uncertainty estimates that consist of Type A uncertainties combined with
Type B uncertainties.
MTE Measuring and Test Equipment
Nominal Units Measurement units in which the nominal value of a parameter is expressed
or specified.
Nominal Value The designated or published value of an artifact or parameter. It may also
sometimes refer to the mode value of an artifact or parameter.
Normal Distribution A symmetric, continuous, unbounded statistical distribution with a central

tendency that represents most random statistical processes. Also referred to
as the Gaussian distribution. Characterized by a mean value and a standard
deviation.
Offset from Nominal The stated offset from nominal of a calibrated artifact.
Operator Bias Error due to the perception or influence of a human operator or other
agency.
Passive Reference Model A measurement configuration in which the value of the subject parameter is
obtained by taking a sample of measurements of the measuring parameter
by the subject parameter. An example would be the calibration of a
precision balance (subject parameter) using a reference mass (measuring
parameter). In this case, it is the subject parameter that is used to collect
measured values of the measuring parameter.
Parameter Often thought of as a specified aspect or feature of an instrument or item.

In general, however, a parameter does not have to be a toleranced quantity
or value. See also Attribute.
71
Parameter Bias A systematic deviation of a parameter value from its nominal or indicated
value.
Pareto Diagram A bar chart that ranks the relative contribution of component uncertainties
to the total combined uncertainty.
PDF Probability Density Function
Percent In-Tolerance The probability, in percent, that a parameter value or the error in the value is
contained within its specified tolerance limits at the time of measurement.
Physical Limit A limiting upper or lower value that bounds values attainable by a given
equipment parameter or attribute.
Population The total set of possible values for a random variable under consideration.
Population Mean It is the expectation value of a random variable described by a population

distribution.
Precision Precision corresponds to how many places past the decimal point we can
express a measurement result. Although higher precision does not
necessarily mean higher accuracy, the lack of precision in a measurement is
a source of measurement error.
Probability The likelihood of the occurrence of a specific event or value from a

population of events or values.
Probability Density Function A mathematical function that describes the relative frequency of occurrence
of the values of a random variable.
Random Error An error that appears as differences in the measured values of a given
artifact or parameter during a measurement session. Sometimes due to
random fluctuations in the subject parameter value, the measuring
parameter value, and the measurement process.
Range In a parameter specification, a designated range of values for which

specified tolerances apply. In a calibration or test procedure, a setting or
designation for a set of measurements.
Reference Standard An artifact used as a measurement reference whose value and uncertainty
have been determined by calibration.
Repeatability The closeness of the agreement between the results of successive

measurements of the value of a parameter carried out under the same
measurement conditions. Repeatability conditions include: the same
measurement procedure, the same observer, the same measuring instrument
used under the same conditions, the same location, repetition over a short
period of time.
Reproducibility The closeness of the agreement between the results of successive

measurements of the value of a parameter carried out under changed
measurement conditions. The changed conditions may include: principle of
measurement, method of measurement, observer, measuring instrument(s),
reference standard, location, conditions of use, time.
Resolution The smallest discernible value indicated by a measuring parameter or

subject parameter.
72
Resolution Error The error due to the finiteness of the precision of a measurement.
RSS (1) Root-sum-square method of combining values. (2) Residual sum of

squares in regression analysis.
Sample A collection of values drawn from a population. Typically, inferences

about a population are made from the sample. Therefore, the sample must
be statistically representative of the population.
Sample Mean The arithmetic average of the measurements of a sample.
Sample Size The number of measured values that comprise a sample.
Single-sided Lower Tolerance Limit A single-sided tolerance limit in which only the lower limit is specified.
Single-sided Upper Tolerance Limit A single-sided tolerance limit in which only the upper limit is specified.
Standard Deviation The square root of the variance of a sample or population of values. A
quantity that represents the spread of values about a mean value. In
statistics, the second moment of a distribution.
Standard Uncertainty A statistic representing spread or uncertainty in the value of a parameter or

error source. If determined statistically from sampled data, the standard
uncertainty is equal to the sample standard deviation.
Statistical Independence A property that describes two error sources as being uncorrelated. See also
Independent Error Sources.
Student's t Distribution A symmetric, continuous distribution characterized by the degrees of

freedom parameter. Used to compute confidence limits for normally
distributed variables whose standard deviation has is based on a finite
degrees of freedom.
SU Subject Unit. The device featuring the subject parameter. Sometimes

called the "unit under test" or UUT.
Subject Parameter An attribute whose value we seek to obtain from a measurement or set of
measurements.
Subject Unit A device or artifact featuring the Subject Parameter.
Symmetric Distribution A statistical distribution of random variables that are equally likely to be
found above or below a mean value. Examples of symmetric distributions
include the normal, cosine, uniform and triangular distributions.
Symmetric Tolerances Two-sided tolerance limits in which the magnitudes of the upper and lower
limits are equal.
Tolerance Limits Limits that bound acceptable parameter values.
Tolerance Units Units for specifying the tolerance limits for a parameter.
Toleranced Parameter A parameter with a set of specified tolerances or limits that bound its value
or the error in this value.
True Value The value that would be obtained by a perfect measurement. True values
are by nature indeterminate.
73
Two-sided Tolerance Limits Upper and lower limits that bound acceptable parameter values. Stated
upper and lower limits that contain biases in parameter value with some
specified level of confidence. Two-sided tolerance limits can be symmetric
or asymmetric.
Type A Estimates Uncertainty estimates obtained by the statistical analysis of a sample of

data.
Type B Estimates Uncertainty estimates obtained by heuristic means.
Uncertainty See Standard Uncertainty.
Uncertainty Component A contribution to total combined uncertainty from an error component.
Uncertainty Units The units in which the uncertainty in each error source is expressed.
Generally, uncertainty units are the same as the tolerance units.
Uniform Distribution A bounded, rectangular-shaped statistical distribution of random variables

in which the probability of obtaining a value within the distribution limits is
uniform (i.e., equally probable).
Upper Range Limit In a parameter specification, the upper limit of a range of values for which
specified tolerances apply.
Upper Tolerance Limit The upper bounding limit that bounds acceptable parameter values.
UUT Unit Under Test. See SU.
Other Error User specified error source other than the measurement process error
sources incorporated into Uncertainty Sidekick.
Welch-Satterthwaite Formula The equation for computing the degrees of freedom for combined
uncertainty estimates as described in the GUM.
74
Suggested Reading
Castrup, H., "Analytical Metrology SPC Methods for ATE Implementation," Proceedings of the NCSLI Workshop &
Symposium, Albuquerque, August 1991.
Castrup, H., "Practical Methods for Analysis of Uncertainty Propagation," Proc. 38th Annual ISA Instrumentation
Symposium, Las Vegas, NV, April 1992.
Castrup, H., "Analyzing Uncertainty for Risk Management," presented at the 49th ASQC Annual Quality Congress, May
1995.
Castrup, H., "Uncertainty Analysis and Parameter Tolerancing," Proc. NCSL Workshop & Symposium, Dallas, TX, July
1995.
ANSI/NCSL Z540-2-1997, U.S. Guide to the Expression of Uncertainty in Measurement, NCSL, Boulder, 1997.
Castrup, H., “Estimating Category B Degrees of Freedom,” Proc. Measurement Science Conference, Anaheim, January
2000.
Spiegel, M. R. et al., Probability and Statistics, Second Edition, Schaum's Outline Series, McGraw-Hill, 2000.
Castrup, H., "Distributions for Uncertainty Analysis," Proc. International Dimensional Workshop, Knoxville, May 2001.
Castrup, H., "Estimating Bias Uncertainty," Proceedings of the NCSLI Workshop & Symposium, Washington D.C.,
July 2001.
Castrup, H., "Selecting and Applying Error Distributions in Uncertainty Analysis," presented at the 2004 Measurement
Science Conference, January 16-17, 2003 in Anaheim, CA.
Castrup, S., "A Comprehensive Comparison of Uncertainty Analysis Tools," Proc. Measurement Science Conference,
Anaheim, January 2004.
Castrup, H., "Estimating and Combining Uncertainties," Presented at the 8th Annual ITEA Instrumentation Workshop,
Lancaster, CA, May 2004.
Castrup, S., "Why Spreadsheets are Inadequate for Uncertainty Analysis," Proc. 8th Annual ITEA Instrumentation
Workshop, Lancaster, May 2004.
75
Subject Index
Measurement Units Database. See
Access to External Applications, 6 Measuring Environment Uncertainty Worksheet, 32, 47
Analysis Notes, 5 measuring parameter, v
Analysis Path Measuring Parameter Bias Uncertainty Worksheet, 17,
Describe the Measurement Reference, 8 30, 45
Describe the Measuring Environment, 8 Measuring Parameter Resolution Error Worksheet, 46
Describe What is Being Measured, 8 Measuring Parameter Uncertainty Worksheet, 18, 31,
Describe Who is Making the Measurement, 8 46
Setup the Analysis, 8
Analysis Path, 4, 8 Nomenclature and Terminology, 66
Analysis Procedures. See Analysis Path
On-Screen Help, 6
Bayesian Analysis. See SMPC Methodology Operator bias uncertainty, 59
Bias uncertainty, 58 Operator Bias Uncertainty Worksheet, 34
Operator Uncertainty Worksheet, 35
Computation uncertainty, 59
confidence limits, 61 Pareto Chart Screen, 22
containment limits, 57 Pareto Diagram. See Pareto Chart Screen
containment probability, 57 Preferred Nomenclature, vi, 37
correlated errors, 61 Print Preview Screen, 52
Correlation Coefficient, 60 Printing Analysis Results, 51
Correlation Coefficients Worksheet, 33 Previewing and Printing Reports, 52
Report Options Screen, 51
Selecting Report Fonts Screen, 51
Definition Path, 4 Setting Reports Nomenclature, 51
Degrees of freedom, 58
Describe the Measurement Reference. See Measuring
Parameter Uncertainty Worksheet Random uncertainty, 59
Digital sampling uncertainty, 59 Report Options Screen, 21, 36, 49
direct measurement, v Reporting Analysis Results. See Report Options Screen
Resolution uncertainty, 59
Effective degrees of freedom, 60
Environmental factors uncertainty, 59 Sample Mean, 56
Error limits, 9 Sample standard deviation, 56
Example 1 – DC Voltmeter Calibration, 11 Saving and Opening Files. See File Menu
Example 2 – Micrometer Calibration, 24 significant digits, v
Example 3 – Thermometer Calibration, 39 SMPC Methodology, 63
Expanded uncertainty, 10, 62 Basic Principles, 63
Cautions and Conditions, 65
Equal Accuracies, 63
Help file. See On-Screen Help Single Sample Case, 64
Two Sample Case, 64
Installation and Startup, 1 Unequal Accuracies, 63
interaction coefficient, 59 Standard uncertainty, 9
Stress response uncertainty, 59
Main Screen, 7 Subject parameter, v
Analysis Details, 9 Subject Parameter Bias Uncertainty Worksheet, 27, 41
Analysis Path, 8 Subject Parameter Resolution Error Worksheet, 16, 28,
Analysis Results, 10 43
Combined Error Distribution, 10 Subject Parameter Uncertainty Worksheet, 15, 28, 43
Error Component Table, 9
Menu, 7 Terminology, vi, 66
Parameter Bias Estimates, 10 Total Uncertainty, 10, 60
Measurement Configuration Screen, 13, 26, 41 t-statistic, 57
Measurement Data Entry Worksheet, 15, 28, 42, 45 Type B Degrees of Freedom Calculator, 6
Measurement Process Overview, 11, 24, 39
76
Uncertainty. See standard uncertainty Type A Estimates, 9, 56
Uncertainty Analysis Basics, 53 Type B Estimates, 9, 57
Combine Uncertainties, 59 Uncertainty Analysis Procedure, 12, 25, 40
Define the Measurement Process, 53 Uncertainty in the Mean Value, 57
Error model, 53 Update Measurement Units Screen.
Rstimate uncertainties, 55
Measurement process errors, 53 Variance Addition Rule, 59
Measurement process uncertainties, 58
Report the Analysis Results, 62 Welch-Satterthwaite formula, 10
Statistical distributions, 54
77

Unc Sidekick User Manual

Uploaded by

Document Information

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

Unc Sidekick User Manual

Uploaded by

Copyright:

Available Formats

User Manual

Integrated Sciences Group

Chapter 1 Installation and Startup 1

Chapter 2 Overview of Features 4

Chapter 3 The Main Screen 7

Chapter 4 Example 1 – DC Voltmeter Calibration 11

Chapter 5 Example 2 – Micrometer Calibration 24

Chapter 6 Example 3 – Thermometer Calibration 39

Chapter 8 Uncertainty Analysis Basics 54

Chapter 9 SMPC Methodology 64

Nomenclature and Terminology 67

Things to Keep in Mind

What You Need to Do

1. Define the Measurement Process

How to Use this Manual

Note on Reporting Results

To continue the installation, click the Next button. To exit the

The User Confirmation screen will be displayed next. If the

Clicking the Next button, closes the Destination Location screen

If you want to backup files that the installation may replace on

Clicking the Next button closes the Select Program Manager

Once the configuration update is completed, a screen will appear

On occasion, the setup program detects whether you need to restart

1.2 Video and Color Recommendations

Video Resolution 1024 x 768 or better with Large Fonts

1.3 System Requirements

1.5 Uncertainty Sidekick Files

Uncertainty Sidekick.exe - The program executable file.

Example 1 - DC Voltmeter Calibration.sdk

Sidekick.hlp - The Uncertainty Sidekick Help file.

Uncertainty Analysis Overview Folder

User Manual Folder

1.6 Program Startup

2.1 Step-by-step Analysis Procedures

2.2 Analysis Screens

2.3 Automatic Uncertainty Combination

2.4 Bayesian Analysis (SMPC)

2.5 Charts and Plots

2.6 Analysis Notes

2.7 Saving and Opening Files

2.8 Reporting Analysis Results

2.9 Measurement Units Database

You can add new measurement areas and units to the

2. Containment limits and history of observations.

3. Containment limits, probability and number of cases.

2.11 Access to External Applications

External programs can be added to the Applications

2.12 On-Screen Help

The contents of the Help file can be accessed from all

Clicking Help on a worksheet or screen menu

3.1 Main Screen Menu

3.2 The Analysis Path

3.3 Analysis Details

3.4 Error Component Table

3.6 Analysis Results

3.7 Combined Error Distribution

The Plot Options menu allows you to display 2-sigma limits

3.8 Parameter Bias Estimates

4.1 Measurement Process Overview

Table 4.1 Repeat Measurements from 973A DMM Calibration

Specifications for Fluke 732B DC Voltage Reference

Specifications for Agilent 973A Digital Multimeter

4.2 Uncertainty Analysis Procedure

Calibration: Philosophy in Practice, 2nd Edition, Copyright © 1994 Fluke Corporation.