Use

R!

Series Editors
Robert Gentleman
Div. Public Health Sciences, San Mateo, California, USA

Kurt Hornik
Accounting & Statistics, WU Wien, Dept of Finance, Wien, Austria

Giovanni Parmigiani
Dana Farber Cancer Institute, Boston, USA

More information about this series at http://www.springer.com/series/6991

Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba

Quality Control with R

An ISO Standards Approach
1st ed. 2015
Emilio L. Cano
Department of Computer Science and Statistics, Rey Juan Carlos University,
Madrid, Spain
Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain

Javier M. Moguerza
Department of Computer Science and Statistics, Rey Juan Carlos University,
Madrid, Spain

Mariano Prieto Corcoba

ENUSA Industrias Avanzadas, Madrid, Spain

ISSN 2197-5736 e-ISSN 2197-5744

ISBN 978-3-319-24044-2 e-ISBN 978-3-319-24046-6

DOI 10.1007/978-3-319-24046-6

Springer Cham Heidelberg New York Dordrecht London

Library of Congress Control Number: 2015952314

© Springer International Publishing Switzerland 2015

Use R!

This work is subject to copyright. All rights are reserved by the Publisher,
whether the whole or part of the material is concerned, specifically the rights of
translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or
by similar or dissimilar methodology now known or hereafter developed.

The use of general descriptive names, registered names, trademarks, service

marks, etc. in this publication does not imply, even in the absence of a specific
statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.

The publisher, the authors and the editors are safe to assume that the advice and
information in this book are believed to be true and accurate at the date of
publication. Neither the publisher nor the authors or the editors give a warranty,
express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer

Science+Business Media (www.springer.com)
To our families.
Foreword
Although it started almost two decades ago as a purely academic project, the R
software has established itself as the leading language for statistical data analysis
in many areas. The New York Times highlighted, in a 2009 article, this transition
and pointed out how important companies, such as IBM, Google, and Pfizer,
have embraced R for many of their data analysis tasks.
It is known that R is becoming ubiquitous in many other commercial areas,
well beyond IT and big pharma companies. This is well described in this book,
which focuses on many of the tools available for quality control (QC) in R and
how they can be of use to the applied statistician working in an industrial
environment. All products that we consume nowadays go through a strict quality
protocol that requires a tight integration with data obtained from the production
line.
The authors have put together a manual that makes Springer’s use R! series
become even more comprehensive as this topic has not been covered before. QC
is an important field because it requires a specific set of statistical methodology
that is often neglected in these times of the Big Data revolution. This volume
could well serve as an accompanying textbook for a course on QC at different
levels, as it provides a description of the main methods in QC and then illustrates
their use by means of examples on real data sets with R.
But this book is not only about teaching QC. In fact, the authors combine an
outstanding academic background with extensive expertise in the industry,
including professional in-company training and an active involvement with the
Spanish Association for Quality (AEC) and with the Spanish Association for
Standardization (AENOR, member of ISO). Thus, the book will also be of use to
researchers on QC and engineers who are willing to take R as their primary
programming language. What makes QC different is that it is at the core of
production and manufacturing. In this context, R provides a suitable
environment for data analysis directly at the production lines. R has evolved in a
way that it can be integrated with other software and tools to provide solutions
and analysis as data (and goods) flow in the lines.
Furthermore, the authors have reviewed ISO standards on QC and how they
have been implemented in R. This is important because it has serious
implications in practice as production is often constrained to fulfill certain ISO
standards. For this reason, I believe that this book will play an important role to
take R even further into the industrial sector.
Finally, I congratulate the authors for continuing the work that they started in
their book on Six Sigma with R. These two books could well be used together
not only to control for the quality of the products but also to improve the quality
of the industrial production processes themselves. With R!
Virgilio Gómez-Rubio
Albacete, Spain
July 2015
Preface
Why Quality Control with R?
Statistical quality control is a time-honored methodology extensively
implemented in companies and organizations all over the world. This
methodology allows to monitor processes so as to detect change and anticipate
emerging problems. Moreover, it needs statistical methods as the building blocks
of a successful quality control planning.
On the other hand, R is a software system that includes a programming
language widely used in academic and research departments. It is currently
becoming a real alternative within corporate environments. With R being a
statistical software and a programming language at the same time, it provides a
level of flexibility that allows to customize the statistical tools up to the
sophistication that every company needs. At the same time, the software is
designed to work with easy-to-use expressions, whose complexity can be scaled
by users as they advance in learning.
Finally, the authors wanted to provide the book with a new flavor, including
the ISO Standards Approach in the subtitle. Standards are crucial in quality and
are becoming more and more important also in academia. Moreover, statistical
methods’ standards are usually less known by practitioners, who will find in this
book a nice starting point to get familiar with them.
Who Is This Book For?
This book is not intended as a very advanced or technical reading. It is aimed at
covering the interest of a wide range of readers, providing something interesting
to everybody. To achieve this objective, we have tried to write the least possible
mathematical equations and formulas. When necessary, we have used formulas
followed by simple numerical examples in order to make them understandable.
The examples clarify the tools explained, using simple language and trying to
transmit the principal ideas of quality control.
As far as the software is concerned, we have not used complicated
programming structures. Most examples follow the structure
function(arguments)  →  results . In this regard, the book is self-contained as it
comprises all the necessary background. Nevertheless, we reference all the
packages used and encourage the reader to consult their documentation.
Furthermore, references both to generic and specific R books are also provided.
Quality control practitioners without previous experience in R will find
useful the chapter with an introduction to the R system and the cheat sheet in the
Appendix. Once the user has grasped the logic of the software, the results are
increasingly satisfactory. For quality control beginners, the introductory chapter
is an easy way to start through the comprehensive intuitive example.
Statistical software users and programmers working in organizations using
quality control and related methodologies will find in this book a useful
alternative way of doing things. Similarly, analysts and advisers of consulting
firms will get new approaches for their businesses beyond the commercial
software approach.
Statistics teachers have in a single book the essentials of both disciplines
(quality control and R). Thus, the book can be used as a textbook or reference
book for intermediate courses in engineering statistics, quality control, or related
topics.
Finally, business managers who want to understand and get the background
to encourage their teams to improve their business through quality control can
read selected chapters or sections of the book, focusing on the examples.

How to Read This Book

In this book, we present the main tools and methodologies used for quality
control and how to implement them using R . Even though a sequential reading
would help in understanding the whole thing, the chapters are written to be self-
contained and to be read in any order. Thus, the reader might find parts of the
contents repeated in more than one chapter, precisely to allow this self-contained
feature. On the other hand, sometimes this repetition is avoided for the sake of
clarity, but we provide a number of cross-references to other chapters. Finally, in
some parts of the book, concepts that will be defined in subsequent chapters are
intuitively used in advance, with a forward cross-reference.
We provide three indices for the book. In addition to the typical subject
index, we include a functions and packages index and an ISO standards index.
Thus, the reader can easily find examples of R code, and references to specific
standards.
The book is organized in four parts. Part I contains four chapters with the
fundamentals of the topics addressed in the book, namely: quality control
(Chapters 1 and 3 ), R (Chapter 2 ), and ISO standards (Chapter 4 ). Part II
contains two chapters devoted to the statistical background applied in quality
control, i.e., descriptive statistics, probability, and inference (Chapter 5 ) and
sampling (Chapter 6 ). Part III tackles the important task of assessing quality
from two different approaches: acceptance sampling (Chapter 7 ) and capability
analysis (Chapter 8 ). Finally, Part IV covers the monitoring of processes via
control charts: Chapter 9 for monitoring variables and attributes quality
characteristics and Chapter 10 for monitoring so-called nonlinear profiles.
Three appendices complete the book. Appendix A provides the classical
Shewhart constants used to compute control chart limits and the code to get them
interactively with R; Appendix B provides the complete list of ISO standards
published by the ISO Technical Committee ISO-TC 69 (Statistical Methods);
and Appendix C is a cheat sheet for quality control with R, containing short
examples of the most common tasks to be performed while applying quality
control with R.
The chapters have a common structure with an introduction to the incumbent
topic, followed by an explanation illustrated with straightforward and
reproducible examples. The material used in these examples (data and code) and
the results (output and graphics) are included sequentially as the concepts are
explained. All figures include a brief explanation to enhance the understanding
of the interpretation. The last section of each chapter includes a summary and
references of the ISO standards relevant for the topics covered in the chapter. 1
We are aware that the book does not cover all the topics concerning quality
control. That was not the intention of the authors. The book paves the way to
encourage readers to go into quality control and R in depth and maybe make
them as enthusiastic as the authors in both topics. The reader can follow the
references provided in each chapter to go into deeper detail on the methods,
especially through the ISO standards.
Finally, if you read the Use R! series book entitled Six Sigma with R , co-
authored by two of this book’s authors, you may find very similar content in
some topics. This is natural, as some techniques in quality control are shared
with Six Sigma methodologies. In any case, we tried to provide a different
approach, with different examples and the ISO standards extent.

Conventions
We use a homogeneous typeset throughout the book so that elements can be
easily identified by the reader. Text in Sans-Serif font is for software (e.g., R ,
Minitab ). Text in teletype font within paragraphs is used for R components
(packages, functions, arguments, objects, commands, variables, etc.).
The commands and scripts are formatted in blocks, using teletype font with
gray background. Moreover, the syntax is highlighted, so the function names,
character strings, and function arguments are colored (in the electronic version)
or with different grayscales (printed version). Thus, an input block of code will
look like this:

The text output appears just below the command that produces it, and with a
gray background. Each line of the output is preceded by two hashes ( ## ):

There are quite a lot of examples in the book. They are numbered and start
with the string Example (Brief title for the example) and finish with a square ( □ )
at the end of the example. In the subsequent evolution of the example within the
chapter, the string (cont.) is added to the example title.
Throughout the book, when we talk about products, it will be very often
suitable for services. Likewise, we use in a general manner the term customer
when referring to customers and/or clients.

The Production
The book has been written in .Rnw files. Both Eclipse + StatET IDE and RStudio
have been used as both editor and interface with R . Notice that if you have a
different version of R or updated version of the packages, you may not get
exactly the same outputs. The session info of the machine where the code has
been run is:
R version 3.2.1 (2015-06-18), x86_64-pc-linux-gnu
Locale: LC_CTYPE=es_ES.UTF-8 , LC_NUMERIC=C , LC_TIME=es_ES.UTF-8 ,
LC_COLLATE=es_ES.UTF-8 , LC_MONETARY=es_ES.UTF-8 ,
LC_MESSAGES=es_ES.UTF-8 , LC_PAPER=es_ES.UTF-8 , LC_NAME=es_ES.UTF-8 ,
LC_ADDRESS=es_ES.UTF-8 , LC_TELEPHONE=es_ES.UTF-8 ,
LC_MEASUREMENT=es_ES.UTF-8 , LC_IDENTIFICATION=es_ES.UTF-8
Base packages: base, datasets, graphics, grDevices, grid, methods, stats,
utils
Other packages: AcceptanceSampling 1.0-3, car 2.0-25, ctv 0.8-1,
downloader 0.3, e1071 1.6-4, Formula 1.2-1, ggplot2 1.0.1, Hmisc 3.16-0,
ISOweek 0.6-2, knitr 1.10.5, lattice 0.20-31, MASS 7.3-42, nortest 1.0-3,
qcc 2.6, qicharts 0.2.0, qualityTools 1.54, rj 2.0.3-1, rvest 0.2.0,
scales 0.2.5, SixSigma 0.8-1, spc 0.5.1, survival 2.38-3, XML 3.98-1.3,
xtable 1.7-4
Loaded via a namespace (and not attached): acepack 1.3-3.3, class 7.3-13,
cluster 2.0.2, colorspace 1.2-6, crayon 1.3.0, curl 0.9.1, digest 0.6.8,
evaluate 0.7, foreign 0.8-64, formatR 1.2, gridExtra 0.9.1, gtable 0.1.2,
highr 0.5, httr 1.0.0, labeling 0.3, latticeExtra 0.6-26, lme4 1.1-8,
magrittr 1.5, Matrix 1.2-0, memoise 0.2.1, mgcv 1.8-6, minqa 1.2.4,
munsell 0.4.2, nlme 3.1-121, nloptr 1.0.4, nnet 7.3-10, parallel 3.2.1,
pbkrtest 0.4-2, plyr 1.8.3, proto 0.3-10, quantreg 5.11, R6 2.1.0,
RColorBrewer 1.1-2, Rcpp 0.11.6, reshape2 1.4.1, rj.gd 2.0.0-1, rpart 4.1-
10, selectr 0.2-3, SparseM 1.6, splines 3.2.1, stringi 0.5-5, stringr 1.0.0,
tcltk 3.2.1, testthat 0.10.0, tools 3.2.1

Resources
The code and the figures included in this book are available at the book
companion website: http://www.qualitycontrolwithr.com . The data sets used in
the examples are available in the SixSigma package. Links and materials will be
updated in a regular basis.

About the Authors

The authors are members of the technical subcommittee AEN CTN66/SC3 at
AENOR (Spanish member of ISO), with Mariano Prieto as the president of such
committee.
Emilio L. Cano is Adjunct Lecturer at the University of Castilla-La Mancha
and Research Assistant Professor at Rey Juan Carlos University. He also
collaborates with the Spanish Association for Quality (AEC) as trainer for in-
company courses. He has more than 14 years of experience in the private sector
as statistician.
Javier M. Moguerza is Associate Professor in Statistics and Operations
Research at Rey Juan Carlos University. He publishes mainly in the fields of
mathematical programming and machine learning. Currently, he is leading
national and international research ICT projects funded by public and private
organizations. He belongs to the Global Young Academy since 2010.
Mariano Prieto Corcoba is Continuous Improvement Manager at ENUSA
Industrias Avanzadas. He has 30 years of experience in the fields of nuclear
engineering and quality. He collaborates with the Spanish Association for
Quality (AEC) as trainer in Six Sigma methodology. Currently, he is president of
the Subcommittee of Statistical Methods in AENOR.
Albacete, Spain, Madrid, Spain, Madrid, Spain
July 2015
Acknowledgments
We wish to thank Virgilio Gómez-Rubio for his kind foreword and the time
devoted to reading the manuscript. We appreciate the gentle review of Iván
Moya Alcón from AENOR on the ISO topics. We thank the Springer staff (Mark
Strauss, Hannah Bracken, Veronika Rosteck, Eve Mayer, Michael Penn, Jay
Popham) for their support and encouragement. A debt of gratitude must be paid
to R contributors, particularly to the R core group ( http://www.r-project.org/
contributors.html ), for their huge work in developing and maintaining the R
project. We also acknowledge projects OPTIMOS 3 (MTM2012-36163-C06-06),
PPI (RTC-2015-3580-7), and UNIKO (RTC-2015-3521-7), Content &
Inteligence (IPT-2012-0912-430000) in which the methodology described in this
book has been applied.
Last but not least, we are eternally grateful to our families for their patience,
forgiving us for the stolen time. Thanks Alicia, Angela, Manuela, Beatriz,
Helena, Isabel, Lucía, Pablo, and Sonia.
Acronyms
AEC Asociación Española para la Calidad
AENOR Asociación Española de NORmalización y certificación
AHG Ad Hoc Group
ANOVA ANalysis Of VAriance
ANSI American National Standards Institute
AQL Acceptable (or Acceptability) Quality Level
ARL Average Run Length
AWI Approved Work Item
BSI British Standards Institution
CAG Chairman Advisory Group
CD Committee Draft
CL Center Line
CLI Command Line Interface
CRAN The Comprehensive R Archive Network
DBMS DataBase Management System
DFSS Design for Six Sigma
DIS Draft International Standard
DoE Design of Experiments
DPMO Defects Per Million Opportunities
ESS Emacs Speaks Statistics
EWMA Exponentially Weighted Moving Average
FAQs Frequently Asked Questions
FDA Federal Drug Administration
FDIS Final Draft International Standard
FOSS Free and Open Source Software
GUI Graphical User Interface
ICS International Classification for Standards
IDE Integrated Development Environment
IEC International Electrotechnical Council
IQR Interquartile range
ISO International Standards Organization
JTC Joint Technical Committee
LCL Lower Control Limit
LL Lower Limit
LSL Lower Specification Limit
LT Long Term
MAD Median Absolute Deviation
MDB Menus and Dialog Boxes
MR Moving Range
NCD Normal Cumulative Distribution
OBP Online Browse Platform (by ISO)
OC Operating Characteristic (curve)
ODBC Open Database Connectivity
OS Operating System
PAS Publicly Available Specification
PLC Programmable Logic Controller
PMBoK Project Management Base of Knowledge
QC Quality Control
QFD Quality Function Deployment
RCA Root Cause Analysis
RNG Random Number Generation
RPD Robust Parameter Design
RSS Really Simple Syndication
RUG R User Group
SC Subcommittee
SDLC Software Development Life Cycle
SME Small and Medium-sized Enterprise
SPC Statistical Process Control
ST Short Term
SVM Support Vector Machine
TC Technical Committee
TMB Technical Management Board
TR Technical Report
TS Technical Specification
UCL Upper Control Limit
UL Upper Limit
URL Uniform Resource Locator
USL Upper Specification Limit
VoC Voice of the Customer
VoP Voice of the Process
VoS Voice of Stakeholders
WD Working Draft
XML eXtended Markup Language
Contents
Part I Fundamentals

1 An Intuitive Introduction to Quality Control with R

1.1 Introduction

1.2 A Brief History of Quality Control

1.3 What Is Quality Control

1.4 The Power of R for Quality Control

1.5 An Intuitive Example

1.6 A Roadmap to Getting Started with R for Quality Control

1.7 Conclusions and Further Steps

References

2 An Introduction to R for Quality Control

2.1 Introduction

2.2 R Interfaces

2.3 R Expressions

2.4 R Infrastructure

2.5 Introduction to RStudio

2.6 Working with Data in R

2.7 Data Import and Export with R

2.8 R Task View for Quality Control (Unofficial)

 2.9 ISO Standards and R

References

3 The Seven Quality Control Tools in a Nutshell: R and ISO Approaches

3.1 Origin

3.2 Cause-and-Effect Diagram

3.3 Check Sheet

3.4 Control Chart

3.5 Histogram

3.6 Pareto Chart

3.7 Scatter Plot

3.8 Stratification

3.9 ISO Standards for the Seven Basic Quality Control Tools

References

4 R and the ISO Standards for Quality Control

4.1 ISO Members and Technical Committees

4.2 ISO Standards and Quality

4.3 The ISO Standards Development Process

4.4 ISO TC69 Secretariat

4.5 ISO TC69/SC1: Terminology

4.6 ISO TC69/SC4: Application of Statistical Methods in Process

Management
 4.7 ISO TC69/SC5: Acceptance Sampling

4.8 ISO TC69/SC6: Measurement Methods and Results

4.9 ISO TC69/SC7: Applications of Statistical and Related Techniques

4.10 ISO TC69/SC8: Application of Statistical and Related Methodology

for New Technology and Product Development

4.11 The Role of R in Standards

References

Part II Statistics for Quality Control

5 Modelling Quality with R

5.1 The Description of Variability

5.1.1 Background

5.1.2 Graphical Description of Variation

5.1.3 Numerical Description of Variation

5.2 Probability Distributions

5.2.1 Discrete Distributions

5.2.2 Continuous Distributions

5.3 Inference About Distribution Parameters

5.3.1 Confidence Intervals

5.3.2 Hypothesis Testing

5.4 ISO Standards for Quality Modeling with R

References
6 Data Sampling for Quality Control with R

6.1 The Importance of Sampling

6.2 Different Kinds of Sampling

6.2.1 Simple Random Sampling

6.2.2 Stratified Sampling

6.2.3 Cluster Sampling

6.2.4 Systematic Sampling

6.3 Sample Size, Test Power, and OC Curves with R

6.4 ISO Standards for Sampling with R

References

Part III Delimiting and Assessing Quality

7 Acceptance Sampling with R

7.1 Introduction

7.2 Sampling Plans for Attributes

7.3 Sampling Plans for Variables

7.4 ISO Standards for Acceptance Sampling and R

References

8 Quality Specifications and Process Capability Analysis with R

8.1 Introduction

8.2 Tolerance Limits and Specifications Design

8.2.1 The Voice of the Customer

 8.2.2 Process Tolerance

8.3 Capability Analysis

8.3.1 The Voice of the Process

8.3.2 Process Performance Indices

8.3.3 Capability Indices

8.4 ISO Standards for Capability Analysis and R

References

Part IV Control Charts

9 Control Charts with R

9.1 Introduction

9.1.1 The Elements of a Control Chart

9.1.2 Control Chart Design

9.1.3 Reading a Control Chart

9.2 Control Charts for Variables

9.2.1 Introduction

9.2.2 Estimation of σ for Control Charts

9.2.3 Control Charts for Grouped Data

9.2.4 Control Charts for Non-grouped Data

9.2.5 Special Control Charts

9.3 Control Charts for Attributes

9.3.1 Introduction
 9.3.2 Attributes Control Charts for Groups

9.3.3 Control Charts for Events

9.4 Control Chart Selection

9.5 ISO Standards for Control Charts

References

10 Nonlinear Profiles with R

10.1 Introduction

10.2 Nonlinear Profiles Basics

10.3 Phase I and Phase II Analysis

10.3.1 Phase I

10.3.2 Phase II

10.4 A Simple Profiles Control Chart

10.5 ISO Standards for Nonlinear Profiles and R

References

A Shewhart Constants for Control Charts

B ISO Standards Published by the ISO/TC69: Application of Statistical

Methods

C R Cheat Sheet for Quality Control

R Packages and Functions Used in the Book

ISO Standards Referenced in the Book

Subject Index
List of Figures
Fig. 1.1 Out of control process

Fig. 1.2 Chance causes variability

Fig. 1.3 Assignable causes variability

Fig. 1.4 Results under a normal distribution

Fig. 1.5 Typical control chart example

Fig. 1.6 R learning curve

Fig. 1.7 R Project website homepage

Fig. 1.8 CRAN web page

Fig. 1.9 Intuitive example control chart

Fig. 1.10 RStudio layout

Fig. 1.11 Example control chart

Fig. 1.12 RStudio new R markdown dialog box

Fig. 1.13 Markdown word report (p1)

Fig. 1.14 Markdown word report (p2)

Fig. 2.1 R GUI for Windows

Fig. 2.2 RStudio Layout

Fig. 2.3 RStudio Console

Fig. 2.4 RStudio Source

Fig. 2.5 RStudio History

Fig. 2.6 RStudio export graphic dialog box

Fig. 2.7 RStudio History

Fig. 2.8 RStudio Workspace

Fig. 2.9 RStudio Files pane

Fig. 2.10 RStudio Packages

Fig. 2.11 RStudio Help

Fig. 2.12 RStudio data viewer

Fig. 2.13 RStudio Import Dataset

Fig. 3.1 Intuitive Cause-and-effect diagram (qcc)

Fig. 3.2 Intuitive Cause-and-effect diagram (SixSigma)

Fig. 3.3 R Markdown Check sheet

Fig. 3.4 Filled Check sheet

Fig. 3.5 Control chart tool

Fig. 3.6 Pellets density basic histogram

Fig. 3.7 A histogram with options

Fig. 3.8 A lattice-based histogram

Fig. 3.9 A ggplot2-based histogram

Fig. 3.10 A simple barplot

Fig. 3.11 Basic Pareto chart

Fig. 3.12 Pareto chart with the qcc package

Fig. 3.13 Pareto chart with the qualityTools package

Fig. 3.14 Pareto chart with the qicharts package

Fig. 3.15 Scatter plot example

Fig. 3.16 Stratified box plots

Fig. 4.1 ISO Standards publication path

Fig. 4.2 ISO TC69 web page

Fig. 5.1 Thickness example: histogram

Fig. 5.2 Thickness example: histograms by groups

Fig. 5.3 Thickness example: simple run chart

Fig. 5.4 Thickness example: run chart with tests

Fig. 5.5 Thickness example: tier chart by shifts

Fig. 5.6 Thickness example: box plot (all data)

Fig. 5.7 Thickness example: box plots by groups

Fig. 5.8 Thickness example: lattice box plots

Fig. 5.9 Histogram with central tendency measures

Fig. 5.10 Normal distribution

Fig. 5.11 Histogram of non-normal density data

Fig. 5.12 Individuals control chart of non-normal density data

Fig. 5.13 Box-Cox transformation plot

Fig. 5.14 Control chart of transformed data

Fig. 5.15 Quantile-Quantile plot

Fig. 5.16 Quantile-Quantile plot (non normal)

Fig. 6.1 Error types

Fig. 6.2 OC Curves

Fig. 7.1 OC curve for a simple sampling plan

Fig. 7.2 OC curve risks illustration

Fig. 7.3 OC curve with the AcceptanceSampling package

Fig. 7.4 OC curve for the found plan

Fig. 7.5 Variables acceptance sampling illustration

Fig. 7.6 Probability of acceptance when p =AQL

Fig. 7.7 Probability of acceptance when p =LTPD

Fig. 8.1 Taguchi’s loss function and specification design

Fig. 8.2 Thickness example: One week data dot plot

Fig. 8.3 Reference limits in a Normal distribution

Fig. 8.4 Histogram of metal plates thickness

Fig. 8.5 Specification limits vs. reference limits

Fig. 8.6 Capability analysis for the thickness example

Fig. 9.1 Control charts vs. probability distribution

Fig. 9.2 Identifying special causes through individual points

Fig. 9.3 Patterns in control charts

Fig. 9.4 Control chart zones

Fig. 9.5 X-bar chart example (basic options)

Fig. 9.6 X-bar chart example (extended options)

Fig. 9.7 OC curve for the X-bar control chart

Fig. 9.8 Range chart for metal plates thickness

Fig. 9.9 S chart for metal plates thickness

Fig. 9.10 X-bar and S chart for metal plates thickness

Fig. 9.11 I & MR control charts for metal plates thickness

Fig. 9.12 CUSUM chart for metal plates thickness

Fig. 9.13 EWMA chart for metal plates thickness

Fig. 9.14 p chart for metal plates thickness

Fig. 9.15 np chart for metal plates thickness

Fig. 9.16 c chart for metal plates thickness

Fig. 9.17 u chart for metal plates thickness

Fig. 9.18 Decision tree for basic process control charts

Fig. 10.1 Single woodboard example

Fig. 10.2 Single woodboard example (smoothed)

Fig. 10.3 Woodboard example: whole set of profiles

Fig. 10.4 Woodboard example: whole set of smoothed profiles

Fig. 10.5 Woodboard example: Phase I

Fig. 10.6 Woodboard example: In-control Phase I group

Fig. 10.7 Woodboard example: Phase II

Fig. 10.8 Woodboard example: Phase II out of control

Fig. 10.9 Woodboard example: Profiles control chart

List of Tables
Table 1.1 CRAN task views

Table 1.2 Pellets density data (g/cm 3 )

Table 4.1 Standard development project stages

Table 5.1 Thickness of a certain steel plate

Table 6.1 Complex bills population

Table 6.2 Pellets density data

Table 7.1 Iterative sampling plan selection method

Table A.1 Shewhart constants

Footnotes
1 ISO Standards are continuously evolving. All references to standards throughout the book are specific for
a given point in time. In particular, this point in time is end of June 2015.

Part I
Fundamentals
 This part includes four chapters with the fundamentals of the three topics
covered by the book, namely: Quality Control, R, and ISO Standards. Chapter 1
introduces the problem through an intuitive example, which is also solved using
the R software. Chapter 2 comprises a description of the R ecosystem and a
complete set of explanations and examples regarding the use of R. In Chapter 3,
the seven basic quality tools are explored from the R and ISO perspectives.
Those straightforward tools will smoothly allow the reader to get used to both
Quality Control and R. Finally, the importance of standards and how they are
made can be found in Chapter 4
© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_1

1. An Intuitive Introduction to Quality

Control with R
Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
This chapter introduces Quality Control by means of an intuitive example.
Furthermore, that example is used to illustrate how to use the R statistical
software and programming language for Quality Control. A description of R
outlining its advantages is also included in this chapter, all in all paving the way
to further investigation throughout the book.

1.1 Introduction
This chapter provides the necessary background to understand the fundamental
ideas behind quality control from a statistical perspective. It provides a review of
the history of quality control in Sect. 1.2. The nature of variability and the
different kinds of causes responsible for it within a process are described in
Sect. 1.3; this section also introduces the control chart, which is the fundamental
tool used in statistical quality control. Sect. 1.4 introduces the advantages of
using R for quality control. Sect. 1.5 develops an intuitive example of a control
chart. Finally, Sect. 1.6 provides a roadmap to getting started with R while
reproducing the example in Sect. 1.5.
1.2 A Brief History of Quality Control
Back in 1924, while working for the Bell Telephone Co. in solving certain
problems related to the quality of some electrical components, Walter Shewhart
set up the foundations of modern statistical quality control [16]. Until that time
the concept of quality was limited to check that a product characteristic was
within its design limits. Shewhart’s revolutionary contribution was the concept
of “process control.” From this new perspective, a product’s characteristic within
its design limits is only a necessary—but not a sufficient—condition to allow the
producer to be satisfied with the process . The idea behind this concept is that the
inherent and inevitable variability of every process can be tracked by means of
simple and straightforward statistical tools that permit the producer to detect the
moment when abnormal variation appears in the process. This is the moment
when the process can be labeled as “out of control,” and some action should be
put in place to correct the situation.
A simple example will help us understand this concept. Let’s suppose a
factory is producing metal plate whose thickness is a critical attribute of the
product according to customer needs. The producer will carefully control the
thickness of successive lots of product, and will make a graphical representation
of this variable with respect to time, see Fig. 1.1. Between points A and B the
process exhibits a small variability around the center of the acceptable range of
values. But something happens after point C, because the fluctuation of values is
much more evident, together with a shift in the average values in the direction of
the Upper Specification Limit (USL) . This is the point when it is said that the
process has gone out of control. After this period, the operator makes some kind
of adjustments in the process (point E) that allows the process to come back to
the original controlled state.
Fig. 1.1 Out of control. Example of an out-of-control process

It is worth noting that none of the points represented in this example are out
of the specification limits, which means that all the production is defect-free .
Although one could think that, after all, what really matters is the distinction
between defects and
non-defects, an out-of-control situation of a process is highly undesirable as
long as it is evident that the producer no longer controls the process and is at the
mercy of chance. These ideas of statistical quality control were quickly
assimilated by industry and even today, almost one century after the pioneering
work of Shewhart, constitute one of the basic pillars of modern quality.

1.3 What Is Quality Control

Production processes are random in nature. This means that no matter how much
care one could place in the process, its response will somewhat vary with time. It
is possible to classify process variability into two main categories: chance
variation and assignable variation . When the variability present in a process is
the result of many causes , having each of them a very small contribution of total
variation, being these causes inherent to the process (i.e., impossible to be
eliminated or even identified in some cases), we say that the process shows a
random normal noise. This comes from the definition of a normal distribution of
random values. In a normal distribution the values tend to be grouped around the
average value, the farther from the average the less probable that a value may
occur. When variability comes only from chance causes (also called common
causes ) the behavior of the process is more predictable; no trends or patterns are
present in the data (Fig. 1.2). In this case the process is said to be under control.

Fig. 1.2 Chance causes. Variability resulting from chance causes. The process is under control

But in certain circumstances processes deviate from this kind of behavior,

some of the causes responsible for the variation become strong enough as to
introduce recognizable patterns in the evolution of data, i.e. step changes in the
mean, tendencies, increase in the standard deviation, etc. This kind of variation
is much more unpredictable than in the previous situation. This special behavior
of the process is the result of a few causes, having each of them a significant
contribution of total variation. These causes are not inherent to the process and
are called assignable causes (also called special causes ). Fig. 1.3 shows a case
where a tendency is clearly observed in the data after point A. In this case the
process is said to be out of control .
Fig. 1.3 Assignable causes. Variability resulting from assignable causes. The process is out of control

From both previous examples it becomes evident that a graphical

representation of the evolution of process data with time is a powerful means of
getting a first idea of the possible state of control of the process. But in order to
give a final judgment over a process’ state of control, something more is needed.
If we suppose that the process is free of assignable causes, thus assuming that
the process is under control, then we would expect a behavior of the process that
could be reasonably described by a normal distribution. A detailed description of
the normal distribution can be consulted in Chapter 5 Under this assumption,
process results become less and less probable as they get farther from the
process mean (μ). If, as it is common practice, we state this distance from the
process mean in terms of the magnitude of the standard deviation (σ) the
probabilities of obtaining a data point in the different regions of the normal
distribution are given in Fig. 1.4. From this figure it comes out that the
probability of obtaining a data point from the process whose distance to the
process mean is larger than 3σ is as small as 0.27 %. This probability is, indeed,
Fig. 1.4 Normal distribution. Probability of a result in different regions of the normal distribution

very small and should lead us to question if the process really is under
control. If we combine this idea with the graphical representation of the process
data with time, we will have developed the first and simplest of the control
charts.
The control chart is the main tool that is used in the statistical processes
control. A control chart is a time series plot of process data to which three lines
are superposed; the mean, the Upper Control Limit (UCL), and the Lower
Control Limit (LCL). As a first approach, upper and lower control limits are
separated from the process mean by a magnitude equal to three standard
deviations (3σ), thus setting up a clear boundary between those values that could
be reasonably expected and those that should be the result of assignable causes.
Figure 1.5 shows all the different parts of a typical control chart: the center line,
calculated as the average value (μ) of the data points, the UCL, calculated as the
sum of the average plus three standard deviations of the data points (μ + 3σ), and
the LCL calculated as the subtraction of the average minus three standard
deviations of the data points (μ − 3σ). A chart constructed in this way is at the
same time a powerful and simple tool that can be used to determine the moment
in which a process gets out of control. The reasoning behind the control chart is
that any time a data point falls outside of the region comprised by both control
limits, there exist a very high probability that an assignable cause has appeared
in the process .
Fig. 1.5 A typical control chart. Data points are plotted sequentially along with the control limits and the
center line

Although the criterion of one data point falling farther than three standard
deviations from the mean is the simplest one to understand based on the nature
of a normal process, some others also exist. For example:
Two of three consecutive data points farther than two standard deviations
from the mean;
Four of five consecutive data points farther than one standard deviation
from the mean;
Eight consecutive data points falling at the same side of the mean;
Six consecutive data points steadily increasing or decreasing;
Etc.
What have all these patterns in common? The answer is simple in statistical
terms; all of them correspond to situations of very low probability if chance
variation were the only one present in the process. Then, it should be concluded
that some assignable cause is in place and the process is out of control .

1.4 The Power of R for Quality Control

Software for Quality Control
The techniques we apply for quality control are based on the data about our
processes. The data acquisition and treatment strategy should be an important
part of the quality control planning , as all the subsequent activities will be based
on such data. Once we have the data available, we need the appropriate
computing tools to analyze them. The application of statistical methods to
Quality Control requires the use of specialized software. Of course we can use
spreadsheets for some tasks, but as we get more and more involved in serious
data analysis for quality control, we need more advanced tools. Spreadsheets can
be still useful for entering the raw data , correct errors, or export results for
further uses.
There exist a wide range of software packages for Statistics in general. Most
of them include specific options for quality control, such as control charts or
capability analysis . Even some of them are focused on quality tools. A thorough
survey of statistical software would be cumbersome, and it is out of the scope of
this book. The reader can find quite a complete list at the Wikipedia entry for Six
Sigma .1 We can see that almost all the available software packages are
proprietary and commercial . This means that one needs to buy a licence to use
them. Nowadays, however, there are more and more Free and Open Source
Software (FOSS ) options for any purpose. In particular, for the scope of this
book, the R statistical software [15] is available.
Before going into the details of R, we would like to make some remarks
about the use of FOSS. Even though reluctance remains for its use within
companies, it is a fact that some FOSS projects are widely used throughout the
World. For example, the use of the Linux Operating System (OS ) is not restricted
to computer geeks anymore thanks to distributions like Ubuntu . Not to mention
Internet software such as php and Apache , or the MySQL database management
system (DBMS ) .
As for the R software and programming language , it is widely spread that it
has become the de-facto standard for data analysis , see, for example, [1]. In fact,
many large companies such as Google, The New York Times, and many others
are already using R as analytic software. Moreover, during the last years some
commercial options have appeared for those companies who need a commercial
licence for any reason, and professional support is also provided by such
companies. Another signal for this trend is the amount of job positions that
include R skills as a requirement. A simple search on the web or professional
social networks is enlightening.
The Free part of FOSS typically implies the following four essential
freedoms [3]2:
The freedom to run the program as you wish, for any purpose (freedom 0);
The freedom to study how the program works, and change it so it does your
computing as you wish (freedom 1);
The freedom to redistribute copies so you can help your neighbor (freedom
 2);
The freedom to distribute copies of your modified versions to others
(freedom 3).
Note that the access to source code , i.e., the OS part of FOSS, is mandatory
for freedom 1 and 3. It is usually said that FOSS means free as in beer and free
as in speech. Therefore, it is apparent that the use of FOSS is a competitive
choice for all kinds of companies, but especially for Small and Medium-sized
Enterprises (SMEs ). One step beyond, we would say that it is a textbook Lean
measure.3

What Is R?
R is the evolution of the S language created in the Bell laboratories in the 1970s
by a group of researchers led by John Chambers and Rick Becker [2]. Note that,
in this sense, quality control and R are siblings, see Sect. 1.2 . Later on, in the
1990s Ross Ihaka and Robert Gentleman designed R as FOSS largely compatible
with S [5]. Definitely, the open source choice encouraged the scientific
community to further develop R, and the R-core was created afterwards. At the
beginning, R was mainly used in academia and research. Nevertheless, as
R evolved it was more and more used in other environments, such as private
companies and public administrations. Nowadays it is one of the most popular
software packages for analytics.4
R is platform-independent, it is available for Linux , Mac , and Windows . It is
FOSS and can be downloaded from the Comprehensive R Archive Network
(CRAN )5 repository. We can find in [4] the following definition of R :

R is a system for statistical computation and graphics. It consists of a

language plus a run-time environment with graphics, a debugger, access to
certain system functions, and the ability to run programs stored in script
files.

Let us go into some interesting details of R from its own definition:

It is a system for statistical computation and graphics. So, we can do
statistics and graphics, but it is more than a statistical package: it is a
system;
It is also a programming language . This means that it can be extended
with new, tailored functionality. Advanced programming features as
debugging or system interaction are available, but just for those users who
 need them;
The run-time environment allows to use the software in an interactive
way;
Writing script files to be run afterwards either in a regular periodic basis or
for an ad-hoc need is the natural way to use R.
From the above definition, we can realize that there are two ways to use R:
interaction and scripting. Surprisingly for the newcomer, interaction means the
use of a console where expressions are entered by the user, resulting on a
response by the system. By creating scripts , expressions can be arranged in an
organized way and stored in files to be edited and/or run afterwards. Interaction
is useful for testing things, learning about the software, or exploring intermediate
results. Nevertheless, the collection of expressions that lead to a given set of
results should be organized by means of scripts. An R script is a text file
containing R expressions that can be run individually or globally.
In addition to a system, R can also be considered a community . Apart from
the formal structure through the R foundation (see below), R Users organize
themselves all over the World to create local R User’s Groups (RUGs ). There is
an updated list6 on the blog of Revolution Analytics,7 which is a company
specialized in analytics with R. They have developed their own interface for R,
and a number of packages to deal with Big Data . Revolution is a usual sponsor
of R events and local groups, and provide commercial support to organizations
using R. Other commercial companies providing R services and support are
RStudio,8 Open Analytics,9 or TIBCO,10 for example. The R community is very
active in the R mailing lists . You can find a relation of the available lists from
the R website . One can subscribe to the suitable list of their interest, place a
question and wait for the solution. However, most of the times the question has
already been posted anywhere and answered by several people. A simple web
search with the question (including “R” on it) will likely return links to
Stackoverflow11 not only with answers, but also with discussions on different
approaches to tackle the problem.
Being R an Open Source project, it is not strange that people ask themselves
who is behind the project, and how it is maintained. We can find out that in the R
website itself (see the following section). Visit the following links in the left side
menu at the home page:
Contributors . The R Development Core Team have write access to the R
source. They are in charge of updating the code. More people contribute by
donating code, bug fixes, and documentation;
 The R Foundation for Statistical Computing . The statutes can be
downloaded from the R website;
Members and Donors. A number of people and institutions support the
project as benefactors, supporting institutions, donors, supporting members,
and ordinary members. We can find relevant companies in the list, such as
AT&T and Google, among others;
The Institute for Statistics and Mathematics of WU (Wirtschaftsuniversität
Wien, Vienna University of Economics and Business) hosts the foundation
and the servers.

Why R?
The ways of using R described above may sound old-fashioned. However, this is
a systematic way of work which, once is appropriately learned, it is far more
effective than the usual point, click, drag, and drop features of a software based
on windows and menus. More often than not, such user-friendly Graphical User
Interfaces (GUIs ) avoid the user to think on what they are actually doing, just
because there is a mechanical sequence of clicks that do the work for them.
When users have to write what they want the machine to do, they must know
what they want the software to do. Still, extra motivation is needed to start using
R. The learning curve for R is very slow at the beginning, and it takes a lot of
time to learn things, see Fig. 1.6. This is discouraging for learners, especially
when you are stressed by the need of getting results quickly in a competitive
environment. However, this initial effort is rewarding. Once one grasps the
basics of the language and the new way of doing things, i.e., writing rather than
clicking, impressive results are get easily. Moreover, the flexibility of having
unlimited possibilities both through the implemented functionality and one’s
own developments fosters the user creativity and allows asking questions and
looking for answers, creating new knowledge for their organization.
Fig. 1.6 R learning curve. It takes a lot of time to learn something about R, but then you create new things
very quickly. The time units vary depending on the user’s previous skills. Note that the curve is asymptotic:
you never become an expert, but are always learning something new

In addition to the cost-free motivation, there are many reasons for choosing
R as the statistical software for quality control. We outline here some of the
strengths of the R project, which are further developed in the subsequent
sections:
It is Free and Open Source;
The system runs in almost any system and configuration and the installation
is easy;
There is a base functionality for a wide range of statistical computation and
graphics, such as descriptive statistics, statistical inference, time series, data
mining, multivariate plotting, advanced graphics, optimization,
mathematics, etc;
The base installation can be enriched by installing contributed packages
devoted to particular topics, for example for quality control;
It has Reproducible Research and Literate Programming capabilities [14];
New functionality can be added to fulfill any user or company
requirements;
Interfacing with other languages such as Python , C , or Fortran is possible, as
well as wrapping other programs within R scripts ;
There is a wide range of options to get support on R, including the extensive
R documentation , the R community, and commercial support.

We provide enough evidence about those advantages of using R throughout

the book. In Sect. 2.8, chapter 2 an overview of the available functions and
packages for quality control are provided. Once the initial barriers have been
overcome, creating quality control reports is a piece of cake as shown in
Sect. 1.6.

How to Obtain R
The official R project website12 is the main source of information to start with R.
Even though the website design is quite austere, it contains a lot of resources, see
Fig. 1.7.

Fig. 1.7 R project website homepage. The left menu bar provides access to basic R information, the
CRAN, and documentation

In the central part of the homepage we can find two blocks of information:
Getting Started: Provides links to the download pages and to the answers
to the frequently asked questions;
News: Feed with the recent news about R: new releases , conferences , and
issues of the R Journal .
In addition, the following sections are available from the left side menu:
About R: Basic information about R and the latest news;
Download, packages: A link to the CRAN repository;
R Project: Varied information about the R Project, its foundation, donors,
conferences, and some tools;
Documentation: It is one of the strengths of R. The Frequently Asked
Questions (FAQs ) is a good starting point. It is a short document with
 general answers about the system, and also to very common questions
arising when starting to use R. The Manuals are quite complete and updated
with each release. There are different manuals for different levels;
Misc: This miscellaneous section provides links to other resources.
The links to download the R software and the link to CRAN lead to the
selection of a mirror. The R project is hosted at the Institute for Statistics and
Mathematics of WU (Wirtschaftsuniversität Wien, Vienna University of
Economics and Business). Mirrors are replicated servers throughout the world
maintained by private and public organizations that kindly contribute to the R
ecosystem. It is recommended to select a mirror near your location when
downloading CRAN files. The main server can be directly accessed without a
mirror selection at the URL : http://cran.r-project.org.
The CRAN web page, see Fig. 1.8, has links to download and install the
software for Linux , Windows , and Mac . For Linux Users, the repository of the
selected mirror can be added to the sources list and then install and get updates
in the usual way using the package management system. For Windows and Mac
users, installation files can be downloaded and installed by double-clicking
them. In any case, the installation is straightforward and the default settings are
recommended for most users.

Fig. 1.8 CRAN web page. Access to the R software, including the sources, documentation, and other
information

Some other interesting resources are available in the CRAN web page. The
source code can be also downloaded, not only for the last release, but also for the
current development of R, and for older versions. From the left side menu, we
can access further resources. The most impressive one is the Packages section.
Add-on packages are optional software that extend the R software with more
code, data, and documentation. The R distribution itself includes about 30
packages. Nevertheless, the number of contributed packages is astonishing. At
the time this is written,13 more than 6500 packages are available at CRAN for a
number of different applications. Each contributed package has a web page at
CRAN with links to download the package (binary or source code), manuals,
and other resources. Moreover, CRAN is not the only repository for R packages.
Other R repositories are Bioconductor14 and Omegahat,15 and more and more
developers are using generic software repositories such as GitHub16 to publish
their packages. In total, the rdocumentation.org17 website records 7393
packages. The installation of add-on packages is straightforward in R, especially
for those available at CRAN, as will be shown in Chapter 2.
Another great resource at CRAN are the Task Views . A Task View is a
collection of resources related to a given topic that bring together R functions,
packages, documentation, links, and other materials, classified and commented
by the Task View maintainer. The task views are maintained by contributors to
the R Project who are experts on the subject. The Task Views available at CRAN
are listed in Table 1.1. Currently, there is not a Task View for Quality Control.
Nevertheless, we include in Chapter 2 a sort of proposal for it.
Table 1.1 CRAN task views
Name Topic
Bayesian Bayesian inference
ChemPhys Chemometrics and computational physics
ClinicalTrials Clinical trial design, monitoring, and analysis
Cluster Cluster analysis and finite mixture models
DifferentialEquations Differential equations
Distributions Probability distributions
Econometrics Econometrics
Environmetrics Analysis of ecological and environmental data
ExperimentalDesign Design of experiments (DoE) and analysis of experimental data
Finance Empirical finance
Genetics Statistical genetics
Graphics Graphic displays and dynamic graphics and graphic devices and
visualization
HighPerformanceComputing High-performance and parallel computing with R
MachineLearning Machine learning and statistical learning
MedicalImaging Medical image analysis
MetaAnalysis Meta-analysis
Multivariate Multivariate statistics
NaturalLanguageProcessing Natural language processing
NumericalMathematics Numerical mathematics
OfficialStatistics Official statistics and survey methodology
Optimization Optimization and mathematical programming
Pharmacokinetics Analysis of pharmacokinetic data
Phylogenetics Phylogenetics, especially comparative methods
Psychometrics Psychometric models and methods
ReproducibleResearch Reproducible research
Robust Robust statistical methods
SocialSciences Statistics for the social sciences
Spatial Analysis of spatial data
SpatioTemporal Handling and analyzing spatio-temporal data
Survival Survival analysis
TimeSeries Time series analysis
WebTechnologies Web technologies and services
gR gRaphical models in R

1.5 An Intuitive Example

We will illustrate quality control and R with an intuitive and comprehensive
example, covering from raw data to reporting using R. The example is also used
in other chapters.

Example 1.1.
Pellets density.
A certain ceramic process produces pellets whose density is a critical quality
characteristic according to customer needs. Current technical specification states
that the density of a pellet is considered acceptable if it is greater than 10.5 
g/cm3. A sample of one pellet is taken and measured, following a standardized
inspection process, after each hour of continuous operation. The complete set of
inspection data for a one-day period is in Table 1.2 (left-to-right ordered).

Table 1.2 Pellets density data (g/cm3)

10.6817 10.6040 10.5709 10.7858 10.7668 10.8101

10.6905 10.6079 10.5724 10.7736 11.0921 11.1023
11.0934 10.8530 10.6774 10.6712 10.6935 10.5669
10.8002 10.7607 10.5470 10.5555 10.5705 10.7723
 What could be said about product quality and process control ? A quick
check of data indicates that all the product is according to specifications; as long
as there is no single data point below 10.5 g/cm3. To determine the control limits
, the following approach should be followed:

In this particular case, μ is estimated by the sample mean and σ is

calculated as , where is the so-called average moving range of the
values, and d 2 is a tabulated constant , see Chapter 9 and Appendix A for details.
In this case, = 10.7342 g/cm3; = 0.1064; and d 2 = 1.1284. Therefore:

The simple observation of the resulting control chart in Fig. 1.9 leads us to

conclude that points 11, 12, and 13 are beyond the upper control limit; therefore,
the process is out of control in that time and some kind of investigation and
corrective action should be implemented. In the following section we will see
how to plot this control chart with R. □ 
Fig. 1.9 Intuitive example control chart. Even though all the points are within the technical specification,
the process is out of control for points 11–13. Further investigation is needed to detect and correct
assignable causes

1.6 A Roadmap to Getting Started with R

for Quality Control
In this section, a step-by-step process is described in order to get ready for using
R for Quality Control. The process starts with the installation of R and RStudio
and ends with the plotting of the quality control chart in the previous section. For
the sake of simplicity, the procedure is described for Windows users. Most of the
explanations are valid for all platforms, but Linux and Mac users might need to
translate some things to their specific system jargon, for example regarding
installation and execution of applications. Please note that the pointers to the
links on the websites might have slightly changed when you are reading this
book.

Download R
1. Go to http://www.r-project.org;

2. Click on the Download R link;

3. Select a mirror close to your location, or the 0-Cloud first one to get
automatically redirected. Any of the links should work, though the
downloading time could vary;

4. Click on the Download R for Windows link;

5. Click on the base link;

6. Click on the Download R 3.x.x for Windows link;

7. Save the.exe file in a folder of your choice, for example the Downloads folder
within your Users folder.

Install R
1. Open the folder where you saved the.exe file;

2. Double-click the R-X.x.x-win.exe file;

3. Accept the default options in the installation wizard.

Download RStudio
1. Go to http://www.rstudio.com;

2. Click on the Powerful IDE for R link;

3. Click on the Desktop link;

4. Click on the DOWNLOAD RSTUDIO DESKTOP button;

5. Click on the RStudio X.xx.xxx - Windows XP/Vista/7/8 link;

6.
Save the.exe file in a folder of your choice, for example the Downloads folder
within your Users folder.

Install RStudio
1. Make sure you have the last version of Java18 installed on your system;

2. Open the folder where you saved the.exe file;

3. Double-click the RStudio-X.xx.xxx.exe file;

4. Accept the default options in the installation wizard.

Start RStudio
After the installation of R and RStudio, you will have on your desktop an icon for
RStudio. As for R, if your system is a 64-bit one, you will have two new icons:
one for R for 32 bits and another for R for 64 bits. In general, it is recommended
to run the version that matches your OS architecture. Double click the RStudio
icon and you should see something similar to the screen capture in Fig. 1.10.
More details about R and RStudio are provided in Chapter 2.
Fig. 1.10 RStudio application. This is what we see when starting RStudio

Install the qcc Package

1. Go to the Packages tab on the lower-right pane;

2. Click on the Install button. A dialog box appears;

3. Type “qcc” on the Packages text box;

4. Click on Install. You will see some messages in the RStudio console.
Packages only need to be installed once, but they need to be loaded in the
workspace in every session that uses functions of the package (see next
steps).

Select and Set Your Working Directory (Optional)
By default, the R working directory is your home directory, e.g., My Documents.
This working directory is shown in the Files tab (lower-right pane) when
opening RStudio. The working directory can be changed following these steps,
even though it is not needed for the purpose of this example:
Select the Files tab in the lower-right pane;
Click the …button on the upper-right side of the Files Pane (see Fig. 1.10);
Look for the directory you want to be the working directory. For example,
create a folder on your home directory called qcrbook;
Click on the Select button;
Click on the More… menu on the title bar of the files tab and select the Set
as working directory option;
Check that the title bar of the console pane (left-down) shows now the path
to your working directory.

Create Your First Script

To reproduce the illustrative example in this chapter you need: (1) the data on
your workspace; (2) the qcc package loaded; and (3) the expression that
generates the plot. Let us write a script with the expressions needed to get the
result.

1. Create a new R Script . You can use the File menu and select New File/R
Script, or click the New File command button (first icon on the toolbar) and
select R Script. A blank file is created and opened in your source editor pane .

2. Save your file. Even though it is empty, it is good practice to save the file at
the beginning and then save changes regularly while editing. By default, the
Save File dialog box goes to the current working directory. You can save the
file there, or create a folder for your scripts. Choose a name for your file, for
example “roadmap” or “roadmap.R”. If you do not write any extension,
RStudio does it for you.

3. Write the expressions you need to get the control chart in the script file:
Create a vector with the pellets data. The following expression creates a
vector with the values in Table 1.2 using the function c, and assigns
(through the operator <-) the vector to the symbol pdensity.
 If you are reading the electronic version of this chapter, you can copy
and paste the code. The code is also available at the book’s companion
website19 in plain text. Please note that when copying and pasting from a
pdf file, sometimes you can get non-ascii characters, spurious text, or
other inconsistencies. If you get an error or a different result when
running a pasted expression, please check that the expression is exactly
what you see in the book. In any case, we recommend typing everything,
at least at the beginning, in order to get used to the R mood.
Load the qcc package . This is done with the function library as
follows:

Use the function qcc to create a control chart for individual values of the
pellets density data:

Remember to save your script. If you did not do that earlier, put a name
to the file.

Run Your Script
Now you have a script with three expressions. To run the whole script, click on
the Source icon in the tool bar above your script. And that is it! You should get
the control chart and output text in Fig. 1.11 in the Plots pane and in the console,
respectively. As you may have noticed, it is not exactly the control chart in
Fig. 1.9. Now we have provided the qcc function with the minimum amount of
information it needs to produce the control chart. Further arguments can be
provided to the function, and we can also add things such as lines and text to the
plot. The following code is the one that produced Fig. 1.9:
Fig. 1.11 Example control chart. Your first control chart with R following the steps of the roadmap
 After loading the qcc package, the first expression produces the plot; the
second one draws a horizontal line at the specification limit; and the third one
puts a text explaining the line. Do not worry for the moment about these details,
just keep in mind that we can go from simple expressions to complex programs
as needed. As for the text output in Figure 1.11, it is the structure of the object
returned by the qcc function. You will learn more about R objects and output
results in Chapter 2.

Make a Reproducible Report (Optional)

One of the strengths of R remarked above was that it has Reproducible Research
and Literate Programming Capabilities. In short, this means that we can include
the data analysis within our reports , so that they can be reproduced later on by
ourselves, or by other collaborators, or by third parties, and so on. Usually, this
is done by including so-called chunks of code within report files. This can be
done with the base installation of R using the LaTeX typesetting system.
However, LaTeX is a complex language that requires practice and, again, a
tough learning time. Fortunately, thanks to some contributed packages such as
knitr and the inclusion of the pandoc software with the RStudio installation,
quality control reports can be easily created. The key point is that instead of
using the complex LaTeX language, RStudio relies on the markdown language,
which is very easy to use for simple formatted text. Furthermore, the output
report can be a.html,.pdf, or.docx file. To generate.pdf files it is needed to have
installed a LaTeX distribution20, even though the report is written in markdown.
Reports with.docx extension can be open with several software packages, such
as Microsoft Office and LibreOffice . Follow these steps to get a Microsoft Word report of
the intuitive example in Sect. 1.5:

1. Create a new R Markdown file. You can use the File menu and select New
File/R Markdown …, or click the New File command button (first icon on the
toolbar) and select R Markdown. A dialog box appears to select the format of
your report, see Fig. 1.12. Type a title and author for your report, select the
option Word and click OK21;

Fig. 1.12 RStudio new R markdown dialog box

2. A new file is shown in the source editor , but this time it is not empty. By
default, RStudio creates a new report based on a template;

3. Keep the first six lines as they are. They are the meta data to create the report,
i.e.: title, author, date, and output format;

4. Take a look at the next two paragraphs. Then delete this text and write
something suitable for your quality control report;

5.
The chunks of code are embraced between the lines ‘‘‘{r} and ‘‘‘. Change the
expression in the first chunk of the template by the first two expressions of the
script you created above;

6. Change the expression in the second chunk of the template by the third
expression of the script you created above;

7. Read the last paragraph of the template to realize what is the difference
between the two chunks of code. Change this paragraph by something you
want to say in your report.

8. You should have something similar to this in your R Markdown file:
 ---
 
title: "Quality Control with R: Intuitive Example"
 author: "Emilio L. Cano"
 date: "31/01/2015"
 output: word_document
 ---
 
This is my first report of quality control using R.
 First, I will create the data I need from the
 example in the Book *Quality Control with R*. I
 also need to load the ‘qcc‘ library:
 ‘‘‘{r}
 
pdensity <- c(10.6817, 10.6040, 10.5709, 10.7858,
 
10.7668, 10.8101, 10.6905, 10.6079,
 
10.5724, 10.7736, 11.0921, 11.1023,
 
11.0934, 10.8530, 10.6774, 10.6712,
  
10.6935, 10.5669, 10.8002, 10.7607,
 
10.5470, 10.5555, 10.5705, 10.7723)
 library(qcc)
 ‘‘‘
 And this is my first control chart using R:
 ‘‘‘{r, echo=FALSE}
 qcc(data = pdensity, type = "xbar.one")
 ‘‘‘
 
It worked! Using R for quality control is great :)

9. Knit the report. Once the R Markdown is ready and saved, click on the knit
Word icon on the source editor toolbar. The resulting report is in Figs. 1.13
and 1.14. Note that everything needed for the report is in the R Markdown
file: the data, the analysis, and the text of the report. Imagine there is an error
in some of the data points of the first expression. All you need to get an
updated report is to change the bad data and knit the report again. This is a
simple example, but when reports get longer, changing things using the
typical copy-paste approach is far less efficient. Another example of a Lean
measure we are applying.
Fig. 1.13 Markdown word report. Page 1
 Fig. 1.14 Markdown word report. Page 2

1.7 Conclusions and Further Steps
In this chapter we have intuitively introduced quality control by means of a
simple example and straightforward R code. Next chapters will develop different
statistical techniques to deal with different quality control scenarios. These
techniques include further types of control charts in order to deal with
continuous and discrete manufacturing, as well as acceptance sampling or
process capability analysis.
On the other hand, every chapter contains a section devoted to the
International Standards Organization (ISO) standards that are relevant to the
chapter contents. Moreover, Appendix B comprises the full list of standards
published by the Application of Statistical Methods ISO Technical Committee
(TC), i.e., ISO TC/69. Relevant ISO standards for this introductory chapter are
those pertaining to vocabulary and symbols, as well as the more general ones
describing methodologies. Even though we will see them in more specific
chapters, these are the more representative generic standards of ISO/TC 69:

ISO 3534-1:2006 Statistics—Vocabulary and symbols—Part 1: General

statistical terms and terms used in probability [6].
ISO 3534-2:2006 Statistics—Vocabulary and symbols—Part 2: Applied
statistics [7].
ISO 11462-1:2001 Guidelines for implementation of statistical process
control (SPC)—Part 1: Elements of SPC [8].
ISO 7870-1:2014 Control charts—Part 1: General guidelines [11].
ISO 7870-2:2013 Control charts—Part 2: Shewhart control charts [9].
ISO 22514-1:2014 Statistical methods in process management—Capability
and performance—Part 1: General principles and concepts [10].
ISO 10576-1:2003 Statistical methods—Guidelines for the evaluation of
conformity with specified requirements—Part 1: General principles [13].
ISO 5725-1:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 1: General principles and definitions [12].

The use of Standards provides benefits to companies, and therefore they

should embrace standards in their quality control strategies. Chapter 4 provides
an overview of ISO Standards, their development process, and committees
structure.
References
1. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. Statistical Engineering for Process
Improvement. Use R!, vol. 36. Springer, New York (2012). http://www.springer.com/statistics/book/
978-1-4614-3651-5

2. Chambers, J.M.: Software for Data Analysis. Programming with R. Statistics and Computing. Springer,
Berlin (2008)
[MATH][CrossRef]

3. Free Software Foundation, Inc.: Free Software Foundation website. http://gnu.org (2014). Accessed 10
July 2014

4. Hornik, K.: R FAQ. http://CRAN.R-project.org/doc/FAQ/R-FAQ.html (2014)

5. Ihaka, R., Gentleman, R.: R: a language for data analysis and graphics. J. Comput. Graph. Stat. 5, 299–
314 (1996)

6. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard. http://www.iso.org/
iso/catalogue_detail.htm?csnumber=40145 (2010)

7. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=
40147 (2014)

8. ISO TC69/SC4–Applications of statistical methods in process management: ISO 11462-1:2001 -

Guidelines for implementation of statistical process control (SPC) – Part 1: Elements of SPC. Published
standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=33381 (2012)

9. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 - Control

charts – Part 2: Shewhart control charts. Published standard. http://www.iso.org/iso/catalogue_detail.
htm?csnumber=40174 (2013)

10. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-1:2014 -

Statistical methods in process management – Capability and performance – Part 1: General principles
and concepts. Published standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=64135
(2014)

11. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-1:2014 - Control
charts – Part 1: General guidelines. Published standard. http://www.iso.org/iso/catalogue_detail.htm?
csnumber=62649 (2014)

12. ISO TC69/SC6–Measurement methods and results: ISO 5725-1:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 1: General principles and definitions. Published
standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=11833 (2012)

13. ISO TC69/SC6–Measurement methods and results: ISO 10576-1:2003 - Statistical methods –
Guidelines for the evaluation of conformity with specified requirements – Part 1: General principles.
Published standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=32373 (2014)

14. Leisch, F.: Sweave: dynamic generation of statistical reports using literate data analysis. In: Härdle, W.,
 Rönz, B. (eds.) Compstat 2002 — Proceedings in Computational Statistics, pp. 575–580. Physica,
Heidelberg (2002). http://www.stat.uni-muenchen.de/~leisch/Sweave

15. R Core Team: R: A Language and Environment for Statistical Computing. R Foundation for Statistical
Computing, Vienna (2015). http://www.R-project.org/

16. Shewhart, W.: Economic Control of Quality in Manufactured Products. Van Nostrom, New York (1931)

Footnotes
1 http://en.Wikipedia.org/wiki/Six_Sigma.

2 See more about free software at http://gnu.org/philosophy/free-sw.en.html.

3 Lean, or Lean Manufacturing, is a quality methodology based on the reduction of waste.

4 r4stats.com/articles/popularity.

5 http://cran.r-project.org.

6 http://blog.revolutionanalytics.com/local-r-groups.html.

7 http://www.revolutionanalytics.com.

8 http://www.rstudio.com.

9 http://www.openanalytics.eu.

10 http://www.tibco.com.

11 http://stackoverflow.com/tags/r.

12 http://www.r-project.org.

13 April 2015.

14 http://bioconductor.org.

15 http://omegahat.org.

16 http://github.org.

17 http://rdocumentation.org.

18 Check at http://www.java.com.

19 http://www.qualitycontrolwithr.com.

20 Visit http://www.miktex.com for a LaTeX distribution for Windows.

21 If there are missing or outdated packages, RStudio will ask for permission to install or update them. Just
answer yes.

© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_2

2. An Introduction to R for Quality Control

Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
This chapter introduces R as statistical software and programming language for
quality control. The chapter is organized as a kind of tutorial with lots of
examples ready to be run by the reader. Moreover, the code is available at the
book’s companion website. Even though the RStudio interface is also introduced
in the chapter, any other user interface can be used, including the R default GUI
and code editor.

2.1 Introduction
In this chapter, the essentials of the R statistical software and programming
language [27] are explained. This provides the reader with the basic knowledge
to start using R for quality control. You should try the code by yourself while
reading this chapter, and therefore you need R and RStudio (optionally but
recommended) installed on your computer before continuing reading the chapter.
Follow the step-by-step instructions explained in Sect. 1.6 of Chapter 1, or just
go to the R website1 and to the RStudio website,2 download the installation files,
and install them to your computer. If you are reading the electronic version of
this chapter, you can copy and paste the code in the examples.3 The code is also
available at the book’s companion website.4 In any case, we recommend typing
everything, at least at the beginning, in order to get used to the R mood.
In Chapter 1, we introduced the power of R for quality control, what it is, its
history, etc. This chapter goes into the details of the software to get advantage of
that power. We highlight here some of the R features explained in Sect. 1.4 of
Chapter 1:
R is the evolution of the S language, developed at Bell Laboratories (then
AT&T and Lucent Technologies) in the 1970s [6]. Note that it is the same
company where Walter Shewhart developed modern statistical quality
control 50 years before [34];
R is maintained by a foundation , a Core Team , and a huge community of
users and stakeholders, including commercial companies that make their
own developments;
R is Free and Open Source Software (FOSS ). Free as in free beer, and free
as in free speech [14];
R is also a programming language, and a system for statistical computing
and graphics;
R is platform independent: it runs in Windows, Mac, and Linux;
The way of interacting with R is by means of expressions , which are
evaluated in the R Console , or can be stored in R scripts to be run as
programs;
R has Reproducible Research and Literate Programming capabilities, which
has proven quite useful for quality control reports in Sect. 1.6, Chapter 1;
R base functionality provides a complete set of tools for statistical
computing and plotting, developed by time-honored experts;
R base functionality is expanded by an increasing number of contributed
packages for a wide range of applications, including some for quality
control;
The software can be customized creating new functions for particular
needs.
The toughest part for new R users is to get used to the interactivity with the
system. Having to write the expressions prompts errors which, especially at the
beginning, are not easy to interpret. Nevertheless, those errors are usually caused
by similar patterns. Find below a list of common errors while writing R
expressions. If you get an error when running an R expression, it very likely can
be classified into one of those categories. Please take into account that those
types of errors are not made only by beginners, but it is part of the normal use of
R. Practice will reduce the number of times errors are produced and, more
importantly, the time one lasts realizing where is the problem and fix the
expression. This list contains concepts that you still do not know about. Note the
list as a reference and come back here whenever you get an error while reading
the chapter and practicing with the code. Once you have completed the chapter,
read the list again to fix concepts.
Missing closing character. You need to close all the parentheses, square
brackets, curly brackets, or quotation marks you had opened in an
expression. Otherwise the expression is incomplete, and the console prompt
keeps waiting for you to finish it with the + symbol. If you are running a
script, R will try to continue the expression with the next one, and the error
message could be uninformative. So always check that you do not have a
missing closing symbol;
String characters without quotation marks. String characters must be
provided in quotation marks ("). Everything in an expression that is not in
quotation marks is evaluated in the workspace, and therefore it should exist,
either in the global environment or in other environments. Usually, the error
message indicates that the object does not exist, or something related to the
class of the object;
Missing parenthesis in functions. Every function must include
parenthesis, even if it does not need any argument;
Missing arguments in a function call. Functions accept arguments, which
sometimes can be omitted either because they are optional or because they
have a default value. Otherwise they are mandatory and a message indicates
so;
Wrong arguments in functions. Sometimes it is due to the missing
quotation marks mentioned above. Check the class of the object you are
using as argument, and the class the function expects as argument;
Incompatible lengths. Data objects have a length that may be critical when
using them. For example, the columns of a data frame must be of the same
length;
Wrong data. If a vector of data is supposed to be, for example, numeric,
but one or more of its components is another thing, for example a character
string, then computations over the vector might result on unexpected
results. This is not always so evident, as it may be a number but the
 computer might interpret a character, for example due to spurious blank
spaces or the like;
Other wrong syntax errors. Check the following:
– The arguments in a function are separated by commas (,);
– The conditions in loops and conditions are in parenthesis;
– You do not have wrong blank spaces, for example in the assignment
operator;
– Use a period (.) as decimal separator in numbers;
– Expressions are in different lines, or separated by a semicolon.
In the remaining of the chapter you will find an overview of R interfaces in
Sect. 2.2; a description of the main R elements in Sect. 2.4; an introduction to
RStudio in Sect. 2.5; Sects. 2.6 and 2.7 describe how to work with data within R
and with external data sources. This is the starting point for the application of the
quality control tools explained throughout the book; a QualityControl task view
is proposed in Sect. 2.8; finally, some ideas and thoughts about R and
Standardization are given in Sect. 2.9. Note that the specific functions and
packages for quality control are not included in this chapter, as they are
explained in detail in the corresponding chapter. For example, functions for
modelling processes are in Chapter 5, and so on. Appendix C is a complete cheat
sheet for quality control with R.

2.2 R Interfaces
The R base installation comes with a Command Line Interface (CLI) . This CLI
allows interacting with R using the R Console as outlined above, by means of
expressions and scripts. This is one of the hardest parts for beginners, especially
for those who do not have experience in programming. Luckily, being R open
source software and a programming language at the same time allows
developing more advanced interfaces to work with R. For the Windows and Mac
versions of R, an extremely simple Graphical User Interface (GUI) is also
included with the base installation. It can be started as any other application in
the system, Figure 2.1 shows the GUI for Windows.
Fig. 2.1 R GUI for windows. The R GUI allows basic interaction with R through the R console; scripts can
be created using the R Editor; and the R Graphics device opens when invoking a plot. The menu bar
contains access to some basic operations such as installing packages, or save and load files

There are a number of projects regarding R interfaces. A list of them can be

found in the R website itself following the “Related Projects” link and then R
GUIs, or just visit http://www.sciviews.org/_rgui. We can find two types of R
interfaces:
Interfaces with menus and dialog boxes (MDB GUIs) . Interfaces of this
type provide the user with menus and dialog boxes to perform statistical
analysis in a similar way other commercial software does. However, only a
limited number of options are included in those menus. They are based on a
common framework, where package developers build functionality for their
functions. The most popular frameworks are R Commander (package Rcmdr )
[13] and Deducer (package Deducer) [11], and they can be loaded inside R as
any other package;
Interfaces for development (Integrated Development Environment,
IDE) . Interfaces of this type provide an environment to make the analyst
life easier, but they do not provide an interface where one can enter data or
click options and then run an analysis. Nevertheless, they allow to exploit
all the capabilities of the R system. Most popular environments include
RStudio 5 [30], Emacs + ESS 6 (Emacs Speaks Statistics), and Eclipse + StatET .7

The approach followed in this book is using an interface of the second type.
This allows to use all the capabilities of R, and the examples provided
throughout the book can be used either in the built-in R GUI, both in the R
console and as scripts in the R Editor, or in other available GUIs. In what
follows, we explain one of the interfaces that has become very popular among a
wide range of R users, including those using R in Industry: RStudio. This choice
does not mean that one interface is better than the others. In fact, we invite the
reader to try out more than one and decide by themselves which one fits better
their needs. In fact, we have been using both RStudio and Eclipse + StatET to write
this book using Reproducible Research and Literate Programming techniques.
The good thing is that we can choose between several alternatives. Moreover, as
we remarked above, all the examples in the book are ready to use in any R
interface, or interactively in the console.

2.3 R Expressions
The way to interact with R is through R expressions, sometimes named as
commands. As explained above, R is interactive, in the sense that it responds to
given inputs. Such inputs are R expressions, which can be of several types,
mainly:
An arithmetic expression;
A logical expression;
A call to a function;
An assignment.
Expressions are evaluated by R when run in the console or through a script. If
the expression is incomplete, the R Console prompt keeps waiting until the
expression is complete. If the expression is complete, R evaluates the expression,
producing a result. This result may show some output to the user, which can be
textual or graphical. Some expressions do not produce any visible output, being
the result, for example, storing data in variables, or writing data to disk.
One of the characteristics of R is that it works with in-memory data.
Nevertheless, we will need to work with expressions containing files in several
ways. Some of them are:
Read data files to use in data analysis;
Write data files to use later on;
Save plots to be included in reports using other software tools;
Create R scripts to write sets of expressions containing a complete analysis;
 Create report files with code, results, data, and text suitable to be compiled
and delivered.
In summary, the purpose of using files in R can be either working with data,
or working with code. When files are involved in R expressions, we can provide
the file location using two approaches:
Through the absolute path, i.e., the location in the computer from the root
file system;
Through the relative path, i.e., the location in the computer from the
working directory of the R session (see below).
File paths must be provided as character strings, and therefore quotation
marks must be used. When using Windows, it is important to note that the
backward slash character (“\”) is reserved for escaping 8 in R, and Windows
paths must be provided using either a forward slash (“/”) or a double backward
slash (“\\”) to separate folders and file names. For relative paths, the usual
symbols for current and parent directories (“.” and “..” respectively) can be used.

2.4 R Infrastructure
The R infrastructure is composed of the following elements:
The console
The editor
The graphical output
The history
The workspace
The working directory
In the R GUI, the console, the editor, and the graphical output are the three
windows that can be seen in Fig. 2.1. However, the history, the workspace, and
the working directory are hidden and we need to use coding to access them. As
remarked above, interfaces like RStudio allow more options in order to work
with those system-related elements. Moreover, advanced functionality is
available to: easily access to objects and functions; syntax highlighting;
contextual menus; access to help; explore data; etc. Nevertheless, the interface is
actually a wrapper for the R system, and the level of interaction for the statistical
analysis is the same: console and scripts.
2.5 Introduction to RStudio
RStudio is a Java-based application, and therefore having Java installed is a
prerequisite. Make sure you have the latest version of Java9 to avoid possible
issues. The RStudio interface is shown in Fig. 2.2.10 It has a layout of four panes
whose dimensions can be adjusted, and each pane can contain different types of
elements by means of tabs. Most of those elements are basic components of the
R system listed above. The default layout is as follows11:

Fig. 2.2 RStudio layout. The RStudio interface is divided into four panes: the console pane, the source
pane, the workspace and history pane, and the files, plots, packages, and help pane. The layout can be
modified through the global options in the Tools menu

1. Lower-left pane. This pane is for the R Console . It can also show system-
related elements such as the output of the system console when calling
system commands, for example to compile a report;

2. Upper-left pane. This pane is for the R Source. R Scripts are managed in this
pane. Other types of files can also be opened in this pane, for example text
files containing data, code in other programming languages, or report files.
Data sets are also shown in this pane;

3. Upper-right pane. This pane is mainly for the R History and the R
Environment Other tabs appear when using certain features of RStudio, such
as packages development, or R Presentations;

4. Lower-right pane. This pane is the most populated. It has the following tabs:
Files. It is a system file explorer with basic functions. It can be linked to
the R working directory;
Plots. It is the RStudio graphics device. The plots generated in R are
shown here;
Packages. Shows the packages available in the system, and we can
install, uninstall, or update them easily;
Help. This tab provides access to all the R Documentation, including the
documentation of the installed contributed packages;
Viewer. This tab is used to develop web applications with RStudio,
which will not be covered in this book.

The R(Studio) Console
The R console in RStudio is located by default in the lower-left pane, see
Fig. 2.3. Its behavior is the same as in the standard R GUI: there is a prompt
identified by the “>” symbol that is waiting for an expression. The user writes an
expression after the prompt and presses the Intro or return key. R evaluates the
expression and produces a result. An important issue puzzling for newcomers
that arises quite often is that if an expression is incomplete, the prompt changes
to the “+” symbol, waiting for the rest of the expression. Most of the times the
user thought that the expression was complete and does not know what to do.
Usually, it is due to a missing closing parenthesis or the like, and the way to
cancel the expression is to press the ESC key. Some details about the RStudio
console:
Fig. 2.3 RStudio console. The RStudio console provides interaction with R

We can go to the console prompt using the keyboard shortcut CTRL+2 from
anywhere;
The title bar of the RStudio console contains a valuable information: the
current working directory (see below);
The arrow next to the working directory path is to show the working
directory in the files pane;
When writing an expression, we can press TAB or CTRL+SPACEBAR to see a
contextual menu and select: available objects in the workspace and
functions; arguments of a function (within a function); or files and folders
of the working directory (within quotation marks);
The ESC key cancels the current expression;
CTRL+L clears the console;
The up and down arrow keys navigate through the history.
In what follows, R code is shown in gray background. Input expressions can
be written directly in the RStudio Console or script editor (or copy-pasted if you
are reading the electronic version of this book). The output produced by R is
shown in the book after two hash symbols (“##”) at the beginning of the line. For
example, the simplest expression we can input is a number. Type the number 1 at
the console and press Intro:
 We can see that the result of this input expression is a line of text with the
number 1 between squared brackets followed by the number 1. The number in
square brackets is an identifier that will be explained later. The result of the
expression is the same number that we wrote. One step beyond would be to ask
for a calculation, for example:

Now the output is the result of the arithmetic expression. What happens if the
expression is incomplete?

As you may have realized, the > symbol changes to +, denoting that the
expression is not complete. The system remains in that state until either the
expression is completed or the ESC key is pressed, cancelling the expression.
Arithmetic expressions return the result of the operation. Another type of R
expressions are logical expressions, which return the TRUE or FALSE value:

An R expression can be a call to a function. In fact, it is the most used type

of expression. The call to a function is always the same: the name of the function
followed (imperatively) by opening and closing parenthesis and, within the
parenthesis, the arguments of the function separated by commas. The function
arguments can be provided in several ways:
Explicitly by the name of the argument in the form “name = value”.
R allows partial matching of names12
Implicitly in the same order they were defined in the function.
Using the default value defined for the function.
We can see the arguments of a function and their default values in the
documentation of the function, or by pressing the TAB key after the opening
parenthesis. The function str using a function name as an argument also returns
the arguments of the function. This is a simple example of the use of a function:

where pi is itself an expression that gets the value of the internal object
containing the value of π = 3,14159….

The log function gets the logarithm of a number. We can see the possible
arguments of the function through the function str 13:

Therefore, log is a function that accepts two arguments: x, that does not have
any default value, and base, whose default value is the expression:

i.e., the e constant. Thus, the value that we get with the log function is the
natural logarithm, i.e., with base e, of the number we pass as first argument, or
with a different base if we pass the base argument. For example, the decimal
logarithm would be:

What happens if we pass no arguments to the function?

What happens is that the expression returns an error because there is no

default value for the first argument (x) and the function needs it. Read carefully
the error messages, they usually have clues to solve the problem.
Some functions do not need any arguments to work. For example, the seq
function generates sequences of numbers.

The dot-dot-dot (…) argument means that the function accepts an undefined
number of arguments. So, will it work without arguments?

It works, we get a sequence of numbers form 1 to 1 by steps of 1, i.e., the

number 1. We can find out more about the arguments a function accepts using
any of the following expressions14:

The documentation for the function is then shown in the Help tab, lower-
right pane. As the function does not need any argument to work, could we use it
without parenthesis?

The answer is no, because in R, every symbol is an object. The expression

seq without parenthesis is the symbol of the function seq, and what R returns is
the content of the function, i.e., its code.15
Let us finish this subsection with an explanation of the mysterious [1] at the
beginning of the R console output. It is meaningless when the output is only one
value, but when the output is a set of values that occupy more than one line, it is
useful. The number at the beginning of each line is the index of the first element
in the row. You will learn more about lengths and indices later on, just see the
following example in which a vector with 20 random variates from a normal
distribution are generated. The [19] at the beginning of the last row indicates
that the first value in that line is the 19th value in the vector.

Note that you might get a different number of elements per row, as the output
width has been set to 50 characters for the generation of the book’s code, you
can set your own preferred output with the options function as follows:

The Source Editor

In the source editor we create text files with R expressions. Expressions can be
more than one line length. In fact, when expressions are too long, it is better to
split it in lines in order to make the code more readable. On the other hand, more
than one expression can be placed in the same line if we separate them with a
semicolon (;). Comments can be included in the code using the # character. R
scripts files have .R extension. The reason for using scripts is to reuse the code
that we write once and ordinarily use it afterwards, maybe with slight changes,
in quality control data analysis.
The RStudio source editor, see Fig. 2.4, has the following improvements with
respect to the R editor in the R GUI:
Code highlighting: functions, objects, numbers, and texts appear with
different colors;
Automatic indentation of code;
Automatic closing of opening embracement characters such as parenthesis,
square brackets, curly braces, and quotation marks;
CTRL+SPACEBAR and TAB keys: Provide some help in the same way that we
described for the console;
Link to the documentation using the F1 key.

Fig. 2.4 RStudio source editor. The RStudio source editor can manage R scripts, reports, and code in other
programming languages, such as C++ and Python

Complex scripts can be run from the console or from other scripts. For
example, if we have a script called dayly.R that performs ordinary tasks that we
want to use in other scripts, e.g., loading packages, import data, etc., we can run
such script with the following expression:

The R Graphical Output

One of the R’s strengths is its graphical system. Publication-quality plots can be
easily produced, including mathematical symbols and formulae. The defaults of
the plotting functions have been set for the minor design choices, but the user
retains full control and elaborated plots can be made by tuning the graphical
parameters up, functions’ arguments, and global options.
The R graphical system is based in so-called devices. Plots are sent to
devices. If no device is open, a new one is open when calling a high-level
graphics function. The grDevices package in the R base includes a number of
devices, including the appropriate one for the user’s OS, and file-format devices
such as pdf, jpeg, or png among others.
Devices can be managed by several functions both interactively and through
scripts, thereby controlling the graphical output of our code. A global option
gives the default device, which is initially set as the most appropriate for each
platform. Some R packages provide further graphics devices. This is the case of
RStudio, which includes its own graphics device in the lower-right pane, see
Fig. 2.5.

Fig. 2.5 RStudio plots tab. Plots generated in RStudio are shown in the plots tab, lower-right pane
 The RStudio graphics device includes a menu bar with several options that
makes life easier with devices management:
Navigation through the graphics history;
Zooming;
Export files to several formats using a dialog box;
Removing and clearing of graphics history.
The export menu includes three options for graphics: save as image, save as
pdf, and copy to clipboard. The former two open a dialog box with export
options such as the file extension (in the case of image), file path, and image
size, see Fig. 2.6.

Fig. 2.6 RStudio export graphic dialog box. A preview of the image is shown along with the export
options: image format and dimensions being the most relevant

The graphics package contains functions for “base” graphics. Those

traditional graphics are enough for most cases. The more recent development of
R graphics relies on the grid package [24]. Both of them are included in the R
base, as well as the lattice package, [32] aimed at elegant plots with a focus on
multivariate representations. Another very popular package for graphics is the
ggplot2 package [36]. In the chapters devoted to specific quality control
modelling and analysis we will see in detail how to make different types of plots
and charts.

The R Commands History

When working in the console, the commands history can be accessed using the
up and down arrow keys, like in the R GUI console. In RStudio, we can also
visualize all the history in the history tab, upper-right pane, see Fig. 2.7.
Expressions in the history can be sent either to the console or to the source
editor. Further actions such as save, open, or clean the history are available in the
menu bar of the history tab.

Fig. 2.7 RStudio history. The R History can be easily consulted, searched, and used during an R session
through the History tab in the upper-right pane

The history can also be accessed via R code, see the documentation of the
functions loadhistory, savehistory, history.

The R Workspace
The objects that are available in R are stored in the workspace. The workspace is
organized in different environments. The Global Environment is the place in
which the objects we create through assignments are stored. Each loaded
package has its own environment. Environments are also created for calls to
functions, and the user can even create environments. For the scope of this book,
just keep track of your Global Environment tab, upper-right pane, see Fig. 2.8
where you will find useful information about the objects that are available to use
in your session. You can save, open, search, and clear objects in the workspace
through the menu bar of the Environment tab. To make actions only over
selected objects in the workspace, change the view using the upper-right icon on
the menu bar from “List” (default) to “Grid,” select the objects you want to save
or clear, and click the appropriate button. Remember to change again to the List
view in order to be able to explore the environment. An icon for importing
datasets stored in text files is also available (we will go over this later on).

Fig. 2.8 RStudio workspace. The R Workspace contains a list of available objects in the global
environment, which can be used in R expressions during the R Session

The R workspace can also be accessed via R code, see the documentation of
the functions ls, str , load, save, save.image, and rm.

The Working Directory

A Working Directory is always associated with an R session. All tasks
related to files would take this directory in the file system as the path where read
or write files, for example, to read data in files, save scripts, or export plots.
When an R expression contains file names, they must be provided in quotation
marks ("). Relative paths from the working directory can be used. A tricky
feature of RStudio is that we can pick folders and files when writing in the
source editor or the console just pressing CTRL+SPACEBAR while the cursor is
between quotation marks. The path to the file or folder is auto-completed as
selected in the contextual menu. Furthermore, the Files tab in the lower-right
pane (see Fig. 2.9) is a file explorer which can be linked to the working
directory. To do so, click on the arrow icon in the title bar of the console, right
after the working directory path. The reverse operation is also possible: Click the
“Go to directory” button, that is, the button with the ellipsis on the upper-right
side of the Files Pane; look for the directory you want to be the working
directory; click on the Select button; click on the More… menu on the title bar of
the Files tab and select the Set as working directory option; now the title bar of
the console shows the path to your working directory, which is also visualized in
the Files pane. The working directory can also be set through the Session menu,
either to the Files pane location, the active source file location, or to a chosen
directory. Basic operations such as creating a new folder and renaming or
deleting items can be done. However, it is usually better to show the folder in a
new window through the More… menu and work with files from there. Please
note that files and folders deleted from the Files pane are permanently deleted,
they do not go to the trash system folder. Finally, the default working directory
can be set in the RStudio global options in the Tools menu.

Fig. 2.9 RStudio files pane. Interaction with the file system is possible through the Files pane, including
the setting and visualization of the R working directory

The R working directory and the file system can also be accessed via R code,
see the documentation of the functions getwd (returns the working directory),
setwd (sets the working directory), list.files, list.dirs, and dir. Actually, it
is common practice to include at the beginning of the scripts an expression to set
the working directory, for example:

Recall that the backslash character (“\”) has a special meaning in R and
Windows paths must be provided using either a forward slash (“/”) or a double
backward slash (“\\”) to separate folders and file names. This is particularly
important when copying and pasting paths from the address bar of the Windows
file explorer.

Packages
R functionality is organized by means of packages. The R distribution itself
includes about 30 packages. Some of them are loaded when starting R. In
addition, a number of contributed packages are available, see Sect. 1.4 in
Chapter 1. In order to use the functions of a package, it must be loaded in
advance. Obviously, the package must be installed in the system in order to be
loaded. The installation of a package is done once, while the package must be
loaded in the R workspace every time we want to use it. Both operations can be
done through the Packages pane of RStudio, see Fig. 2.10.
Fig. 2.10 RStudio packages. The Packages tab in the lower-right pane shows a list of installed packages
with links to the documentation and command buttons to manage the packages

To install a package, click on the Install icon in the Packages tab menu bar. A
dialog box opens where we can select whether to install the package from
CRAN or from a local file. To install a package from CRAN, type the name of
the package (or just the first letters to get a list) and click on the Install button. If
you select to install it from a local file, a dialog box appears to search the file.
This is useful for packages that are not published in official repositories, but are
available from other sources. Similarly to the R software, add-on packages are
regularly updated by their authors. Installed packages can be updated by clicking
the Update button in the command bar of the Packages tab. From the list of
installed packages we can also go to the documentation of the package by
clicking on its name, remove the package from the system clicking on the icon
on the right, or load the package in the workspace by selecting the check-box on
the left. Nevertheless, even though the installation of packages is comfortable
through the RStudio interface, it is more convenient to load the packages in the
scripts as they are needed in the code using the library function. For example,
to load the qcc package:

Packages management can also be performed with the install.packages and

remove.packages functions. Other functions related to packages are require,
detach, search, installed.packages, and available. packages, check their
documentation for details. An example of use could be to get the number of
packages available at CRAN on a given date, try out by yourself and check how
R grows:

R and RStudio Help

The R documentation can be accessed through the Help tab in the lower-right
pane of RStudio, see Fig. 2.11. You can go there (there is even a keyboard
shortcut: CTRL+3) and browse the help documentation: Manuals, packages
reference, search engine and keywords, and miscellaneous material. There are
other several ways of getting help in RStudio:
The keyboard shortcut CTRL+SPACEBAR inside a function shows the basic
documentation of that function;
Pressing the F1 key when the cursor is over the name of a function or any
other R object with documentation, for example a dataset;
Using the search text box in the Help tab toolbar. A list of topics starting
with the text of the search appears. If a topic is selected on this list, the
documentation for the topic is shown. Otherwise, a search over all the
documentation is done and the topics in which the search string appears are
listed.
Fig. 2.11 RStudio help. The R documentation is easily available from the RStudio interface

Typically, the documentation of a function contains the following sections:

Description: a paragraph with a description of the function;
Usage: The function name and the arguments it expects as they are defined
in the code;
Arguments: a detailed description of each argument. This is very important
in order to provide the function with objects of the correct class (see the
most common errors in Sect. 2.1);
Value: a description of the returned value. Such value can be stored in R
objects;
References;
See Also: links to related functions or topics;
Examples: reproducible examples which can be copy-pasted or executed by
calling the example function with the function name as argument.
There is a special type of documentation called vignettes, which can include
examples with output and extended explanations. If a package contains
vignettes, then they are available from the package documentation index page.
R help can also be interactively accessed through R expressions. Some
interesting functions to get help in such a way are (check their documentation for
more details): apropos, help, help.search, example, demo, vignette,
browseVignettes.
As for RStudio, there is a “Help” menu in the menu bar. There you can access
to the R help home in the help tab, information about the RStudio version, check
for updates, diagnostics, and a keyboard shortcut quick reference. Regarding
help for RStudio itself, there are links to RStudio Support and to RStudio docs in
the RStudio website. Note that both R and RStudio are continuously evolving,
and there may have been changes with respect to what we are showing in the
book, highly likely improvements for the user benefit. Check the latest
documentation.

2.6 Working with Data in R

Data Structures
In R, data regarding a given variable can be of a specific type, for example
numeric or character. Those variables, in turn, can be included in other data
structures. The simplest form is a vector for the data of a single variable. A
matrix is a vector with more than one dimension. Lists can contain objects of
different type and length. Objects whose class is data.frame are composed by
variables which can be of different type, but with the same length. This is the
most common way of organizing information for data analysis, and also for
quality control. Each column of a data frame is for a variable, and each row
represents an observation (item, individual, etc.) for which all the variables have
been measured.

Classes and Data Types

Regarding data objects in R, we can talk about them in two different senses: their
class and their data type. For example, an object whose class is data.frame may
have columns that are of type numeric, logical, or character, for example. The
main basic data types in R are:
logical: TRUE/FALSE;
integer: Integer number;
double: Real number. It can also appear as numeric or real;
character: String character;
Thus, vectors whose elements are of any of those data types can be created.
There are some other basic types for objects that we do not use in the book, see
the documentation of the typeof function to learn more about them. On the other
hand, those basic data are organized in data structures of different classes. The
most important classes available in R to organize the information are:
vector: One dimensional variable, all the values of the same type;
matrix: Vector organized in rows and columns;
list: List of objects that can be of different types and lengths;
data.frame: Dataset organized in columns of the same length but may have
different type, and rows;
factor: One dimensional categorical variable. In addition to values, a factor
contains information about levels and labels;
POSIXct, Date: Special classes for temporal data.
The classes listed above are enough for the scope of this book, but there are
many more classes in R, and new classes can be created through the
programming capabilities of R. For example, objects of class ts are useful for
working with time series. And remember that R functions are R objects which
class is “function”; Most of those extended classes are actually containers for
objects of other classes, including the basic classes enumerated above.
Regarding quality control, control charts can be stored in objects of class qcc in
order to use the information of the control chart afterwards for further analysis,
for example the points out of control. Such objects are usually a list of other
objects of different classes and types. If in doubt, you can find out the class of an
object through the class function. .

Vectors

Creating Vectors
The most basic classes in R are vectors. They are also very important because
more complex data structures usually are composed by vectors. For example, the
columns of a data frame with the data of a process are actually vectors with the
values of different variables. Therefore, the explanations in this subsection are
mostly valid for working with objects whose class is data.frame or list.
There are several ways of creating vectors. The most basic one is entering
the values interactively in the console using the scan function. If you type on the
console:

then the console prompt changes to “1:” waiting for the first element of the
vector. Type, for example, 10 and press RETURN. Now the prompt changes again
to “2:” and waits for the second element of the vector, and so on. Enter, for
example, two more values: 20 and 30. When you have finished, press INTRO
without any value and the scanning of values finishes. Your output should look
like this:

Now you have a vector whose name is “x” on your workspace. This is what
the assignment operator (“<-”) did: to assign the result of the scan function to the
“x” symbol.16 If you are using RStudio, check the Environment tab in the upper-
right pane, and see the information you have at a glance. Under the “Values”
group, you have the object x and some information about it: the data type (num),
its length (from index 1 to index 3, i.e., 3 elements), and the first values of the
vector (in this case all of them as there are few of them). You can always access
this information from code either in the console or within a script. The following
expression gets the list of objects in your workspace using the ls function17:

And now you can ask for the structure of the x object with the str function:

If you input the variable symbol as a expression, you get its contents as
output:

When using scripts, creating vectors interactively is not practical. Instead,

vectors are created using the c function, which combines its arguments into a
vector. For example, the following expression is equivalent to the above process:

We can also create vectors using operators and functions to generate

sequences. For example, the seq function generates sequences of numbers, and
the following expression is also valid to create our vector:

Sequences of integers can also be created using the colon operator (“:”)
between the first and last numbers of the intended sequence. For example, the
following expression creates a vector with the integer numbers from 1 to 10:

Notice how in the above code we have typed two expressions in the same
line, but a semicolon was used to separate them. Another useful function to
generate vectors is the rep function, that repeats values. When working with
vectors, it is common practice to combine the different ways of creating vectors:

The sequence of indices along a vector can also be generated with the
seq_alog function:

Check that you have all the five new vectors in your workspace. We have
created numeric and character vectors. Logical vectors can also be created:

Let us study this expression. We are assigning to the “logicalVector” symbol

what we have on the right-hand side of the assignment expression. There, we
first have the vector “1:6”, which is compared to the number “3”. This
comparison is done for all the elements in the vector, and the result is another
vector with the results of those comparisons, and this logical vector is assigned
to the “logicalVector” object:

The TRUE and FALSE values are coerced to 1 and 0, respectively, when trying
to operate with them. This is useful, for example, to get the number of elements
that are true in a logical vector18:

Vectors and Factors

Vectors and factors are different classes in R. But, actually, a factor is a kind of
vector which contains information about the possible values we can find in it
(levels), and the identifying labels for those possible values. For example, we
might have a variable for the machine that operates a given process, being those
machines identified by letters. This identification is the label.

In the above expression, we used the internal object letters, which is

actually a vector with the letters of the alphabet. There is also a LETTERS object,
guess the difference and try them in the console. We can also generate factors for
a given number of replications of each level using the gl function:

Lengths and Names

Vectors (and factors) lengths can be get using the length function. Moreover, we
can assign names to each element of a vector. For example, the following
expression gets the length of our first vector:

If we want to label each element of this vector, for example because the
numbers are for different weeks, we can do so using the names function:

Accessing Vector Items

Data objects in R are indexed, and we can access each element of a vector (or
factor, or any other R object as we will see later) either through its index or
through its name, if such name exists. Vector indices are indicated through the
square brackets symbols (“[ ]”). We can access elements of a vector for either
extracting or replacing their content. For example, the following expression
extracts the third element of the “x1” vector:
 while the following one replaces the content of the third element by the
number 50:

remaining the rest of the items unchanged. We can include integer vectors as
index to select more than one element. For example the following expression
gets the first and third elements of the x1 vector:

We can also exclude elements from the selection instead of specifying the
included elements. Thus, the previous expression is equivalent to this one:

New elements can be added to a vector either creating a new vector with the
original one and the new element(s) or assigning the new element to the index
greater than the last one, for example:

To delete a vector item, we re-assign the vector resulting of the exclusion of

such element:
 If the elements of a vector has names, then the selection can also be done
through such names as follows:

When working with data in R, it is very common to select elements of an

object through logical vectors. Hence, instead of using numerical vectors as
indices, we can use logical vectors of the same length than the vector, and the
result will be a vector with the elements of the original vector whose indices are
TRUE in the logical vector. For example, for the above selection we could make
the following selection using logical indices:

The combination of logical expressions and index selection is what makes

this strategy powerful for data analysis. For example, to get the values of the
vector that are greater than 15, we would use the following expression:

First, the expression x1 > 15 is evaluated, returning the logical vector

{FALSE, TRUE, TRUE}. Then, the selection is done returning only the second and
third elements of the vector, which are the ones that fulfill the condition. See
Appendix C for further logical operators.

Ordering Vectors
Two functions are related with the ordering of vectors. Let us create a random
vector to illustrate them. The following expressions are to get a random sample
of size 10 from the digits 0 to 9. The set.seed function sets the seed in order to
make the example reproducible, see ?RNG to get help about random numbers
generation with R.

The sort function returns the values of the vector ordered:

 The order function returns the indices of the ordered values of the original
vector, i.e., the first element is the index of the minimum value in the original
vector, and so on:

This function is very useful for sorting datasets as we will see later. Both the
order function and the sort function accept a “decreasing” argument to get the
reverse result. In addition, the rev function reverses the order of any vector, for
example, the following expressions are equivalent:

Operating with Vectors

There are two types of operations we can perform over a vector, namely:
Operations over all elements of a vector as a whole. A function is applied
using all the elements in the vector to produce a given result, which can be
a computation, some other values, a plot, etc. For example, to compute the
average of all the elements in vector x1 we can apply the mean function
passing the vector as first argument:

Operations over each element of the vector, resulting on a vector of the

same length with a computation over each value of the vector. For example,
arithmetic operations and some mathematical functions work like that:

At this point, let us introduce one interesting feature of R: recycling. The

first expression in the above chunk of code is a sum of a vector whose length is 3
and another vector whose length is 1. To do that operation, the vector of length
1, i.e., the number 5, is coerced to a vector of length 3 recycling the number 5
twice. If we add a vector of length 2, recycling is also done, but we get a
warning because the length of the first vector is not a multiple of the second one:

In this case, the 5 has been recycled once to complete a 3-length vector.
Mathematical functions which require a single value as argument return vectors
with the result of the function over each value of the original vector. For
example, the sqrt function returns the square root of a number:

Matrices

Creating and Accessing Matrices

A matrix is actually a vector organized in rows and columns. All the elements
must be of the same type. The most common way of creating matrices is through
the matrix function, whose main arguments are: (1) the vector with all elements
of the matrix; (2) the number of rows; and (3) the number of columns. The data
are added by columns, unless the byrow argument is set to TRUE:

We can extract and replace parts of a matrix in the same way as in vectors.
The only difference is that now we have two indices rather than one inside the
squared brackets, separated by a comma. The first one is for the row index, and
the second one is for the column index. We can extract a whole row (column) by
leaving the second (first) index empty:
 Notice that in the Environment tab of the upper-right pane of RStudio, the
matrix is under the “Data” group, instead of the “Values” one. As matrices have
two dimensions, i.e., rows and columns, they can be visualized in the RStudio
data viewer by clicking on the icon on the right of the list. The structure of the
matrix can be also get using the str function. See how now the lengths of the
two dimensions are shown, i.e., four rows and two columns:

Basic Matrix Operations

We can assign names to rows and/or columns of matrices:

Marginal sums and means can be computed using the rowSums, colSums,
rowMeans, and colMeans functions, for example:

See Appendix C for more examples of matrix operations. Arrays of higher

dimensions are possible in R through the array function.

Lists

Creating Lists
Lists are data structures that can contain any other R objects of different types
and lengths. Such objects can be created within the own definition of the list, or
taken from the workspace. The elements of a list can also be named, typically
this is done when creating the list. In the following example, we create a list
whose name is “myList,” and has three components.

See the printing of the list. The first two elements are shown with its name
preceded by a $ symbol. This is because we named them in the list definition.
The third element had no name and it is identified by its index between double
square brackets [[3]].

Accessing Lists
Similarly to vectors, the components of a list are indexed, and we can extract
each element of the list either by its index or by its name. In the latter case, we
can use the $ operator. See the following examples:
 The difference between simple and double squared brackets is that when
using double squared brackets, we get the original object that is within the list, of
its own class, e.g., matrix. On the contrary, if we do the extraction using the
single squared brackets like in vectors, we get an object of class list. This makes
possible to select more than one element in the list, for example:

Notice that we can extract elements from the inner components of a list, for
example a column of the matrix that is the first element of the list. We can also
replace parts of an object as we had done with vectors and matrices:
 You can see the structure of a list in the workspace by looking at the
Environment tab in the upper-right pane of RStudio, or using the str function:

Notice that the structure of a list shows the structure of each element of the
list. In the Environment tab, RStudio upper-right pane, the number of elements
of the list is shown, and by clicking on the left-side icon next to the name of the
list, the list is expanded to show the structure of each element of the list.

Data Frames
The usual way of working with data is by organizing them in rows and columns.
It is common that we have our data in such a way, either from spreadsheets, text
files, or databases. Columns represent variables, which are measured or observed
in a set of items, represented by rows. The class of R objects with such structure
is the data.frame class. We refer to them as data frames hereon. Recall that
matrices are also organized in rows and columns. The difference is that a matrix
can only contain data of the same type, for example numbers or character
strings. However, the columns of a data frame can be of different types, e.g., a
numerical column for the measurement of a quality characteristic, another one
logical stating whether the item is nonconforming, another one a factor for the
machine where the item was produced, and so on.

Creating Data Frames

Normally, we will import data to data frames from files. Nevertheless,
sometimes we need to create data frames from other R objects or by generating
vectors. We create data frames with the function data.frame

Accessing Data Frames

Data frames are actually a sort of combination of lists, matrices, and vectors.
Look at the Environment tab in the upper-right pane of RStudio. Equally to
matrices, data frames are under the “Data” group and can be visualized in the
RStudio data viewer: click on the right icon to show the data as a new tab in the
RStudio source pane, see Fig. 2.12. In that sense, a data frame is a matrix with
rows and columns. On the other hand, the information shown about the data
frame is the number of observations (rows) and the number of variables
(columns). Notice that the expand/collapse icon for list objects explained above
is also next to the data frame name. If you click on it, the structure of each
column is shown. Let us see the structure of the data frame using the str
function:
Fig. 2.12 RStudio data viewer. Matrices and data frames can be visualized in the data viewer. A new tab is
open when clicking the icon right to the object in the Environment tab

Therefore, a data frame is a list of columns, and each column is a vector.

Similarly to lists, we can access data frame columns by names using the $
operator, or by index:

Notice that the access by name is equivalent to the access using double
squared brackets. The difference is whether the result is a data frame or a vector.
As a two dimensional data object, we can also access data frame elements in the
matrix fashion:

Sometimes, we need to get the number of rows or the number of columns of

a data frame to be used in expressions. We can get them with the following
expressions:

Data frames rows and columns have always names. Even if they are not
available when creating the data frame, R assign them: for columns, using the
letter V followed by a number (V1, V2, …); for rows, the default names are row
indices. Rows and column names can be consulted and changed afterwards in
the same way we explained above for factors and vectors, see the following
examples19 (we first create a copy of the data frame):
Ordering, Filtering, and Aggregating Data Frames
We already know that data frame columns are vectors. Therefore we can use the
functions explained for vectors in data frames. For example, to sort the data
frame created above by the weight column, we use the extracting strategy by
means of the squared brackets, passing as row indices the result of the order
function over the column (or columns) of interest:

For filtering (subsetting in R jargon) data frames, in addition to the use of

indexing, we can use the subset function, whose use is more intuitive: the first
argument is the data frame to be subset, and the second one a logical expression
with the condition. Further options can be used, see the documentation of the
function:
 On the other hand, the aggregate function allows us to get subtotals of
numerical variables by categorical variables in a data frame. A special type of
expression is used as the first argument of the function: a formula. A formula is
an expression with two sides, separated by the symbol ~. It is mainly used to
specify models (see Chapter 5) in the form of y ~ model, where y is the response
variable and model can include several independent variables and their
relationship. For aggregating data, the idea is that the y in the formula left-hand
side is the variable that we want to aggregate, and the model are the criteria by
which we want to aggregate the data. For example, if we want to get the sum of
weight by type in our data frame:

where the third argument can be any function over a vector, typically
aggregation functions, see Appendix C.

Editing Data Frames

We use assignment expressions to edit, add, or remove elements of a data frame.
Changing values in a data frame is done in the same way as in vectors or
matrices. For example, to change the third observation of the second column:
 We can add new columns to a data frame as follows. For example, imagine
we want to inspect the items in the data frame at a random order to check the
measurements. Then we add the randomorder column as follows20:

Note that if we do the assignment over an existing column, it is overwritten.

Sometimes this is what we want to do, but some others we are unexpectedly
losing data. To remove a column, we assign the special value NULL to it:
 Computed columns are easy to add to our data frames. The operation is
similar to what we do in spreadsheets with formulas, for example to add values
in columns, or any other operations over data, and then copy the formulas
throughout the rows, and so on. In this case it is more straightforward. Imagine
we want to compute a column with the proportion over the total each item
represents. The following simple expression does that, and it is ready to further
work with it:

Special Data Values

Missing Values
Missing values treatment is a quite important topic in data analysis in general,
and in quality control in particular, especially in early stages of data cleaning.
Missing values are represented in R by the special value NA (not available). If we
try to do computations over vectors that include NAs, for example the mean, we
will get NA as a result, unless the argument na.rm (remove NAs) is set to TRUE.
Such argument is available in a number of functions and methods, but not
always. It may happen that NA values should actually have a value, but it was not
correctly identified when creating the data object. Then we can assign other
values to NAs. For that purpose (and others) the is.na function is very useful.
First, let us create a new column in our data frame to illustrate NAs. Suppose we
measured the content of salt of each element in the data frame in addition to the
weight. Unfortunately, for some reason the measurements could have not be
taken for all of the items. We add this new information as we learnt above:

Let us compute the means of the two numerical variables in the data frame:

There was no problem to compute the mean weight, as all the observations
are available. However, the mean salt could not be computed because there is a
missing value. To overcome this situation, we must tell the mean function to
omit the missing values:

Another possible action over NAs is to assign a value. Let us suppose that the
missing value is due to the fact that the item had no salt at all, i.e., the correct
value should be zero. We can turn all the NAs values into zeros (or any other
value) as follows:

Other Special Values in R

In addition to the NA and NULL values we have seen so far, there are other special
values in R. For example, the Inf value represents the infinity:

Sometimes we get NaNs (not a number) when an operation cannot be done:

The i symbol is used to represent complex numbers:

The following built-in constants are also available:

Data Types Conversion
When creating objects, sometimes data types can be specified somehow, for
example via the creating function arguments. Thus, the data.frame function
accepts the stringsAsFactors argument to determine whether strings should be
created as factors (default) or as character (setting the argument to FALSE). R
tries to figure out what is the best type for a data set. For example, when creating
a vector, if the input data includes only numbers, it creates a numeric vector; if
the input data includes only character strings, it creates a character vector; if the
input data includes both numbers and character strings, it creates a character
vector to preserve all the information: numbers can be converted to strings, but
strings cannot be converted to numbers. We can see the type of data in the vector
using the class function:
 We can also check whether an object is of a given type:

In any case, data structures and types can be converted from one type to
another. For example, if we want vector3 to be a numeric vector, we coerce the
object to numeric:

Note that, as the third element cannot be converted to a number, NA is

introduced by coercion. Functions as.xxx and is.xxx are available for a number
of types and classes, type apropos("^as[.]") for a list.

Working with Dates

Dates and times are important types of data. As was shown in Chapter 1, in
quality control it is important to keep track of the sequential order in which the
data were produced. Dates and times are usually stored in form of character
strings, and they can be expressed in varied formats. For example, suppose we
know the manufacturing date for some of the items in our data frame
(DD/MM/YYYY format):
 We have added a character vector with the dates to the data frame. If we had
included the column when creating the data frame, the column would have
factor class. If we keep this variable as is, all the operations we do with it are
referred to characters. For example, try to sort the data frame by date:

It did not work because the string “13/02/2015” is the first one in a by-
character order. To make R understand that a variable is a date, we need to
convert the character string into a date. As you have likely guess, we do that
with an as.xxx function. But in this case we need an important additional
argument: the format in which the date is stored in the character vector. In the
case at hand, we have a day/month/year format, which must be specified as
follows (we overwrite the date variable):

Note that now the date column is of Date type, and the data is represented in
ISO format, i.e., “YYYY-MM-DD”. The format argument expects a character
string indicating the pattern used in the character strings that store the dates. In
our example, we are specifying that the string is formed by: (1) the day of the
month in decimal format (%d); (2) a forward slash; (3) the month of the year in
decimal format (%m); (4) another forward slash; and (5) the year with century
(%Y). Check the documentation for the strptime topic for more options. Now
we can sort the data frame by date:

It can also be useful to create variables for the year, month, etc. for
aggregation, classification, stratification, or any other purpose. For example, if
we store the week we can plot control charts where the groups are the weeks. We
use the format function in the reverse sense, i.e., we turn dates into character
strings, see the following examples:
2.7 Data Import and Export with R
In the previous section we have created all data from scratch. There are many
situations in which we will use such strategy. However, raw data usually comes
from external sources, either because they are automatically recorded during the
process, or stored in databases, or manually entered in spreadsheets. The easiest
way to import data in R is using.csv files. CSV stands for Comma Separated
Values, and.csv files are text files in which each line corresponds to an
observation of a dataset, and the values for each column are separated by a
comma. Actually, the comma can be substituted by another value, for example
when the comma is used as decimal point, then semicolons are used instead of
commas to separate columns. The main advantage of using.csv files is that they
can be generated by most of the applications that storage data, such as
spreadsheets, databases, etc. Furthermore,.csv files can be opened and edited in
spreadsheets applications such as Microsoft Office or LibreOffice, for which most of
the users are already trained.
In the following, we will explain how to get data into R from.csv files. At the
end of the section, some directions are provided to import data from other
sources. A.csv file is available for downloading from the book’s companion
website.

Importing.csv Files
In manufacturing it is common that PLCs (Programmable Logic Controllers)
record data regarding product quality features. Quite often such recording
machines can automatically generate data in.csv files. In such a case the files are
ready to work with them in R. However, if we are exporting data from
spreadsheets, we must take into account that the resulting file will only contain
text, and what we see on the screen may be different than what we get on the
file. If the data on the file does not correspond with what we want, then formats,
formulas, or other application-specific options might be the cause. Remove all
the formats in numbers and characters. It is also recommended to do the
computations in R rather than using formulas in the spreadsheet. Make sure the
data in each column are consistent, for example you do not use different data
types in the same columns (text and numbers). Once you have your data ready
for exporting, select the “Save as …” menu option of your spreadsheet
application and select the.csv format in the “File type” list. Search the location
where you want to save the file, for example your R working directory, choose a
name, and save the file. Depending on your system locale configuration, the
software usually decides the decimal point symbol and the separator for values.
For example, if your system is in English, the decimal point symbol will be the
period, and the separator, the comma; but if your system is, for example, in
Spanish, then the decimal point symbol will be the comma, and the separator, the
semicolon. These two formats are the most common ones.
For the examples below, you need to download the file http://www.
qualitycontrolwithr.com/lab.csv to your working directory. You can go to your
browser and download it as any other file. Alternatively, you can use the
download.file function21:

Now that you have a.csv file on your working directory, you can explore it.
If you click the file in the Files pane of RStudio, the text file is opened in the
source pane. This format is difficult to manage from a text editor, so take a look
just to see how it looks like, and close it. Before importing the data into R, open
the.csv file with your spreadsheet application, for example Microsoft Excel.
Double-clicking the file in a files explorer window should work, but if it does
not, use the “File / Open …” menu of your spreadsheet application and search
the file. It is possible that the spreadsheet application asks for the format of your
data. If so, just select the period as decimal symbol and the comma as values
separator. See how the data inside the.csv file looks like your usual spreadsheets,
without formats, though. Now you can close the file, from now on we will work
with the data in R.

Importing Data from Text Files

RStudio includes a functionality to import data from files. In the Environment
tab, upper-right pane (see Fig. 2.8), the “Import Dataset” menu has two options:
“From Text File …” and “From Web URL …”. The former opens a dialog box to
search a text file. Select the lab.csv file that you downloaded and click on Open.
A dialog box appears, see Fig. 2.13. In this dialog box, we can see on the right
how the text input file looks like (top), and a preview of the data frame that is to
be created when clicking on the Import button. On the left side we can tune the
import options, which are automatically detected:

Fig. 2.13 RStudio import dataset dialog box. From the Environment tab we can import data in text files
through the Import Dataset menu

The name of the data frame in your workspace;

 Whether the file contains headings, i.e., the first row contains the variables’
names;
The characters that define the separator, the decimal point symbol, and the
text quotes;
The value for empty strings, NA by default;
Whether importing strings as factors (default). Unchecking the box, string
columns are imported as character vectors.
Accept the default settings and click on the Open button. Several things
happen after importing data. Check your workspace in the Environment tab,
upper-right pane. Now you have a data frame named lab in your workspace,
under the “Data” group. Expand the structure using the left icon to see the
variables data types. Moreover, RStudio opens automatically the data viewer to
visualize the data frame in the source pane. On the other hand, take a look to
your console. After importing the data, you should have something similar to
this:

where <your_path> is the path where you downloaded the.csv file, i.e., your
working directory if you used the previous code. The second expression is the
one that opened the data frame in the data viewer. It is just what RStudio does
when clicking on the icon to the right of a data frame or matrix in the
Environment tab. The first expression is the interesting one. Importing files from
the import dataset menu is useful to explore data files, or to import a static data
file once and then save the data processing as specific R data files
(extension.RData). However, the usual way of working is that data files are
regularly updated, either adding rows to the files or adding files to folders. Thus,
it is more efficient to automate the data import in the scripts, and we do that with
the read.csv function above. Note that the only argument that was included was
the file path, as the rest of the default options are valid for standard.csv files as it
is the case. The read.csv function is actually a wrapper of the read.table
function, check the documentation for more details and options.

Data Cleaning
Now we have the data available in our system, but raw data is likely to contain
errors. Before applying the methods described in the following chapters, make
sure that your data are ready for quality control. An exploratory data analysis
should be made to detect possible errors, find outliers, and identify missing
values. Some examples are given below. The first thing we must do is to verify if
the data frame is what we expect to be. Check the number of rows, number of
columns and their type, either in the RStudio environment tab or in the console:

Exploratory Data Analysis

Exploratory data analysis should include descriptive statistics, which is
described in Chapter 5. This exploratory data analysis should also include
techniques to detect errors like the ones presented hereon. Once the data has
been cleaned, exploratory data analysis would continue with statistics,
probability, and plotting techniques. Let us illustrate the data cleaning part of
exploratory data analysis using the data we have imported from the.csv file. For
the purpose of this illustration, we will use the summary function, which will be
further explained in Chapter 5. This function produces result summaries of a
given R object. Applied over a numeric vector, summary statistics are shown: the
minimum, first quartile, median, mean, third quartile, and the maximum. This is
enough for the moment.

Missing Values
Let us get a summary of the ph variable of the lab data frame.

Notice that in addition to the summary statistics mentioned above, we are

also informed about the number of mission values (NA’s). It is strange that in a
dataset of 1259 observations there is one and only one NA. We can look for that
value using the extraction techniques learned in this chapter:
 Let us suppose that we do some research and find out that the measurement
was taken but the operator forgave to record it. We know that the value is 6.6.
Again, using objects’ assignment and replacement we can fix that (note that we
know from the previous output that row number 22 was the wrong one):

and now we have no missing values for the ph variable. It is not always
necessary to assign missing values. For example, for the other two numerical
variables of the data set, there are missing values, but we know from the own
process that ph is always measured, but fat and salt are only measured for some
items. Thus, it is normal having NAs in those columns. We just must be aware
and take that into account when making computations:

In addition to the is.na function, the functions any.na and complete. cases
functions are useful to manage missing values. The former returns TRUE if at least
one of the values of the first argument is NA. The latter gets the row indices of a
data frame whose columns are all not NAs. This can be useful to get only the
rows of a data frame that are complete.
Outliers
Outliers are another type of particular data that must be examined before
applying statistics to quality control. An outlier of a data set is, as defined in
[18], “a member of a small subset of observations that appears to be inconsistent
with the remainder of a given sample.” Outliers or outlying observations can be
typically attributed to one or more of the following causes:
Measurement or recording error;
Contamination;
Incorrect distributional assumption;
Rare observations.
Similarly to missing values, sometimes outliers are correct, or we just cannot
remove them or assign a different value. In such cases, robust techniques should
be applied, see [18] . But in some other cases, the value is either impossible or
extremely unlikely to occur, and it should be corrected or removed. In a dataset
with more variables, this removal means assigning the NA value.
To illustrate outliers, let us go back to the ph variable of our lab data frame.
You have probably already realized that there is a strange number in the
summary:

As you might have guessed, the median and mean are close to 6.6, the
minimum is 6.63, but the maximum is ten times those values. It looks like there
is something inconsistent. There is obviously something wrong with that value.
We can descendently sort the dataset and check the first values to see if there are
more extreme values:
 We see that it is just row number 16 who has the maximum value. After
some investigation, it was detected a wrong recording of the value, the value
should have been 6.63. Note that this is a very common error when recording
data. Again, we can fix the problem as follows:

In addition to the statistics summary, a powerful tool to identify outliers is

the box plot. It will be described in Chapter 5, and a thorough explanation can
also be found in [18] . It is basically the representation of the numbers in the
summary, but all the possible outliers are also identified.

Wrong Values
Finally, other wrong values may arise in the data. It is not always easy to detect
wrong values. For categorical variables, a frequency table is a good way to find
possible errors. Let us get a frequency table for the analyst variable using the
function table. The result is the count of rows for each possible value of the
variable used as argument:

Notice that there is an “analyst4” and an “analyst_4”. The former has only
one count, and the rest are 288 or above. Apparently, “analyst4” and “analyst_4”
are the same person, and we have again a recording error. Unfortunately, these
types of errors are quite common when manually recording data in spreadsheets.
A more difficult to detect error is the one we have in the date column. There
is no value for row 24, but it is not detected as missing value because it was
imported as an empty string rather than a missing value:
 The which function returns the TRUE indices of a logical vector, very useful
to use for data extraction. In this case, we should have created a column of type
Date in advance, and then look for missing data over that column, because the
empty string is coerced to NA:

Let us fix it, supposing the correct date is 2012-01-10:

Exporting Data from R

So far in this section, we have imported data from text files, and clean the data to
get it ready for quality control. From here, we could follow two approaches:
Save the import and data cleaning code and run it again at the beginning of
the quality control data analysis;
Save the clean data in a data file and import the data at the beginning of the
quality control data analysis.
For the first approach, the expressions corresponding to the data import and
replacement are to be saved in a script and then include an expression in the
quality control analysis script to run the script via the source function as
explained in Sect. 2.5. For the second approach, we can save the clean data in a
file and then include an expression in the quality control analysis script to import
the clean data. For that purpose, the counterpart function of the read.csv
function is write.csv. The following expression saves our clean data in the file
lab_clean.csv:

The first argument of the function is the object in the workspace that contains
the data to be exported, preferably a matrix or data frame; The second argument
is the path to the output file; and the third argument avoids to create a column
with the row names, typically the row index unless row names have been set.
Thus, the lab_clean.csv has the same structure as lab.csv but with the wrong
data fixed. In fact, it is the same result as if we had edited the.csv file with a
spreadsheet application such as Microsoft Excel to correct the data and then
saved the file with a new name.
A strategy mixing both approaches is however the most efficient. The
following step-by-step procedure can be followed as a guide when planning
quality control data analysis:

1. Create folders structure for your quality control data analysis project. The
following could be a general proposal which should be adapted to the project
specifics, if any:
data: This folder shall contain the data files. It could contain sub-folders
such as “rawdata” and “cleandata”, “yyyy_mm_dd” (one folder per day),
etc.;
code: This folder shall contain the scripts;
reports: This folder shall contain the.Rmd files and their counterpart
compiled reports as shown in Sect. 1.6, Chapter 1;
plots: This folder could contain the exported plots to be used by other
programs;
other…: Any other folder useful for the purpose of the quality control
data analysis.

2. Save the raw data file;

3. Create a script for data cleaning. This allows to keep track of the changes
made, even including comments in the code for further reference and lessons
learned;

4. Export the clean data in a data file with a new name (included in the data
 cleaning script);

5. Create scripts for the quality control data analysis. There might be several
different scripts, for example for exploratory data analysis, control charts,
capability analysis, etc.;

6. Create report files with the relevant results.

In addition to.csv files, data can be exported to many other formats. For
example, one or more objects in the workspace can be saved in a.RData file
using the save function. The following expression saves the lab data frame in the
lab.RData file:

Later on, the data in a.RData file can be imported to the workspace with the
load function:

It is up to the user which data and file formats to choose for their quality
control data analysis. All of them have advantages and disadvantages.
Depending on the use that will be done over the data, it could be better to use.csv
files, e.g., when the data are bound to be used by other applications, or.RData
files, if only R will make use of them. In addition to.csv and.RData, many other
formats can be used in R for data import and export, see the following
subsection.

Importing Data from Other Sources

Importing files from.RData or text files is the easiest and less prone-error way of
getting data into R. Nevertheless, there are many more ways of importing data
from different sources. Check the “R Data Import/Export” manual enclosed in
the R documentation or at the R project website. The following is a list of the
functions and packages that deal with importing data from common sources,
check their documentation for details if you need to import data from the sources
they manage:
The foreign package [26] can read data coming from the main statistical
 packages, such as Minitab, S, SAS, SPSS, Stata, or Systat, among others;
The RODBC package [28] deals with Open Database Connectivity (ODBC)
sources. It originated on Windows but is also implemented on Linux / Unix
/ OS X. The supported databases include Microsoft SQL Server, Access,
MySQL, PostgreSQL, Oracle, and IBM DB2;
RMySQL [25], RSQLite [37], and RPostgreSQL [9] are the appropriate packages
for their counterpart FOSS database management systems;
ROracle [23] and RJDBC [35] work with Oracle and Java databases,
respectively;
The XML package [21] can make many operations with XML files;
The XLConnect package [22] can read and write Microsoft Excel files
directly22;
Unstructured and distributed databases are also accessible, for example via
the RMongo [7] and h5 [2] packages.
More and more institutions are making their data available on the Internet.
A general approach to deal with those data is to download the source file to disk
as explained above to download the “lab.csv” file and then import the data into
R.

2.8 R Task View for Quality Control (Unofficial)

If there were a Task View for Quality Control at CRAN, it should include the
following resources.
This Task View collects information on R packages for Statistical Quality
Control. Statistical Quality Control applies statistics to process control and
improvement. The main tools used in statistical quality control are control
charts, capability analysis, and acceptance sampling. All statistical tools may be
useful at some point in a quality control planning.

Modeling Quality
The packages in this paragraph are installed with the base installation of R.
The base package contains basic functions to describe the process
variability. The summary function gets a numerical summary of a variable.
The function table returns frequency tables. The functions mean, median,
var, and sd compute the mean, median, variance, and standard deviation of
a sample, respectively. For two variables, we can compute the covariance
 and the correlation with the functions cov and cor, respectively.
The stats package includes functions to work with probability
distributions. The functions for the density/mass function, cumulative
distribution function, quantile function, and random variate generation are
named in the form dxxx, pxxx, qxxx, and rxxx respectively, where xxx
represents a given theoretic distribution, including norm (normal), binom
(binomial), beta, geom (geometric), and so on, see ?Distributions for a
complete list. Linear models can be adjusted using the lm function. Analysis
of Variance (ANOVA) can be done with the anova function. The ts and
arima functions are available for time series analysis.

Visualizing Quality
Standard plots can be easily made with the graphics package. It basically works
as a painter canvas: you can start plotting a simple plot and then add more
details. The graphics, grid, and lattice packages are included in the R base
installation. The grid and lattice packages must be loaded before use, though.
The graphics package allows to build standard plots using the plot (scatter
plots), hist (histograms), barplot (bar plots), boxplot (box plots) functions.
Low-level graphics can also be drawn using the functions: points, lines,
rect (rectangles), text, and polygon. Those functions can also be used to
annotate standard plots. Functions of x can be drawn with the curve
function.
The grid package implements a different way to create and modify plots in
run time, including support for interaction.
The lattice package [32] can plot a number of elegant plots with an
emphasis on multivariate data. It is based in Trellis plots.
ggplot2 is another package [36] providing elegant plots through the
grammar of graphics.
Cause-and-effect diagrams can be drawn with the cause.and.effect (qcc
package [33]) and the ss.ceDiag (SixSigma package [5]) functions.
To make Pareto charts the functions pareto.chart (qcc package),
paretoChart (qualityTools package [29]) and paretochart
(qicharts package [1]) can be used.

Control Charts
The qcc package [33] can perform several types of control charts, including:
 xbar (mean), R (range), S (standard deviation), xbar.one (individual values),
p (proportion), np, c, u (nonconformities), and g (number of non-events
between events). The function qcc plots a control chart of the type specified
in the type argument for the data specified in the data argument. For charts
expecting data in groups, i.e., xbar, R, and S charts, the input data must be
prepared with the function qcc.groups, whose arguments are the vector with
the measurements and the vector with the groups identifiers. For attribute
charts where the size of groups is needed, e.g., p, np, and u, the sizes
argument is mandatory.
The qcc package allows to implement customized control charts, see
demo("p.std.chart").
The functions ewma, cusum, and mqcc in the qcc package are for
exponentially weighted moving average control charts, cumulative sums
control charts, and multivariate control charts, respectively.
The SixSigma package can plot moving range control charts with the ss.cc
function.
The qicharts package provides the qic to plot control charts and run charts
. It has also the trc function for multivariate data run charts.
The IQCC package [3] implements qcc control charts with a focus on Phase I
and Phase II analysis.
The qcr package [12] provides quality control charts and numerical results.
The MSQC package [31] is a toolkit for multivariate process monitoring.
Control Charts Operating Characteristic (OC) curves. The qcc package
oc.curves function draws operating characteristic curves which provide
information about the probability of not detecting a shift in the process.

Capability Analysis
The qcc package process.capability function performs a capability
analysis over a qcc object previously created.
The qualityTools package cp function returns capability indices and charts.
The SixSigma package contains functions to individually get the indices
(ss.ca.cp, ss.ca.cpk, ss.ca.z ). A complete capability analysis including
plots can be done with the ss.ca.study function.
The mpcv package [8] performs multivariate process capability analysis
using the multivariate process capability vector.
 The tolerance package [41] contains functions for calculating tolerance
intervals, useful to set specifications.

Acceptance Sampling
The AcceptanceSampling package [20] provides functionality for creating
and evaluating single, double, and multiple acceptance sampling plans. A
single sampling plan can be obtained with the find.plan function
The acc.samp function in the tolerance package provides an upper bound
on the number of acceptable rejects or nonconformities in a process.
The Dodge package [15] contains functions for acceptance sampling ideas
originated by Dodge [10].

Design of Experiments
Please visit the ExperimentalDesign Task View to see all resources
regarding this topic.

Quality Control Reports

The Sweave function can produce.pdf files from.Rnw files, which can
contain LaTeX and R code.
The knitr package [38–40] can produce.pdf,.html, and.docx files
from.Rmd files, which can contain markdown text and R code.

CRAN Packages
AcceptanceSampling
base
Dodge
edcc
ggplot2
graphics
grid
IQCC
knitr
lattice
mpcv
 MSQC
qcc
qcr
qicharts
qualityTools
SixSigma
spc
spcadjust
stats
tolerance

Books
Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. Statistical
Engineering for Process Improvement, Use R!, vol. 36. Springer, New York
(2012).
Cano, E.L., Moguerza, J.M., Prieto, M.: Quality Control with R. An ISO
Standards Approach, Use R!. Springer, New York (2015).
Dodge, H., Romig, H.: Sampling Inspection Tables, Single and Double
Sampling. John Wiley and Sons (1959)
Montgomery, D.C. Statistical Quality Control, Wiley (2012)

Links
http://www.qualitycontrolwithr.com
http://www.sixsigmawithr.com
http://www.r-qualitytools.org

2.9 ISO Standards and R

This book follows an ISO Standards approach for quality control using R.
The process of creating international standards is explained in Chapter 4 The aim
of this approach is to present the standards relevant to the quality control topics,
such as statistics, control charts, capability analysis, and acceptance sampling. In
this section we reference some ISO Standards related to software and data, as
well as the Certification issue.
ISO Standards and Data
In Sect. 2.6 it was shown how R represents dates in ISO format. In particular,
according to the strptime topic documentation:

The default formats follow the rules of the ISO 8601 international standard
which expresses a day as “2001-02-28” and a time as “14:01:02” using
leading zeroes as here. (The ISO form uses no space to separate dates and
times: R does by default.)

Check ISO 8601 international standard [17] for more details on date and time
data representation. As explained in Sect. 2.6, using the format function and the
% operator, any format can be obtained. The ISOweek package [4] could be useful
if you are in trouble to get weeks in ISO format when using Windows.
As referenced in Sect. 2.7, part 4 of ISO 16269, Statistical interpretation of
data—Part 4: Detection and treatment of outliers [18], “provides detailed
descriptions of sound statistical testing procedures and graphical data analysis
methods for detecting outliers in data obtained from measurement processes. It
recommends sound robust estimation and testing procedures to accommodate the
presence of outliers.”
Some examples in this chapter generated random numbers. ISO 28640 [19],
“Random variate generation methods,” specifies methods for this technique.
Regarding data management and interchange, there are a number of
international standards developed by the ISO/IEC JTC 1/SC 32, check the
available standards in the subcommittee web page23 for further details.
Data is becoming a relevant topic in standardization. Recently, a Big Data
Study Group has been created within the ISO/IEC JTC 1 Technical Committee
(Information Technology).24 Keep updated on this standardization topic if your
quality control data is big.

R Certification
Even though there is not a specific reference to ISO Standards in the R project
documentation regarding the software, we can find in the R website homepage a
link to “R Certification.” There we can find “A Guidance Document for the Use
of R in Regulated Clinical Trial Environments,” a document devoted to
“Regulatory Compliance and Validation Issues.” Even though the document
focuses on the United States Federal Drug Administration (FDA) regulations,
many of the topics covered can be applied to or adopted for other fields. In
particular, the Software Development Life Cycle (SDLC) section, which is also
available in the R Certification web page as a standalone document, represents
“A Description of R’s Development, Testing, Release and Maintenance
Processes,” which can be used for certification processes if needed, for example,
in relation to ISO/IEC 12207 [16], Systems and software engineering—Software
life cycle processes.

References
1. Anhoej, J.: qicharts: quality improvement charts. http://www.CRAN.R-project.org/package=qicharts
(2015). R package version 0.2.0

2. Annau, M.: h5: interface to the ‘HDF5’ library. http://www.CRAN.R-project.org/package=h5 (2015). R

package version 0.9

3. Barbosa, E.P., Barros, F.M.M., de Jesus Goncalves, E., Recchia, D.R.: IQCC: improved quality control
charts. http://www.CRAN.R-project.org/package=IQCC (2014). R package version 0.6

4. Block, U.: Using an algorithm by Hatto von Hatzfeld: ISOweek: week of the year and weekday
according to ISO 8601. http://www.CRAN.R-project.org/package=ISOweek (2011). R package version
0.6-2

5. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six sigma with R. In: Statistical Engineering for Process
Improvement, Use R!, vol. 36. Springer, New York (2012). http://www.springer.com/statistics/book/
978-1-4614-3651-5

6. Chambers, J.M.: Software for data analysis. In: Programming with R. Statistics and Computing.
Springer, Berlin (2008)
[MATH][CrossRef]

7. Chheng, T.: RMongo: MongoDB client for R. http://www.CRAN.R-project.org/package=RMongo

(2013). R package version 0.0.25

8. Ciupke, K.: Multivariate process capability vector based on one-sided model. Qual. Reliab. Eng. Int.
(2014). doi:10.1002/qre.1590. R package version 1.1

9. Conway, J., Eddelbuettel, D., Nishiyama, T., Prayaga, S.K., Tiffin, N.: RPostgreSQL: R interface to the
PostgreSQL database system. http://www.CRAN.R-project.org/package=RPostgreSQL (2013). R
package version 0.4

10. Dodge, H., Romig, H.: Sampling Inspection Tables, Single and Double Sampling. Wiley, New York
(1959)
[MATH]

11. Fellows, I.: Deducer: a data analysis gui for R. J. Stat. Softw. 49(8), 1–15 (2012). http://www.jstatsoft.
org/v49/i08/

12. Flores, M., Naya, S., Fernandez, R.: qcr: quality control and reliability. http://www.CRAN.R-project.
org/package=qcr (2014). R package version 0.1-18

13. Fox, J.: The R commander: a basic statistics graphical user interface to R. J. Stat. Softw. 14(9), 1–42
(2005). http://www.jstatsoft.org/v14/i09
14. Free Software Foundation, Inc.: Free Software Foundation website. http://www.gnu.org (2014)
[Retrieved 2014-07-10]

15. Godfrey, A.J.R., Govindaraju, K.: Dodge: functions for acceptance sampling ideas originated by H.F.
Dodge. http://www.CRAN.R-project.org/package=Dodge (2013). R package version 0.8

16. ISO/IEC JTC 1 – Information Technology: ISO/IEC 12207:2008, Systems and Software Engineering –
Software Life Cycle Processes. ISO – International Organization for Standardization (2008). http://
www.iso.org/iso/catalogue_detail?csnumber=43447

17. ISO TC154 – Processes, Data Elements and Documents in Commerce, Industry and Administration:
ISO 8601 Data Elements and Interchange Formats – Information Interchange – Representation of Dates
and Times. ISO – International Organization for Standardization (2004)

18. ISO TC69/SCS–Secretariat: ISO 16269-4:2010 – Statistical Interpretation of Data – Part 4: Detection
and Treatment of Outliers. Published Standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=
44396 (2010)

19. ISO TC69/SCS–Secretariat: ISO 28640:2010 – Random Variate Generation Methods. Published
Standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=42333 (2015)

20. Kiermeier, A.: Visualizing and assessing acceptance sampling plans: the R package
AcceptanceSampling. J. Stat. Softw. 26(6) (2008). http://www.jstatsoft.org/v26/i06/

21. Lang, D.T., The CRAN Team: XML: tools for parsing and generating XML within R and S-plus. http://
www.CRAN.R-project.org/package=XML (2015). R package version 3.98-1.3

22. Mirai Solutions GmbH: XLConnect: excel connector for R. http://www.CRAN.R-project.org/package=

XLConnect (2015). R package version 0.2-11

23. Mukhin, D., James, D.A., Luciani, J.: ROracle: OCI based oracle database interface for R. http://www.
CRAN.R-project.org/package=ROracle (2014). R package version 1.1.12

24. Murrell, P.: R Graphics, 2nd edn. Chapman & HallCRC, Boca Raton (2011)
[MATH]

25. Ooms, J., James, D., DebRoy, S., Wickham, H., Horner, J.: RMySQL: database interface and MySQL
driver for R. http://www.CRAN.R-project.org/package=RMySQL (2015). R package version 0.10.1

26. R Core Team: Foreign: read data stored by minitab, S, SAS, SPSS, Stata, Systat, Weka, dBase, ….
http://www.CRAN.R-project.org/package=foreign (2015). R package version 0.8-64

27. R Core Team: R: A Language and Environment for Statistical Computing. R Foundation for Statistical
Computing, Vienna (2015). http://www.R-project.org/

28. Ripley, B., Lapsley, M.: RODBC: ODBC database access. http://www.CRAN.R-project.org/package=
RODBC (2015). R package version 1.3-12

29. Roth, T.: qualityTools: statistics in quality science. http://www.r-qualitytools.org (2012). R package
version 1.54 http://www.r-qualitytools.org

30. RStudio Team: RStudio: Integrated Development Environment for R. RStudio Inc., Boston, MA
(2012). http://www.rstudio.com/
31. Santos-Fernández, E.: Multivariate Statistical Quality Control Using R, vol. 14. Springer, Berlin
(2013). http://www.springer.com/statistics/computational+statistics/book/978-1-4614-5452-6

32. Sarkar, D.: Lattice: Multivariate Data Visualization with R. Springer, New York (2008). http://www.
lmdvr.r-forge.r-project.org. ISBN 978-0-387-75968-5

33. Scrucca, L.: qcc: an r package for quality control charting and statistical process control. R News 4/1,
11–17 (2004). http://www.CRAN.R-project.org/doc/Rnews/

34. Shewhart, W.: Economic Control of Quality in Manufactured Products. Van Nostrom, New York (1931)

35. Urbanek, S.: RJDBC: provides access to databases through the JDBC interface. http://www.CRAN.R-
project.org/package=RJDBC (2014). R package version 0.2-5

36. Wickham, H.: ggplot2: elegant graphics for data analysis. In: Use R! Springer, Berlin (2009)

37. Wickham, H., James, D.A., Falcon, S.: RSQLite: SQLite interface for R. http://www.CRAN.R-project.
org/package=RSQLite (2014). R package version 1.0.0

38. Xie, Y.: Dynamic Documents with R and Knitr. Chapman and Hall/CRC, Boca Raton, FL (2013).
http://www.yihui.name/knitr/. ISBN 978-1482203530

39. Xie, Y.: knitr: a comprehensive tool for reproducible research in R. In: Stodden, V., Leisch, F., Peng,
R.D. (eds.) Implementing Reproducible Computational Research. Chapman and Hall/CRC, Boca Raton
(2014). http://www.crcpress.com/product/isbn/9781466561595. ISBN 978-1466561595

40. Xie, Y.: knitr: a general-purpose package for dynamic report generation in R. http://www.yihui.name/
knitr/ (2015). R package version 1.10.5

41. Young, D.S.: tolerance: an R package for estimating tolerance intervals. J. Stat. Softw. 36(5), 1–39
(2010). http://www.jstatsoft.org/v36/i05/

Footnotes
1 http://www.r-project.org.

2 http://www.rstudio.com.

3 Please note that sometimes what you paste could not be exactly what you see in the book and some
modifications could be needed.

4 http://www.qualitycontrolwithr.com.

5 http://www.rstudio.com.

6 http://www.ess.r-project.org.

7 http://www.walware.de/goto/statet.

8 Escaping means to provide a character string with special characters. For example, ∖n is for the special
character new line.

9 http://www.java.com.

10 The version used while writing this book was 0.99.xxx.

11 It can be changed through the Tools > Global options menu.

12 This means that only the first letters of the argument name can be provided. We do not recommend that,
though.

13 This function returns the structure of any R object.

14 Pressing the F1 key when the cursor is in a function name also works.

15 In this case not all the code is shown because seq is a built-in, compiled function.

16 The scan function also accepts arguments to scan data from files and text, check the function
documentation.

17 Note that the output might have more elements if further objects were created beforehand.

18 The sum function will be explained later.

19 We use the paste function to get a sequence of character strings, see Appendix C to see more functions
to work with strings.

20 We first fix the seed in order to make the example reproducible.

21 We use a different URL within the download.file function as it fails in redirecting URLs.

22 There are more packages able to deal with Excel files, check http://www.thertrader.com/2014/02/11/a-
million-ways-to-connect-r-and-excel/.

23 http://www.iso.org/iso/home/store/catalogue_tc/catalogue_tc_browse.htm?commid=45342.

24 http://www.jtc1bigdatasg.nist.gov.

© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_3

3. The Seven Quality Control Tools in a

Nutshell: R and ISO Approaches
Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
The aim of this chapter is to smoothly introduce the reader to Quality Control
techniques from the so-called Seven Basic Quality Control tools: Cause-and-
effect-diagram, check sheet, control chart, histogram, Pareto chart, scatter
diagram, and stratification. These are basic but powerful tools when used wisely.

3.1 Origin
Kaoru Ishikawa is one of the main Japanese figures in the quality area. He
chose a set of seven very simple tools, which as such a set constitute an
improvement methodology [5]. Using these tools, most standard quality and
productivity problems become affordable, see [19]. The seven quality Ishikawa
tools are:

1. Cause-and-effect diagram1

2.
Check sheet

3. Control chart

4. Histogram

5. Pareto chart

6. Scatter diagram

7. Stratification2

Next, we provide brief R examples of the use of each tool.

3.2 Cause-and-Effect Diagram

The cause-and-effect diagram, also known as Ishikawa diagram, or
“fishbone” diagram is used to identify the causes of a “problem” (referred in the
quality control literature as “effect”). Usually, once a problem has been detected,
in order to mitigate its effects, the problem causes are to be identified. Often, at a
first stage, some groups of causes are identified, and then, at a second stage,
concrete causes within each group are detected. A way to begin is to use the five
M’s method as groups of causes, namely Manpower, Materials, Machines,
Methods, and Measurements. Sometimes a sixth M is used referring to Mother
Nature, and in some cases, even two more M’s are considered, Money and
Maintenance. In fact, ISO 13053-2 Standard [18] recommends to use the 5Ms+E
(Environment), see Sect. 3.9 for more on ISO standards for the seven Ishikawa
quality tools.
Example 3.1.
Pellets density.
Let us consider the introductory example in Chapter 1, in which three
measurements of pellets density were out of control. You can also see the control
chart of the process later in this chapter, Sect. 3.4. The Ishikawa diagram could
be used in this context to find out the reason of such out-of-control state (the
effect ). Interviews with operators, brainstorming meetings, checking of records
or any other method can be used to determine the possible causes for this
problem. The following list is a suitable example for this situation:
Manpower
– Receptionist
– Recording Operator
– Storage operators
Materials
– Supplier
– Transport agency
– Packing
Machines
– Compressor
– Operation conditions
– Machine adjustment
Methods
– Raw materials reception
– Transport method
Measurements
– Recording method
– Measurement apparatus
The above list can be graphically represented in R with the qcc and the
SixSigma packages. First, let us save the data into R character vectors. In this
way, we only have to type them once. This is useful not only for the two cause-
and-effect diagrams below, but for further purposes as we will see later.
 The following code produces the Ishikawa diagram in Fig. 3.1 using the
cause.and.effect function of the qcc package [23].

Fig. 3.1 Cause-and-effect diagram for the intuitive example. A horizontal straight line is drawn and the
cause is put on the right side. Then, lines for the groups stem from the center line, looking like a fishbone.
The possible causes within each group are finally printed besides those lines

A more elaborated visualization can be produced with the ss.ceDiag function

in the SixSigma package [3] through the following code, see Fig. 3.2:

Fig. 3.2 Cause-and-effect diagram for the intuitive example. The SixSigma package produces a more
elaborated diagram

 □ 

3.3 Check Sheet

The check sheet is a very important tool for the compilation of data. This tool
allows obtaining data from people whose knowledge about the project at hand is
thorough. In practice, although most data are automatically recovered, some data
are compiled by hand in order to be analyzed later on. A straightforward
example is the compilation of stopping times in production machinery.
The compilation of data using check sheets and its subsequent manipulation
should be reflected in the quality improvement plan. In this regard, the check
sheets must be adequately treated by introducing its data into a database or a
spreadsheet, in order to be analyzed at a subsequent step. Check sheets should be
part of the data collection plan, see ISO 13053-2 [18].
There are as many check sheet models as possible problems. The kind of
data to be introduced in a check sheet may be very different depending on the
problem at hand: from a simple event to a group of variables. It is important
always to include dates, machine operator and operator taking the information if
different.
Let us focus on the simplest and most widely used check sheets type, i.e., the
so-called tally sheet, in which an operator makes tick marks whenever a given
event occurs. This event can be a machine stop, error, interruption, or any other
event relevant for the process. The aim is to record a count of the times the event
occur for each identified category, e.g., what cause the error. Note that such
categories might have been identified in advance using the first essential quality
control described in this chapter: the cause-and-effect diagram. As remarked
above, it is also important to record any additional information for future data
analysis. For example, when recording defects on parts, the check sheet might
include room to identify the location of the defect within the part, e.g., top,
bottom, center, and so on.

Example 3.2.
Pellets density (cont.)
Although a check sheet can be designed with text processing or spreadsheet
software, we will use R to produce a check sheet with the data we have from the
previous examples. This has the advantage of reusing data, which reduces errors
and is part of the reproducible research approach explained in Chapter 1. Recall
from the previous section that we saved the causes and the effect in character
vectors in order to use them when necessary. Now we consolidate all those
vectors in a data frame:
 The rbind function binds rows of data frames with identical columns to
create larger data frames. Additional columns can be added to register events by
different criteria. For example: a column for each day of the week and then
analyze the check sheets weekly; a column for the machine where the pellet was
compacted; etc. In this example, let us suppose that the raw material could come
from three different suppliers, and we want to count the point out of control by
each supplier. Then we need to add three columns to the data frame that will
form the check sheet. For the moment, NA values can be assigned, see Chapter 2
to find out more about NAs.

Examine the data_checkSheet object in your workspace. Now we have a data

structure in rows and columns that could be exported to a.csv file, and opened in
a spreadsheet application to print the check sheet, see Chapter 2 to find out more
about csv files. However, our approach is to generate the check sheet from R
itself. Thus, we can create an R Markdown file in RStudio to generate the check
sheet. The following text is a complete R Markdown file to produce a.html file.
---
title: "Out of control pellets density check sheet"
author: "Quality Control Department"
date: "30/06/2015"
output: html_document
---
Instructions: Mark ticks for the more likely cause
of the out-of-control point. Cross every four ticks
to make five.
‘‘‘{r, echo=FALSE, results=’asis’}
 source("checksheet_data.R")
library(xtable)
print(xtable(data_checkSheet), type = "HTML",
html.table.attributes =
"border=1 width=100% cellpadding=10")
‘‘‘
|Week|Operator|Signature|
|----|--------|---------|
| | | |
We use the xtable package [4] to generate tables in.html and.pdf reports.
Notice that the checksheet_data.R file is sourced. It contains the code to create
the vectors and data frame above. Fig. 3.3 shows the check sheet in the RStudio
viewer after knitting the R Markdown file. You can click on the “Open in
Browser” button and print the check sheet for the operator. A filled-out check
sheet could be that in Fig. 3.4, we will use it to illustrate Pareto charts in
Sect. 3.6. □ 

Fig. 3.3 R Markdown check sheet. We can produce tables in reports and in this way a check sheet is easy
to generate
Fig. 3.4 Filled check sheet. An operator registers the events per potential cause with simple tick marks

3.4 Control Chart

The control charts underlying idea is that measurements corresponding to an in-
control process live between some natural limits (referred as control limits).
When measurements go out of these limits, the cause of this especial variability
should be explored and eliminated, see Chapter 1 for an introduction. As we will
explain in Chapter 9, where we will treat control charts in detail, it is important
not to confuse control limits with specification limits.

Example 3.3.
Pellets density (cont.)
In our illustrative example, the control chart for pellets density
measurements plotted in Chapter 1 is reproduced in Fig. 3.5 in order to make this
chapter self-contained. To make such control chart, we need to have the data in
an R object as follows:
Fig. 3.5 Control chart tool. The control chart advances possible problems before they reach the customer

and then plot the control chart for individual values using the qcc package as
follows:
 A summary of the object shows basic information. The assigned object is a
list whose content can be used for further analysis, for example to get the out-of-
control points:

In this case, three points are beyond the control limits, namely measurements
number 11, 12, 13. This means that they are highly unlikely to occur, and
therefore the cause of this situation should be investigated. This investigation
might result in a cause-and-effect diagram as explained in Sect. 3.2. □ 

3.5 Histogram
A histogram provides an idea of the statistical distribution of the process data,
that is, whether the data are centered or not, or even if the data are concentrated
around the mean or sparse. To build a histogram, at a first stage some intervals
are calculated and then, at a second stage, the number of observations inside
each interval has to be accounted. This number of observations is known as the
“frequency.” Finally, the frequencies are represented using vertical adjacent bars.
The area of each bar is proportional to its frequency, being its width the length of
the corresponding interval.

Example 3.4.
Pellets density (cont.)
Basic histograms can be generated with short R expressions and the standard
graphics. For example, the histogram of the pellets density in the illustrative
example explained in Chapter 1 shown in Fig. 3.6 is obtained with the following
simple expression:

Fig. 3.6 Pellets density basic histogram. A basic histogram is generated just with the hist function with
a numerical vector as argument

Notice that histograms generated with statistical software are usually built
with constant width intervals, being the height of the bars for the frequency
(absolute or relative) of data points within each interval. The number and width
of intervals are decided by the software using one of the accepted rules, see
Chapter 5 for details.
The R Core, see Chapter 1, is very concerned about making possible to draw
plots with few options. This allows to generate plots with such short expressions
as the one that generated the extremely simple histogram in Fig. 3.6.
Nevertheless, one of the strengths of R is its graphical capability. Thus, more
elements and styles can be added to the histogram just with the graphics base
package, see, for example, the following code that generates a more elaborated
version of our histogram, shown in Fig. 3.7:

Fig. 3.7 A histogram with options. Graphical options can be added to the hist function to generate nicer
histograms

See the documentation for the topics hist, plot, and par to find out more
about those and other options. Check also how to include expressions through
the expression function. On the other hand, there are some packages specialized
in elegant graphics that use specific syntax. The lattice package [22] is
included in the R base installation, but it is not loaded at the start-up. The
histogram function is the one to generate lattice-based histograms. The panel
argument accepts a number of functions to make complex plots. For example, in
the following code we add a density line to the histogram, see Fig. 3.8 for the
result:
Fig. 3.8 A lattice-based histogram. A density line has been added to the histogram

Consult the lattice package documentation and its functions histogram and
xyplot for graphical options, and trellis.par.set, trellis.par.get for themes and
styles.
The ggplot2 package [24] can also plot histograms using the grammar of
graphics, see [24]. The following code creates the histogram in Fig. 3.9.3 On the
other hand, the special grammar of ggplot2 builds the chart by adding
components with the “+” operator.
Fig. 3.9 A ggplot2-based histogram. In this function, the number of bars is determined by the binwidth
argument, which sets the interval width to create the frequency table

Note that the number of bars in a histogram is an arbitrary decision. Different

functions use different rules. All of them are correct, and they can be adjusted
via the functions’ arguments. The key part is that the histogram should tell us
something about our process. If the default options are meaningless, try to
change the criteria for constructing the histogram. □ 

3.6 Pareto Chart

Cause-and-effect diagrams and check sheets data are usually represented using
Pareto charts. These types of charts plot sorted bars (from the highest bar to the
shortest bar) representing the variable measured, in this case, the counts for
plausible causes of an effect. Pareto analysis, based on the Pareto principle or
80-20 rule, consists of identifying those few vital causes (20 %) producing the
main part of the problem (80 %), in order to avoid assigning resources to the
many trivial causes (the remaining 80 %). This kind of analysis can also be used
to prioritize, for instance, those improvement projects leading to the largest
savings.
There are several alternatives to build Pareto charts in R. The simplest one
would be to create a data frame or table with the data, sort the observations, and
call the barplot function of the base graphics package.

Example 3.5.
Pellets density (cont.)
Let us illustrate Pareto charts with the example we used in sections 3.2 and
3.3. The data gathered by the process owner in the check sheet shown in Fig. 3.4
can be saved into an R data frame with the following code. Note how we are re-
using the data recorded in the previous tools.

Now we have all data in the same place:

 A simple bar plot for the total variable in the data frame could be the one in
Fig. 3.10 using this short expression:
Fig. 3.10 A simple barplot. A simple barplot is not useful for Pareto Analysis

This bar plot is useless for Pareto analysis. The following changes result in
Fig. 3.11, which is an actual Pareto Chart. Note that the options have been tuned
up to improve the readability of the plot, check the documentation of the par
function.
Fig. 3.11 A basic Pareto chart. Simply sorting the bars and reorganizing the axis information a simple
bar plot becomes a useful Pareto chart

Even though we can use standard plots for Pareto charts, there are some
functions in contributed packages that can be useful. The pareto.chart in the
qcc package returns the plot and a table containing the descriptive statistics used
to draw the Pareto chart. This table can be stored in an R object for further use.
The data input should be a named vector, so we first prepare our data. The result
is the chart in Fig. 3.12.
Fig. 3.12 Pareto chart with the qcc package. A dot and line plot is plotted with the cumulative
percentage
 The qualityTools package [21] also includes a function for Pareto charts,
namely paretoChart. It also uses a named vector and returns a frequency table,
see Fig. 3.13.
Fig. 3.13 Pareto chart with the qualityTools package. The table below the chart can be removed by
setting the showTable argument to FALSE
 Another option is the paretochart function in the qicharts package [1]. This
function expects a factor or character vector to make the counts by itself. We can
easily create this data structure with the rep function, see Chapter 2. The
following code produces Fig. 3.14:
Fig. 3.14 Pareto chart with the qicharts package. The output shows the frequency table used to plot the
chart
 Note that when using specific functions we usually lose control over the
graphics. Such specific functions are often convenient, but we may need
something different in the output for our quality control report. For example, we
could split and color the bars in Fig. 3.11 according to the suppliers information
in the check sheet. On the other hand, adding lines and points with the
cumulative percentages in the Pareto chart in Fig. 3.11 is straightforward with
the points function. Moreover, customizing packages’ functions is also possible
as the source code is available. □ 

3.7 Scatter Plot

The scatter plot (or scatter diagram) is used to discover relations between
variables. Once potential cause-and-effect relations are identified, they should be
validated through an experimental design. Consider again the pellets example.
Imagine that, in addition to the density, the temperature of the product is also
available. In order to check the relation between both variables a scatter plot can
be used. This plot is a two dimensional graph where one variable is represented
in the horizontal axis and the other one in the vertical axis. In this way, each
point in the scatter plot represents the value of the pair of variables measured for
each item.

Example 3.6.
Pellets density (cont.)
To illustrate the example, we simulate the temperature with the following
code:

In this simulation, we have added random noise, i.e., values of a normal

standard distribution, with the rnorm function (see Chapter 5 for more about
probability distributions), fixing the seed to an arbitrary value to make the
example reproducible (see Chapter 6 for more references about random number
generation). Now we have two variables: density and temperature, and we can
check with the scatter plot if there is some relation between them. We use the
generic plot function to generate scatter plots, the following code produces
Fig. 3.15:

Fig. 3.15 Scatter plot example. Relations between variables can be found by using scatter plots. Cause-
and-effect relations must be validated through designed experiments, though

The input of the plot function can be a formula, as in this example, where the
left-hand expression is for the response variable (vertical axis) and the right-
hand expression is for the predictive variable (horizontal axis). Similarly to
histograms and other plots, scatter plots can be also generated with the lattice
and ggplot2 packages, check the functions xyplot and geom_point, respectively.
In this example,4 it is apparent that when one variable increases, the other
one also grows in magnitude. Further investigation is usually needed to
demonstrate the cause-and-effect relationship, and to eventually set the optimal
values of the factors for the process optimization.  □ 

3.8 Stratification
In many cases, some numerical variables may have been measured for
different groups (also referred to as factors). When this information is available,
the quality control analysis should be made by these groups (strata). We illustrate
the stratification strategy, by means of the box plot, one of the most useful
graphical tools that will be described in detail in Chapter 5. Using the same scale
in the vertical axis, the data distribution for each factor can be visualized at a
glance. It is important to register as much information as possible about the
factors which data can be split into (operator, machine, laboratory, etc.).
Otherwise, masking effects or mixtures of populations may take place, slowing
down the detection of possible problems.

Example 3.7.
Pellets density (cont.)
In the case of the example we have been using throughout this chapter, let us
assume that the observations of the density measurements correspond to the
three suppliers A, B, and C:

Now we have a numerical variable (density) and a categorical variable

(supplier) in which our data can be grouped. Now we can make stratified
analysis, for example to check if there are differences among the groups. The
above-mentioned box plot by supplier can be easily plotted with R as follows,
see the result in Fig. 3.16. The counterpart functions in the lattice and ggplot2
packages are bwplot and geom_boxplot.
Fig. 3.16 Stratified box plots. Box plots by groups provide quick insights about the differences between
groups, regarding both central tendency and variability

 □ 

Actually, stratification is a strategy that is used throughout the rest of the seven
basic quality tools, and in the application of any other statistical technique in
quality control. For example, it was stratification what we did when designing
the check sheet to gather information by supplier. Stratification can also be
applied to histograms, scatter plots, or Pareto charts. We will see in Chapter 6
that stratified sampling is also a way to improve our estimations and predictions
about the process.
The seven basic quality tools is a topic covered by a number of authors, as it
is an effective and easy to implement problem-solving technique, even being
included in the Project Management Base of Knowledge (PMBoK) [20]. In this
regard, most of the lists keep the six previous tools. However, some of them
replace “stratification” by “run chart” or “flow chart.” A run chart is actually a
simplified version of the control chart, where data points are plotted sequentially.
Flow charts and similar diagrams such as process maps are in fact a previous-to-
stratification step. Those problem-structuring tools allow to divide the process
into steps or sub-processes, and identify the different factors that could influence
the output, thereby defining the groups in which perform the stratified analysis.
A detailed explanation of process maps and how to get them with R can be found
in [3].

3.9 ISO Standards for the Seven Basic Quality Control

Tools
The cause-and-effect diagram is one of the tools in the Six Sigma quality
improvement methodology. There is a subcommittee devoted to this
methodology within the ISO/TC 69 that has developed the ISO 13053 Series,
Quantitative methods in process improvement – Six Sigma. According to part 1
of ISO 13053, DMAIC methodology [17], the cause-and-effect diagram should
be one of the outputs in the Analyze phase of the DMAIC5 methodology. In this
part, the cause-and-effect diagram is also included in the typical training agendas
for both SixSigma black belts and green belts, see [2] for a brief introduction on
the Six Sigma methodology. On the other hand, in part 2 of ISO 13053, Tools
and techniques [18] , a complete factsheet for the tool can be found, pointing to
[6] as a key reference. This standard relates the cause-and-effect diagram with
the brainstorming tool as a possible input. On the other hand, ISO/IEC 31010
[7], Risk management – Risk assessment techniques, includes the cause-and-
effect diagram as one of the tools to be used in root cause analysis (RCA), as
well as in the cause-and-effect analysis, both of them being part of the risk
assessment techniques covered by that standard.
Regarding check sheets , they should be part of the Data collection plan,
also included in ISO 13053-2 [18] as a DMAIC methodology tool. You could
also check clause 7 (data collection) of ISO/IEC 19795-1 [8]. Pareto charts and
Pareto analysis are also included as Six Sigma tools in ISO 13053 series.
Histograms are defined in ISO 3534-1 [9] , Statistics – Vocabulary and
symbols – Part 1: General statistical terms and terms used in probability. This
standard “defines general statistical terms and terms used in probability which
may be used in the drafting of other International Standards. In addition, it
defines symbols for a limited number of those terms”.
There is a series of standards for control charts, developed by ISO/TC 69
SC 4. The following parts have been already published at the time this is
written6:
ISO 7870-1:2014 [15] , Control charts – Part 1: General guidelines. It
presents key elements and philosophy of the control chart approach;
ISO 7870-2:2013 [14] , Control charts – Part 2: Shewhart control charts. It
is a guide to the use and understanding of the Shewhart control chart
 approach to processes’ statistical control;
ISO 7870-3:2012 [13] , Control charts – Part 3: Acceptance control charts.
This part gives guidance on the uses of acceptance control charts and
establishes general procedures for determining sample sizes, action limits
and decision criteria;
ISO 7870-4:2011 [12] , Control charts – Part 4: Cumulative sum charts.
This part provides statistical procedures for setting up cumulative sum
(cusum) schemes for process and quality control using variables (measured)
and attribute data;
ISO 7870-5:2014 [16] , Control charts – Part 5: Specialized control charts.
Specialized control charts should be used in situations where commonly
used Shewhart control chart approach to the methods of statistical control
of a process may either be not applicable or less efficient in detecting
unnatural patterns of variation of the process;
Part 6 of the 7870 series is in preparation for Exponentially Weighted
Moving Average (EWMA) control charts, which will be likely already published
when you are reading this chapter.7
Stratification is defined in ISO 3534-1 [9] , and then this definition is used
in other ones to bound the use of some techniques such as sampling, e.g. in ISO
3534-4, Survey Sampling [10] . As a crossing topic, stratification can also appear
in different ISO standards to apply in other tools and techniques. For example, in
ISO 13053-2 [18] , stratified data collection is needed, and descriptive statistics
visualization may involve stratifying by levels of a factor.
Finally, ISO 11462-2 [11] is a catalogue of tools and techniques for
Statistical Process Control (SPC) that includes all the 7 basic quality control
tools in such a catalogue. There you can find a short description, application, and
references (including related ISO Standards).

References
1. Anhoej, J.: Qicharts: quality improvement charts, url http://CRAN.R-project.org/package=qicharts. R
package version 0.2.0 (2015)

2. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six sigma in a nutshell. In: Six Sigma with R, Use R!, vol.
36, pp. 3–13. Springer, New York (2012). doi:10.1007/978-1-4614-3652-2_1. url http://dx.doi.org/10.
1007/978-1-4614-3652-2_1

3. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six sigma with R. Statistical Engineering for Process
Improvement, Use R!, vol. 36. Springer, New York (2012). url http://www.springer.com/statistics/book/
 978-1-4614-3651-5

4. Dahl, D.B.: Xtable: export tables to LaTeX or HTML, url http://CRAN.R-project.org/package=xtable.

R package version 1.7-4 (2014)

5. Ishikawa, K.: What is total quality control? The Japanese way. Prentice Hall Business Classics.
Prentice-Hall, Englewood Cliffs (1985)

6. Ishikawa, K.: Guide to Quality Control. Asian Productivity Organisation, Tokyo (1991)

7. ISO: ISO/IEC 31010:2009, Risk management – Risk assessment techniques. International standard
(2010)

8. ISO: ISO/IEC 19795-1:2006, Information technology – Biometric performance testing and reporting –
Part 1: Principles and framework. International standard (2016)

9. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). url http://
www.iso.org/iso/catalogue_detail.htm?csnumber=40145

10. ISO TC69/SC1–Terminology and Symbols: ISO 3534-4:2014 - Statistics – Vocabulary and symbols –
Part 4: Survey sampling. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=56154

11. ISO TC69/SC4–Applications of statistical methods in process management: ISO 11462-1:2010 -

Guidelines for implementation of statistical process control (SPC) – Part 2: Catalogue of tools and
techniques. Published standard (2010). url http://www.iso.org/iso/home/store/catalogue_tc/catalogue_
detail.htm?csnumber=42719

12. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-4:2011 - Control
charts – Part 4: Cumulative sum charts. Published standard (2011). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40176

13. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-3:2012 - Control
charts – Part 3: Acceptance control charts. Published standard (2012). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40175

14. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 - Control
charts – Part 2: Shewhart control charts. Published standard (2013). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40174

15. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-1:2014 - Control
charts – Part 1: General guidelines. Published standard (2014). url http://www.iso.org/iso/catalogue_
detail.htm?csnumber=62649

16. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-5:2014 - Control
charts – Part 5: Specialized control charts. Published standard (2014). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40177

17. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six Sigma:
ISO 13053-1:2011 - Quantitative methods in process improvement – Six Sigma – Part 1: DMAIC
methodology. Published standard (2011). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=
52901
18.
 ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six Sigma:
ISO 13053-2:2011 - Quantitative methods in process improvement – Six Sigma – Part 2: Tools and
techniques. Published standard (2011). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=
52902

19. Kume, H.: Statistical Methods for Quality Improvement. The Association for Overseas Technical
Scholarships, Tokyo (1985)

20. PMI: A guide to the projet management body of knowledge. Project Management Institute (PMI),
Newton Square (2013)

21. Roth, T.: Qualitytools: Statistics in Quality Science, url http://www.r-qualitytools.org. R package
version 1.54 (2012)

22. Sarkar, D.: Lattice: Multivariate Data Visualization with R. Springer, New York (2008). url http://
lmdvr.r-forge.r-project.org. ISBN 978-0-387-75968-5

23. Scrucca, L.: Qcc: an R package for quality control charting and statistical process control. R News 4/1,
11–17 (2004). url http://CRAN.R-project.org/doc/Rnews/

24. Wickham, H.: Ggplot2: Elegant Graphics for Data Analysis. Use R!. Springer, New York (2009)

Footnotes
1 Also known as Ishikawa diagram or “fishbone” diagram.

2 Some authors replace “stratification” by “flow chart” or “run chart” in the original list.

3 The ggplot function requires a data frame object as input data, so the pdensity vector is converted to a
data frame in the first argument.

4 As expected from the simulation!

5 Define, Measure, Analyze, Improve, and Control.

6 Descriptions are from the standards summaries at the ISO website http://www.iso.org.

7 At the time of writing, the development stage is Final Draft International Standard (FDIS) , see Chapter 4
 to find out more about standards development stages.

© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_4

4. R and the ISO Standards for Quality

Control
Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
This chapter details the way ISO international standards for quality control are
developed. Quality Control starts with Quality, and standardization is crucial to
deliver products and services where quality satisfies final users, whatever they
are customers, organizations, or public bodies. The development process, carried
out by Technical Committees (TCs), entails a kind of path until the standard is
finally adopted, including several types of intermediate deliverables. The work
of such TCs is outlined along with the general structure of ISO, and with a focus
on the TC in charge of statistical methods. Finally, the current and potential role
that R can play, not only as statistical software, but also as programming
language, is shown.

4.1 ISO Members and Technical Committees

The International Organization for Standardization (ISO) is an independent, non-
governmental membership organization and the world’s largest developer of
voluntary International Standards. Note that ISO is not a proper acronym.
Actually, ISO founders decided to give it the short form ISO from the Greek
isos, whose meaning is equal. Thus, whatever the country, whatever the
language, the International Organization for Standardization is always ISO.
ISO is a network of national standards bodies. Each member represents ISO
in its country, and there is only one member per country. At the time this book is
being written, 163 national bodies are members of ISO.1 For example, ANSI is
the USA member, BSI the UK member, AENOR the Spanish member, etc. ISO
central secretariat is in Geneva, Switzerland, and the ISO Council governs the
operations of ISO.
ISO technical committee structure is managed by the Technical Management
Board (TMB) . Its role is contained in the ISO statutes, including the following:
“Technical committees shall be established by the Technical Management Board
and shall work under its authority. The TMB deals with appeals in accordance
with the ISO/IEC Directives, Part 1”. The TMB also approves the programme of
work for each TC. The TMB reports to the ISO Council.
A technical committee works in a specific field of technical activity. New
technical committees can be proposed by different stakeholders, such as national
bodies or existing technical committees, among others. The main duty of
technical committees is the development and maintenance of International
Standards, but they can also publish other types of deliverables, namely:
Technical Specifications (TS), Publicly Available Specifications (PAS) or
Technical Reports (TR). Technical committees are defined by a number, a title,
and a scope. The scope precisely defines the limits of the work of a TC. Within a
TC, subcommittees (SC) can be established, upon certain conditions, see
ISO/IEC Directives Part 1 [95] for details. Subcommittees also have a number,
title and scope, and their structure and procedures are similar to TCs. Technical
committees and subcommittees are organized as follows:
Secretariat, allocated to a national body. An individual is to be appointed
as secretary. Provides technical and administrative services to its TC or SC
and ensures that the ISO/IEC Directives and the decisions of the technical
management board are followed. It is responsible for monitoring, reporting,
and ensuring active progress of the TC or SC work.
Chair, nominated by the secretariat. Responsible of the TC or SC overall
management.
Working Groups (WG), established by a TC or SC for specific tasks, for
example for preparing working drafts (WD). A restricted number of experts,
appointed by but independent of a national body, compose a WG.
Other structures are editing committees, for the purpose of updating and
editing drafts at different stages, and project committees, established by the
TMB to prepare standards out of the scope of existing TCs and SCs.
National bodies participate in the work of technical committees. For each
technical committee or subcommittee, each national body can have two different
roles, which must be clearly indicated. Thus, one of the following roles can be
assumed:
P-member (participating country), if the national body intends to
participate actively in the work. P-members are obliged to vote on the
different standards development stages, and to contribute to meetings;
O-member (observing country), if the national body intends just to follow
the work as an observer. They have the right to attend meetings, vote at
some standards development stages, and submit comments.
P-members can have automatically changed their status to O-member if
failing at their obligations, i.e., meetings and votes. An O-member can become a
P-member if they want to participate more actively. A member of a committee
must notify if they want to contribute to a given subcommittee, regardless of
their status in the committee. When a subcommittee is established, all the
committee members are given the opportunity of joining it.
Other organizations may participate in the technical work of technical
committees by means of liaisons. Such organizations must be representative of
their technical or industrial field. They are required to be willing to contribute to
the technical work, assuming the ISO/IEC directives and the rest of the TC rules.
There may be liaisons at two levels: (a) technical committees or subcommittees,
called category A (active contribution) and category B (keep informed) ; (b)
working groups, called category D . Liaisons can also be arranged between TCs
working in related fields. Details about liaisons can be found in the ISO/IEC
Directive Part 1 [95].

4.2 ISO Standards and Quality

Quality Control starts with Quality. So let us take a look at what quality means.
Quality is, in general, a subjective term which perception vary from one person
to another. That is to say, something that is of good quality for one person could
be of poor quality for another one. Nevertheless, we can find different
definitions of quality. If we look at what the gods say, the meaning of quality is
also viewed from different perspectives, for example “fitness for use” for Joseph
Juran [98], or “conformance to requirements” for Philip Crosby [4]. Quality is
defined relative to the need to improve for W. Edwards Deming [100], whilst
Taguchi sees quality as “loss given to a society” [102]. In summary, different
approaches trying to reach the fuzzy concept of being good enough. And at the
end, George P. Box, the consummate ‘Renaissance man’ in the field of the
quality sciences2 wrote “I often think the Quality Gurus do not help”[2].
We can look for more formal definitions in dictionaries or encyclopedias. For
example, quality is defined on Cambridge Dictionaries3 as how good or bad
something is, which matches with the subjective perception. The Wikipedia
article4 for Quality (Business) says that Quality “has a pragmatic interpretation
as the non-inferiority or superiority of something; it is also defined as fitness for
purpose.”
But this chapter is about standards, so let us look for a standardized
definition of quality. Undoubtedly, the most popular ISO standards family is ISO
9000—Quality management. This family of standards includes the following:

ISO 9000:2005 Quality management systems—Fundamentals and

vocabulary [7]. This standard covers the basic concepts and language.
ISO 9001:2008 Quality management systems—Requirements [8]. This
standard sets out the requirements of a quality management system.
ISO 9004:2009 Managing for the sustained success of an organization—A
quality management approach [9]. This standard focuses on how to make
a quality management system more efficient and effective.
ISO 19011:2011 Guidelines for auditing management systems [11].
Importantly, the popularity of ISO 9000 is due to the certification
process. This standard sets out guidance on internal and external audits
of quality management systems.

It is in ISO 9000 where quality is defined as “degree to which a set of

inherent characteristics fulfils requirements.” The bolded terms are also
defined, being a characteristic a “distinguishing feature” and a requirement a
“need or expectation.” On the other hand, the role of statistical techniques
deserves a subclause (2.10) in ISO 9000, where it is remarked how statistical
techniques help in understanding variability, Statistics’ reason for being [3].
ISO TR/10017 [10] provides guidance on statistical techniques for quality
management systems.
The importance of standardization is twofold. On the one hand, the
fulfilment of requirements by the product or service characteristic must be based
in standard values that can be assessed. Sometimes this is stated by other
standards or regulations, for example for the average weight in packages. On the
other hand, the use of standardized procedures is a requirement for quality
assurance.
At this point, it is important to remark that, even though ISO 9000 family
and other certifiable rock stars5 are known for almost everybody, the ISO
Standards catalogue6 contains over 19,500 Standards. In particular, there are a
number of ISO Standards regarding statistical techniques used in quality control,
published by the ISO/TC69 Technical committee, Applications of Statistical
Methods. In Sec. 4.3, the standard development process is explained. The
structure of ISO Technical committees is detailed in Sect. 4.1. Details of TC69
secretariat and subcommittees (SCs) and their published standards are given in
Sects. 4.4–4.10. Sect. 4.11 outlines the role of R in ISO standards.

4.3 The ISO Standards Development Process

The procedures used to develop and maintain ISO standards and other
technical work are described in ISO/IEC Directives. Even though such directives
apply to ISO, IEC, and ISO/IEC Standards, we will refer hereon to ISO
terminology for the sake of clarity, although there could be slight differences at
the IEC scope. ISO Standards are developed and maintained by ISO technical
committees, see Sect. 4.1. ISO/IEC Directives are published in two parts:
Part 1 and Consolidated ISO Supplement: Official procedures to be
followed when developing and maintaining an International Standard and
procedures specific to ISO [95].
Part 2: Principles to structure and draft documents intended to become
International Standards, Technical Specifications or Publicly Available
Specifications. [94]
The process of elaborating and publishing an ISO standard is quite similar to
the academic peer-reviewed process when publishing scientific papers. After
someone proposes a new standard, possibly from outside the corresponding TC,
a draft (manuscript) is prepared, circulated, voted, and revised, throughout a
series of steps until the standard is published. Of course it might happen that a
proposal or draft is rejected at some point and eventually not published.
However, in contrast with academic publications, ISO standards are also
bounded to maintenance procedures to keep the applicable bulk of standards
alive. The different stages that an ISO standard passes through are summarized
in Table 4.1.
Table 4.1 Standard development project stages
Acronym Description
PWI Preliminary Work Item
NP New Work Item Proposal
WD Working Draft
CD Committee Draft
DIS Draft International Standard
FDIS Final Draft International Standard
ISO International Standard
SR Systematic Review

Once a proposal (NP) is accepted, it is included as a project in the

programme of work of the corresponding TC or SC. As such a project, it must
include target dates for each subsequent stage, a project leader, and procedures
for project management and progress control. Let us take a walk for the usual
stages of an ISO Standard development. Please note that some stages are
mandatory, whilst others could be skipped, see [95].

1. PWI. Preliminary Work Items that are not yet mature enough to be
incorporated to a programme of work, for example relating to emerging
technologies or recent discoveries. If the preliminary work item has not
progressed to the proposal stage in 3 years, it is automatically deleted from
the programme of work.

2. NP. A new Work Item Proposal can be for a new standard, a new part of an
existing standard, a technical specification (TS) or a publicly available
specification (PAS) . The proposal can be made by different stakeholders,
such as the own TC, a national body, or an organization in liaison, among
others. It must include at least an outline and a project leader. The (of course
standardized) form is circulated to the TC members to vote. Approval
requires simple majority of P-members (see Sect. 4.1) and the commitment to
participate by some of them. Once approved, it is included in the TC
programme of work as a project.

3.
WD. A first version of the Working Draft could have been submitted with the
NP. Once the project is accepted, the project leader and the experts nominated
during the approval work together to prepare/improve a working draft
conforming to Part 2 of ISO/IEC Directives [94]. A working group can be
proposed by the TC Secretariat. ISO/IEC Directive Part 2 assures that all
standards have the same structure and style. ISO Standards are published in
English and French7, so all efforts must be made to have English and French
versions of the text in order to avoid delays. When the WD is finished, it is
circulated to TC members as a first committee draft (CD).

4. CD. At this stage, national bodies provide comments on the CD. It is quite an
active stage in which technical details are discussed within the TC or SC both
electronically and in-person meetings. Comments are compiled by the
secretariat until an appropriate level of consensus is attained. In case of
doubt, a two-thirds majority is usually sufficient. During this stage the CD
can be discussed and revised until it is proposed as a DIS.

5. DIS. A draft international standard is circulated for voting and commenting to
all national bodies, not only to those involved in the TC/SC. At this stage,
technical comments, mandatory in case of negative vote, can be made.
Comments can be addressed by the secretary for the final draft. Before
stepping into the next stage, a report on the voting and decisions on
comments is circulated again, and finally an FDIS is prepared.

6. FDIS. This is the last stage before publication. The procedure follows a
similar procedure to the one in DIS. However, editorial comments are
expected rather than technical comments.

7. ISO. The international standard is eventually published once the comments in
FDIS has been addressed.

8. SR. After publication, an ISO Standard and other deliverables such as TR are
subject to systematic review in order to determine whether it should be
confirmed, revised/amended, converted to another form of deliverable, or
withdrawn. For an ISO Standard, the maximum elapsed time before
systematic review is 5 years.

Figure 4.1 summarizes the standards development process, including
approximate target dates, see [95] for details. Please note that at any voting
stage, the document can be rejected and referred back to the TC/SC, that may
decide to resubmit a modified version, change the type of document (e.g., a
technical specification instead of an international standard), or cancel the project.
Fig. 4.1 ISO Standards publication path. Standardized process

4.4 ISO TC69 Secretariat

The scope of technical committee ISO TC69 is the “standardization in the
application of statistical methods, including generation, collection (planning and
design), analysis, presentation and interpretation of data.” Most of the standards
stemmed from ISO TC69 are developed by the subcommittees, at the scope of
their specific field of activity. Nevertheless, some standards are under the direct
responsibility of the TC secretariat. It is important to remark that ISO TC69 has
also “the function of advisor to all ISO technical committees in matters
concerning the application of statistical methods in standardization.”
 ISO TC69 is structured in six subcommittees, a Working Group (WG), a
Chairman Advisory Group (CAG), and an Ad-Hoc Group (AHG). The role of
each group and subcommittee are as follows:
ISO/TC 69/CAG: Chairman Advisory Group
ISO/TC 69/AHG 1: Documents to support the application of statistical
methods standards;
ISO/TC 69/WG 3: Statistical interpretation of data;
ISO/TC 69/SC 1: Terminology and symbols;
ISO/TC 69/SC 4: Applications of statistical methods in process
management;
ISO/TC 69/SC 5: Acceptance sampling;
ISO/TC 69/SC 6: Measurement methods and results;
ISO/TC 69/SC 7: Applications of statistical and related techniques for the
implementation of Six Sigma;
ISO/TC 69/SC 8: Application of statistical and related methodology for
new technology and product development.
The following standards and TRs have been published under the direct
responsibility of ISO TC69 Secretariat. A short description of each document
can be found at the standard webpage within the ISO website, see the references
section at the end of this chapter.
ISO 11453:1996 Statistical interpretation of data—Tests and confidence
intervals relating to proportions [83].
ISO 11453:1996/Cor 1:1999 [80].
ISO 16269-4:2010 Statistical interpretation of data—Part 4: Detection and
treatment of outliers [82].
ISO 16269-6:2014 Statistical interpretation of data—Part 6: Determination
of statistical tolerance intervals [88].
ISO 16269-7:2001 Statistical interpretation of data—Part 7: Median—
Estimation and confidence intervals [84].
ISO 16269-8:2004 Statistical interpretation of data—Part 8: Determination
of prediction intervals [87].
ISO 2602:1980 Statistical interpretation of test results—Estimation of the
mean—Confidence interval [89].
 ISO 2854:1976 Statistical interpretation of data—Techniques of estimation
and tests relating to means and variances [90].
ISO 28640:2010 Random variate generation methods [91].
ISO 3301:1975 Statistical interpretation of data—Comparison of two
means in the case of paired observations [92].
ISO 3494:1976 Statistical interpretation of data—Power of tests relating to
means and variances [93].
ISO 5479:1997 Statistical interpretation of data—Tests for departure from
the normal distribution [85].
ISO/TR 13519:2012 Guidance on the development and use of ISO
statistical publications supported by software [86].
ISO/TR 18532:2009 Guidance on the application of statistical methods to
quality and to industrial standardization [81].

4.5 ISO TC69/SC1: Terminology

Terminology standards are very important not only for the rest of ISO TC69,
but also to other committees that use statistical terminology. In fact, this SC has
liaisons with several TCs and organizations. ISO TC69/SC1 has currently the
following Working Groups:
ISO/TC 69/SC 1/WG 2: Revisions of ISO 3534;
ISO/TC 69/SC 1/WG 5: Terminology liaison;
ISO/TC 69/SC 1/WG 6: Terminology for emerging areas of statistical
applications.
The following standards have been published by ISO TC69/SC1. A short
description of each document can be found at the standard webpage within the
ISO website, see the references section at the end of this chapter.
ISO 3534-1:2006 Statistics—Vocabulary and symbols—Part 1: General
statistical terms and terms used in probability [12].
ISO 3534-2:2006 Statistics—Vocabulary and symbols—Part 2: Applied
statistics [14].
ISO 3534-3:2013 Statistics—Vocabulary and symbols—Part 3: Design of
experiments [13].
ISO 3534-4:2014 Statistics—Vocabulary and symbols—Part 4: Survey
 sampling [15].

4.6 ISO TC69/SC4: Application of Statistical Methods

in Process Management
This subcommittee develops standards regarding Statistical Process Control,
capability analysis, and control charts. There are three working groups, namely:
ISO/TC 69/SC 4/WG 10: Revision of control charts standards;
ISO/TC 69/SC 4/WG 11: Process capability and performance;
ISO/TC 69/SC 4/WG 12: Implementation of statistical Process Control.
ISO TC69/SC4 has published the following documents. Note that one of
them is a TR, probably due to the fact that during the standard development
process, see Sect. 4.3, the document did not reach the appropriate level to be a
standard. A short description of each document can be found at the standard
webpage within the ISO website, see the references section at the end of this
chapter.
ISO 11462-1:2001 Guidelines for implementation of statistical process
control (SPC)—Part 1: Elements of SPC [18].
ISO 22514-1:2014 Statistical methods in process management—Capability
and performance—Part 1: General principles and concepts [25].
ISO 22514-2:2013 Statistical methods in process management—Capability
and performance—Part 2: Process capability and performance of time-
dependent process models [21].
ISO 22514-3:2008 Statistical methods in process management—Capability
and performance—Part 3: Machine performance studies for measured data
on discrete parts [16].
ISO 22514-6:2013 Statistical methods in process management—Capability
and performance—Part 6: Process capability statistics for characteristics
following a multivariate normal distribution [22].
ISO 22514-7:2012 Statistical methods in process management—Capability
and performance—Part 7: Capability of measurement processes [19].
ISO 22514-8:2014 Statistical methods in process management—Capability
and performance—Part 8: Machine performance of a multi-state production
process [26].
 ISO 7870-1:2014 Control charts—Part 1: General guidelines [27].
ISO 7870-2:2013 Control charts—Part 2: Shewhart control charts [23].
ISO 7870-3:2012 Control charts—Part 3: Acceptance control charts [20].
ISO 7870-4:2011 Control charts—Part 4: Cumulative sum charts [17].
ISO 7870-5:2014 Control charts—Part 5: Specialized control charts [28].
ISO/TR 22514-4:2007 Statistical methods in process management—
Capability and performance—Part 4: Process capability estimates and
performance measures [24].

4.7 ISO TC69/SC5: Acceptance Sampling

There are four working groups in the acceptance sampling SC:
ISO/TC 69/SC 5/WG 2: Sampling procedures for inspection by attributes
(Revision of ISO 2859);
ISO/TC 69/SC 5/WG 3: Sampling procedures and charts for inspection by
variables for percent nonconforming (Revision of ISO 3951);
ISO/TC 69/SC 5/WG 8: Sampling by attributes;
ISO/TC 69/SC 5/WG 10: Audit sampling.
Acceptance Sampling is a popular topic, as the number of standards
published by ISO TC69/SC5 shows. A short description of each document can
be found at the standard webpage within the ISO website, see the references
section at the end of this chapter.
ISO 13448-1:2005 Acceptance sampling procedures based on the
allocation of priorities principle (APP)—Part 1: Guidelines for the APP
approach [34].
ISO 13448-2:2004 Acceptance sampling procedures based on the
allocation of priorities principle (APP)—Part 2: Coordinated single
sampling plans for acceptance sampling by attributes [37].
ISO 14560:2004 Acceptance sampling procedures by attributes—Specified
quality levels in nonconforming items per million [38].
ISO 18414:2006 Acceptance sampling procedures by attributes—Accept-
zero sampling system based on credit principle for controlling outgoing
quality [39].
ISO 21247:2005 Combined accept-zero sampling systems and process
control procedures for product acceptance [40].
ISO 24153:2009 Random sampling and randomization procedures [48].
ISO 2859-10:2006 Sampling procedures for inspection by attributes—Part
10: Introduction to the ISO 2859 series of standards for sampling for
inspection by attributes [41].
ISO 2859-1:1999 Sampling procedures for inspection by attributes—Part 1:
Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot
inspection [42].
ISO 2859-1:1999/Amd 1:2011 [31].
ISO 2859-3:2005 Sampling procedures for inspection by attributes—Part 3:
Skip-lot sampling procedures [43].
ISO 2859-4:2002 Sampling procedures for inspection by attributes—Part 4:
Procedures for assessment of declared quality levels [29].
ISO 2859-5:2005 Sampling procedures for inspection by attributes—Part 5:
System of sequential sampling plans indexed by acceptance quality limit
(AQL) for lot-by-lot inspection [44].
ISO 28801:2011 Double sampling plans by attributes with minimal sample
sizes, indexed by producer’s risk quality (PRQ) and consumer’s risk quality
(CRQ) [32].
ISO 3951-1:2013 Sampling procedures for inspection by variables—Part 1:
Specification for single sampling plans indexed by acceptance quality limit
(AQL) for lot-by-lot inspection for a single quality characteristic and a
single AQL [35].
ISO 3951-2:2013 Sampling procedures for inspection by variables—Part 2:
General specification for single sampling plans indexed by acceptance
quality limit (AQL) for lot-by-lot inspection of independent quality
characteristics [36].
ISO 3951-3:2007 Sampling procedures for inspection by variables—Part 3:
Double sampling schemes indexed by acceptance quality limit (AQL) for
lot-by-lot inspection [30].
ISO 3951-4:2011 Sampling procedures for inspection by variables—Part 4:
Procedures for assessment of declared quality levels [33].
ISO 3951-5:2006 Sampling procedures for inspection by variables—Part 5:
Sequential sampling plans indexed by acceptance quality limit (AQL) for
inspection by variables (known standard deviation) [45].
 ISO 8422:2006 Sequential sampling plans for inspection by attributes [46].
ISO 8423:2008 Sequential sampling plans for inspection by variables for
percent nonconforming (known standard deviation) [47].

4.8 ISO TC69/SC6: Measurement Methods and Results

The following working groups can be found at ISO TC69/SC6:
ISO/TC 69/SC 6/WG 1 Accuracy of measurement methods and results;
ISO/TC 69/SC 6/WG 5 Capability of detection;
ISO/TC 69/SC 6/WG 7 Statistical methods to support measurement
uncertainty evaluation;
ISO/TC 69/SC 6/WG 9 Statistical methods for use in proficiency testing.
The following standards and documents have been published by ISO TC
69/SC6. In addition to international standards we can found technical reports and
technical specifications. A short description of each document can be found at
the standard webpage within the ISO website, see the references section at the
end of this chapter.
ISO 10576-1:2003 Statistical methods—Guidelines for the evaluation of
conformity with specified requirements—Part 1: General principles [61].
ISO 10725:2000 Acceptance sampling plans and procedures for the
inspection of bulk materials [49].
ISO 11095:1996 Linear calibration using reference materials [62].
ISO 11648-1:2003 Statistical aspects of sampling from bulk materials—
Part 1: General principles [63].
ISO 11648-2:2001 Statistical aspects of sampling from bulk materials—
Part 2: Sampling of particulate materials [52].
ISO 11843-1:1997 Capability of detection—Part 1: Terms and definitions
[64].
ISO 11843-2:2000 Capability of detection—Part 2: Methodology in the
linear calibration case [50].
ISO 11843-3:2003 Capability of detection—Part 3: Methodology for
determination of the critical value for the response variable when no
calibration data are used [65].
ISO 11843-4:2003 Capability of detection—Part 4: Methodology for
 comparing the minimum detectable value with a given value [66].
ISO 11843-5:2008 Capability of detection—Part 5: Methodology in the
linear and non-linear calibration cases [53].
ISO 11843-6:2013 Capability of detection—Part 6: Methodology for the
determination of the critical value and the minimum detectable value in
Poisson distributed measurements by normal approximations [57].
ISO 11843-7:2012 Capability of detection—Part 7: Methodology based on
stochastic properties of instrumental noise [67].
ISO 21748:2010 Guidance for the use of repeatability, reproducibility and
trueness estimates in measurement uncertainty estimation [51].
ISO 5725-1:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 1: General principles and definitions [54].
ISO 5725-2:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 2: Basic method for the determination of
repeatability and reproducibility of a standard measurement method [58].
ISO 5725-3:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 3: Intermediate measures of the precision of a
standard measurement method [59].
ISO 5725-4:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 4: Basic methods for the determination of the
trueness of a standard measurement method [60].
ISO 5725-5:1998 Accuracy (trueness and precision) of measurement
methods and results—Part 5: Alternative methods for the determination of
the precision of a standard measurement method [68].
ISO 5725-6:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 6: Use in practice of accuracy values [55].
ISO/TR 13587:2012 Three statistical approaches for the assessment and
interpretation of measurement uncertainty [56].
ISO/TS 21749:2005 Measurement uncertainty for metrological
applications—Repeated measurements and nested experiments [70].
ISO/TS 28037:2010 Determination and use of straight-line calibration
functions [69].

4.9 ISO TC69/SC7: Applications of Statistical and

Related Techniques for the Implementation of Six Sigma
Six Sigma is a breakthrough methodology that is in part extending the use of
statistics throughout companies all over the World. Even though Six Sigma was
born in the 1980s, this ISO SC was recently created, specifically in 2008. The
work is organized in three WGs and one AHG as follows:
ISO/TC 69/SC 7/WG 1: Design of experiments;
ISO/TC 69/SC 7/AHG 1: Strategic planning and working practice;
ISO/TC 69/SC 7/WG 2: Process measurement and measurement capability;
ISO/TC 69/SC 7/WG 3: Six sigma methodology.
Given its young age, there are few ISO TC 69/SC7 standards published.
However, the SC work programme works in new standards.8 Technical Reports
produced by this SC illustrate different techniques for some applications, using
different software packages. A short description of each document can be found
at the standard webpage within the ISO website, see the references section at the
end of this chapter.
ISO 13053-1:2011 Quantitative methods in process improvement—Six
Sigma—Part 1: DMAIC methodology [75].
ISO 13053-2:2011 Quantitative methods in process improvement—Six
Sigma—Part 2: Tools and techniques [76].
ISO 17258:2015 Statistical methods—Six Sigma—Basic criteria
underlying benchmarking for Six Sigma in organisations [78].
ISO/TR 12845:2010 Selected illustrations of fractional factorial screening
experiments [73].
ISO/TR 12888:2011 Selected illustrations of gauge repeatability and
reproducibility studies [77].
ISO/TR 14468:2010 Selected illustrations of attribute agreement analysis
[74].
ISO/TR 29901:2007 Selected illustrations of full factorial experiments
with four factors [71].
ISO/TR 29901:2007/Cor 1:2009 [72].

4.10 ISO TC69/SC8: Application of Statistical and

Related Methodology for New Technology and Product
Development
ISO TC69/SC8 is the youngest of the SCs in TC69. The following three
WGs are working on developing innovative standards:
ISO/TC 69/SC 8/WG 1: Sample survey;
ISO/TC 69/SC 8/WG 2: Transformation;
ISO/TC 69/SC 8/WG 3: Optimization;
Established in 2009, at the time this book is being written only one
international standard has been published by ISO/TC 69 SC 8. However, one
DIS and seven Approved Work Items (AWIs) can be found in the work
programme, some or all of them will very likely be published when you are
reading this book. A short description of that standard can be found at the
standard webpage within the ISO website, see the references section at the end
of this chapter.
ISO 16336:2014 Applications of statistical and related methods to new
technology and product development process—Robust parameter design
(RPD) [79].

4.11 The Role of R in Standards

Nowadays, we cannot think about applying statistical techniques without using a
statistical software. Even though international standards tend to be software-
agnostic, sometimes software tools and packages appear within them. Market-
leaders such as Minitab and JMP (by SAS) are usually the most referred. For
example, if we look at ISO TC69/SC7 TRs, they are used in all of them: ISO/TR
29901 [71], ISO/TR 12888 [77] , ISO/TR 29901 [71] , and ISO/TR 12845 [73] .
In ISO/TR 12845 also R is referred, which is a signal of the increasing interest in
the industry on the R statistical software and programming language. In fact, in
the comments submitted by ISO members during the standards development
process, see Sect. 4.3, R is more and more mentioned by experts. On the other
hand, R has become a kind of de facto standard for data analysis, see, for
example, [1, 3, 5, 6, 101]. Being R open source, it is just common sense to use it
within standards in order to allow all the users test methods and reproduce
examples.
Furthermore, R is not just statistical software. R is a programming language
that can be used to illustrate new methods or algorithms. For example, ISO
28640 [91] is for random variate generation methods. Annex B of that standard
lists C code for several algorithms, including the Mersenne-Twister algorithm
[99], the default one to generate random numbers in R, see the documentation
for the RNG topic (?RNG). R could be used also for this purpose, i.e., illustrate
algorithms and programming code. However, R is not still an ISO standardized
language like C. Perhaps some day R and other emerging programming
languages for data analysis like Python will join the club at ISO/IEC JTC 1/SC
22—Programming languages, their environments and system software
interfaces. In addition to ISO/IEC 9899 for C [96] and ISO/IEC 14882 for C++
[97] , this Joint Technical Committee has published a total of 96 ISO standards
for a number of programming languages, including Prolog, Pascal, Ruby,
Fortran, and Ada, among others. Check the JTC web page9 to find out more
about them.
To finish this chapter, let us use R to retrieve information about ISO
standards from the web. The ISO catalogue can be browsed in the website
(http://www.iso.org) in the usual way. The search box at the top right side allows
to find a standard immediately. The full standards catalogue can be browsed by
TC or by International Classification for Standards (ICS) at http://www.iso.org/
iso/home/store/catalogue_tc.htm. Furthermore, a great resource is the ISO
Online Browse Platform (OBP) (https://www.iso.org/obp/). Previews of ISO
standards are available, both from the individual standard web page, that
redirects to OBP, and from the searchable data base. In addition to this manual
search, information may be retrieved from the ISO RSS channels. There is a link
to subscribe RSS channels at a TC or SC web page and at a single standard
webpage. For example, Figure 4.2 shows the ISO/TC69 web page where you can
find a link to subscribe to updates. This link leads to an XML file with the
standards of the TC. As XML files are data structures, they can be imported into
R objects for further use. For example, the following code retrieves TC69/SC1
standards into an R data.frame, via the XML package:
Fig. 4.2 ISO TC69 web page. A link to subscribe RSS channels is available at each TC, SC, and individual
standard web pages

Now we have a data.frame called DATA with ten observations and six
variables, namely: title, link, description, category, guid, pubDate. From the
information of each variable we can extract information to have more useful
variables, for example splitting the title variable to get the standard code as

follows:

Finally, we can use also XML syntax to do what is usually called

webscrapping, i.e., retrieving information from websites accessing the pages
HTML structure. One of the fields retrieved from the RSS channel is the URL of
each standard web page. In such standard web pages we can find interesting
information, for example an abstract for the standard. Let us get the abstract of
ISO 3534-1:2006.10 First, we need the URL for the standard11:

Then, following a similar approach of that for an XML file, the information
from an HTML file can be get. However, errors may arise due to the strict rules
of the XML specification. Webscrapping is easier to do using the rvest package,
see the package documentation for details. In our example, the following code
gets the abstract for ISO 3534-1:200612:

Now you can use the text for any purpose, for example to print out the
abstract with the cat function:

Output:

Abstract
ISO 3534-1:2006 defines general statistical terms and terms used in
probability which may be used in the drafting of other International
Standards. In addition, it defines symbols for a limited number of these
terms.

In any case, webscrapping requires practice and knowledge on HTML

documents structure. We have just illustrated how R can help beyond statistical
data analysis, with an application on catching up with the standards development
world.
References
1. Baclawski, K.: Introduction to Probability with R. Chapman & Hall/CRC Texts in Statistical Science.
Taylor & Francis, New York (2008)
[MATH]

2. Box, G.: Total quality: its origins and its future. In: Total Quality Management, pp. 119–127. Springer,
Berlin (1995)

3. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. Statistical Engineering for Process
Improvement, Use R!, vol. 36. Springer, New York (2012). http://www.springer.com/statistics/book/
978-1-4614-3651-5

4. Crosby, P.: Quality is Free: The Art of Making Quality Certain. Mentor Book. McGraw-Hill, New
York (1979). https://books.google.es/books?id=n4IubCcpm0EC

5. Eubank, R., Kupresanin, A.: Statistical Computing in C++ and R. Chapman & Hall/CRC The R
Series. Taylor & Francis, New York (2011). https://books.google.es/books?id=CmZRui57cWkC

6. Gardener, M.: Beginning R: The Statistical Programming Language. ITPro collection. Wiley, New
York (2012). https://books.google.es/books?id=iJoKYSWCubEC

7. ISO TC176/SC1 – Concepts and terminology: ISO 9000:2005 - Quality management systems –
Fundamentals and vocabulary. Published standard (2005). http://www.iso.org/iso/home/store/
catalogue_tc/catalogue_detail.htm?csnumber=42180

8. ISO TC176/SC2 – Quality systems: ISO 9001:2008 - Quality management systems – Requirements.
Published standard (2008). http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?
csnumber=46486

9. ISO TC176/SC2 – Quality systems: ISO 9004:2009 - Managing for the sustained success of an
organization – A quality management approach. Published standard (2009). http://www.iso.org/iso/
home/store/catalogue_tc/catalogue_detail.htm?csnumber=41014

10. ISO TC176/SC3 – Supporting technologies: ISO/TR 10017:2003 - Guidance on statistical techniques
for ISO 9001:2000. Published Technical Report (2003). http://www.iso.org/iso/home/store/catalogue_
tc/catalogue_detail.htm?csnumber=36674

11. ISO TC176/SC3 – Supporting technologies: ISO 19011:2011 - Guidelines for auditing management
systems. Published standard (2011). http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.
htm?csnumber=50675

12. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). http://www.
iso.org/iso/catalogue_detail.htm?csnumber=40145

13. ISO TC69/SC1–Terminology and Symbols: ISO 3534-3:2013 - Statistics – Vocabulary and symbols –
Part 3: Design of experiments. Published standard (2013). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=44245

14. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
 csnumber=40147

15. ISO TC69/SC1–Terminology and Symbols: ISO 3534-4:2014 - Statistics – Vocabulary and symbols –
Part 4: Survey sampling. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=56154

16. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-3:2008 -

Statistical methods in process management – Capability and performance – Part 3: Machine
performance studies for measured data on discrete parts. Published standard (2011). http://www.iso.
org/iso/catalogue_detail.htm?csnumber=46531

17. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-4:2011 -

Control charts – Part 4: Cumulative sum charts. Published standard (2011). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40176

18. ISO TC69/SC4–Applications of statistical methods in process management: ISO 11462-1:2001 -

Guidelines for implementation of statistical process control (SPC) – Part 1: Elements of SPC.
Published standard (2012). http://www.iso.org/iso/catalogue_detail.htm?csnumber=33381

19. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-7:2012 -

Statistical methods in process management – Capability and performance – Part 7: Capability of
measurement processes. Published standard (2012). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=54077

20. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-3:

2012 - Control charts – Part 3: Acceptance control charts. Published standard (2012). http://www.iso.
org/iso/catalogue_detail.htm?csnumber=40175

21. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-2:2013 -

Statistical methods in process management – Capability and performance – Part 2: Process capability
and performance of time-dependent process models. Published standard (2013). http://www.iso.org/
iso/catalogue_detail.htm?csnumber=46530

22. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-6:2013 -

Statistical methods in process management – Capability and performance – Part 6: Process capability
statistics for characteristics following a multivariate normal distribution. Published standard (2013).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=52962

23. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 -

Control charts – Part 2: Shewhart control charts. Published standard (2013). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40174

24. ISO TC69/SC4–Applications of statistical methods in process management: ISO/TR 22514-4:2007 -

Statistical methods in process management – Capability and performance – Part 4: Process capability
estimates and performance measures. Published standard (2013). http://www.iso.org/iso/catalogue_
detail.htm?csnumber=46532

25. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-1:2014 -

Statistical methods in process management – Capability and performance – Part 1: General principles
and concepts. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=
64135

26. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-8:2014 -

 Statistical methods in process management – Capability and performance – Part 8: Machine
performance of a multi-state production process. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=61630

27. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-1:2014 -

Control charts – Part 1: General guidelines. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=62649

28. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-5:2014 -

Control charts – Part 5: Specialized control charts. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40177

29. ISO TC69/SC5–Acceptance sampling: ISO 2859-4:2002 - Sampling procedures for inspection by
attributes – Part 4: Procedures for assessment of declared quality levels. Published standard (2010).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=36164

30. ISO TC69/SC5–Acceptance sampling: ISO 3951-3:2007 - Sampling procedures for inspection by
variables – Part 3: Double sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot
inspection. Published standard (2010). http://www.iso.org/iso/catalogue_detail.htm?csnumber=40556

31. ISO TC69/SC5–Acceptance sampling: ISO 2859-1:1999/Amd 1:2011. Published standard (2011).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=53053

32. ISO TC69/SC5–Acceptance sampling: ISO 28801:2011 - Double sampling plans by attributes with
minimal sample sizes, indexed by producer’s risk quality (PRQ) and consumer’s risk quality (CRQ).
Published standard (2011). http://www.iso.org/iso/catalogue_detail.htm?csnumber=44963

33. ISO TC69/SC5–Acceptance sampling: ISO 3951-4:2011 - Sampling procedures for inspection by
variables – Part 4: Procedures for assessment of declared quality levels. Published standard (2011).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=44806

34. ISO TC69/SC5–Acceptance sampling: ISO 13448-1:2005 - Acceptance sampling procedures based on
the allocation of priorities principle (APP) – Part 1: Guidelines for the APP approach. Published
standard (2013). http://www.iso.org/iso/catalogue_detail.htm?csnumber=37429

35. ISO TC69/SC5–Acceptance sampling: ISO 3951-1:2013 - Sampling procedures for inspection by
variables – Part 1: Specification for single sampling plans indexed by acceptance quality limit (AQL)
for lot-by-lot inspection for a single quality characteristic and a single AQL. Published standard
(2013). http://www.iso.org/iso/catalogue_detail.htm?csnumber=57490

36. ISO TC69/SC5–Acceptance sampling: ISO 3951-2:2013 - Sampling procedures for inspection by
variables – Part 2: General specification for single sampling plans indexed by acceptance quality limit
(AQL) for lot-by-lot inspection of independent quality characteristics. Published standard (2013).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=57491

37. ISO TC69/SC5–Acceptance sampling: ISO 13448-2:2004 - Acceptance sampling procedures based on
the allocation of priorities principle (APP) – Part 2: Coordinated single sampling plans for acceptance
sampling by attributes. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=37430

38. ISO TC69/SC5–Acceptance sampling: ISO 14560:2004 - Acceptance sampling procedures by

attributes – Specified quality levels in nonconforming items per million. Published standard (2014).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=37869
39. ISO TC69/SC5–Acceptance sampling: ISO 18414:2006 - Acceptance sampling procedures by
attributes – Accept-zero sampling system based on credit principle for controlling outgoing quality.
Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=38684

40. ISO TC69/SC5–Acceptance sampling: ISO 21247:2005 - Combined accept-zero sampling systems
and process control procedures for product acceptance. Published standard (2014). http://www.iso.org/
iso/catalogue_detail.htm?csnumber=34445

41. ISO TC69/SC5–Acceptance sampling: ISO 2859-10:2006 - Sampling procedures for inspection by
attributes – Part 10: Introduction to the ISO 2859 series of standards for sampling for inspection by
attributes. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=39991

42. ISO TC69/SC5–Acceptance sampling: ISO 2859-1:1999 - Sampling procedures for inspection by
attributes – Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot
inspection. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=1141

43. ISO TC69/SC5–Acceptance sampling: ISO 2859-3:2005 - Sampling procedures for inspection by
attributes – Part 3: Skip-lot sampling procedures. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=34684

44. ISO TC69/SC5–Acceptance sampling: ISO 2859-5:2005 - Sampling procedures for inspection by
attributes – Part 5: System of sequential sampling plans indexed by acceptance quality limit (AQL) for
lot-by-lot inspection. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=39295

45. ISO TC69/SC5–Acceptance sampling: ISO 3951-5:2006 - Sampling procedures for inspection by
variables – Part 5: Sequential sampling plans indexed by acceptance quality limit (AQL) for
inspection by variables (known standard deviation). Published standard (2014). http://www.iso.org/
iso/catalogue_detail.htm?csnumber=39294

46. ISO TC69/SC5–Acceptance sampling: ISO 8422:2006 - Sequential sampling plans for inspection by
attributes. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=39915

47. ISO TC69/SC5–Acceptance sampling: ISO 8423:2008 - Sequential sampling plans for inspection by
variables for percent nonconforming (known standard deviation). Published standard (2014). http://
www.iso.org/iso/catalogue_detail.htm?csnumber=41992

48. ISO TC69/SC5–Acceptance sampling: ISO 24153:2009 - Random sampling and randomization
procedures. Published standard (2015). http://www.iso.org/iso/catalogue_detail.htm?csnumber=42039

49. ISO TC69/SC6–Measurement methods and results: ISO 10725:2000 - Acceptance sampling plans and
procedures for the inspection of bulk materials. Published standard (2010). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=33418

50. ISO TC69/SC6–Measurement methods and results: ISO 11843-2:2000 - Capability of detection – Part
2: Methodology in the linear calibration case. Published standard (2010). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=20186

51. ISO TC69/SC6–Measurement methods and results: ISO 21748:2010 - Guidance for the use of
repeatability, reproducibility and trueness estimates in measurement uncertainty estimation. Published
standard (2010). http://www.iso.org/iso/catalogue_detail.htm?csnumber=46373

52. ISO TC69/SC6–Measurement methods and results: ISO 11648-2:2001 - Statistical aspects of
sampling from bulk materials – Part 2: Sampling of particulate materials. Published standard (2012).
 http://www.iso.org/iso/catalogue_detail.htm?csnumber=23663

53. ISO TC69/SC6–Measurement methods and results: ISO 11843-5:2008 - Capability of detection – Part
5: Methodology in the linear and non-linear calibration cases. Published standard (2012). http://www.
iso.org/iso/catalogue_detail.htm?csnumber=42000

54. ISO TC69/SC6–Measurement methods and results: ISO 5725-1:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 1: General principles and definitions. Published
standard (2012). http://www.iso.org/iso/catalogue_detail.htm?csnumber=11833

55. ISO TC69/SC6–Measurement methods and results: ISO 5725-6:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 6: Use in practice of accuracy values. Published
standard (2012). http://www.iso.org/iso/catalogue_detail.htm?csnumber=11837

56. ISO TC69/SC6–Measurement methods and results: ISO/TR 13587:2012 - Three statistical approaches
for the assessment and interpretation of measurement uncertainty. Published standard (2012). http://
www.iso.org/iso/catalogue_detail.htm?csnumber=54052

57. ISO TC69/SC6–Measurement methods and results: ISO 11843-6:2013 - Capability of detection – Part
6: Methodology for the determination of the critical value and the minimum detectable value in
Poisson distributed measurements by normal approximations. Published standard (2013). http://www.
iso.org/iso/catalogue_detail.htm?csnumber=53677

58. ISO TC69/SC6–Measurement methods and results: ISO 5725-2:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 2: Basic method for the determination of
repeatability and reproducibility of a standard measurement method. Published standard (2013). http://
www.iso.org/iso/catalogue_detail.htm?csnumber=11834

59. ISO TC69/SC6–Measurement methods and results: ISO 5725-3:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 3: Intermediate measures of the precision of a
standard measurement method. Published standard (2013). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=11835

60. ISO TC69/SC6–Measurement methods and results: ISO 5725-4:1994 - Accuracy (trueness and
precision) of measurement methods and results – Part 4: Basic methods for the determination of the
trueness of a standard measurement method. Published standard (2013). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=11836

61. ISO TC69/SC6–Measurement methods and results: ISO 10576-1:2003 - Statistical methods –
Guidelines for the evaluation of conformity with specified requirements – Part 1: General principles.
Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=32373

62. ISO TC69/SC6–Measurement methods and results: ISO 11095:1996 - Linear calibration using
reference materials. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=1060

63. ISO TC69/SC6–Measurement methods and results: ISO 11648-1:2003 - Statistical aspects of
sampling from bulk materials – Part 1: General principles. Published standard (2014). http://www.iso.
org/iso/catalogue_detail.htm?csnumber=33484

64. ISO TC69/SC6–Measurement methods and results: ISO 11843-1:1997 - Capability of detection – Part
1: Terms and definitions. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=1096
65. ISO TC69/SC6–Measurement methods and results: ISO 11843-3:2003 - Capability of detection – Part
3: Methodology for determination of the critical value for the response variable when no calibration
data are used. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=
34410

66. ISO TC69/SC6–Measurement methods and results: ISO 11843-4:2003 - Capability of detection – Part
4: Methodology for comparing the minimum detectable value with a given value. Published standard
(2014). http://www.iso.org/iso/catalogue_detail.htm?csnumber=34411

67. ISO TC69/SC6–Measurement methods and results: ISO 11843-7:2012 - Capability of detection – Part
7: Methodology based on stochastic properties of instrumental noise. Published standard (2014).
http://www.iso.org/iso/catalogue_detail.htm?csnumber=53678

68. ISO TC69/SC6–Measurement methods and results: ISO 5725-5:1998 - Accuracy (trueness and
precision) of measurement methods and results – Part 5: Alternative methods for the determination of
the precision of a standard measurement method. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=1384

69. ISO TC69/SC6–Measurement methods and results: ISO/TS 28037:2010 - Determination and use of
straight-line calibration functions. Published standard (2014). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=44473

70. ISO TC69/SC6–Measurement methods and results: ISO/TS 21749:2005 - Measurement uncertainty
for metrological applications – Repeated measurements and nested experiments. Published standard
(2015). http://www.iso.org/iso/catalogue_detail.htm?csnumber=34687

71. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO/TR 29901:2007 - Selected illustrations of full factorial experiments with four factors.
Published standard (2007). http://www.iso.org/iso/catalogue_detail.htm?csnumber=45731

72. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO/TR 29901:2007/Cor 1:2009. Published standard (2009). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=54056

73. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO/TR 12845:2010 - Selected illustrations of fractional factorial screening experiments.
Published standard (2010). http://www.iso.org/iso/catalogue_detail.htm?csnumber=51963

74. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO/TR 14468:2010 - Selected illustrations of attribute agreement analysis. Published
standard (2010). http://www.iso.org/iso/catalogue_detail.htm?csnumber=52900

75. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO 13053-1:2011 - Quantitative methods in process improvement – Six Sigma – Part 1:
DMAIC methodology. Published standard (2011). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=52901

76. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO 13053-2:2011 - Quantitative methods in process improvement – Six Sigma – Part 2: Tools
and techniques. Published standard (2011). http://www.iso.org/iso/catalogue_detail.htm?csnumber=
52902
77. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO/TR 12888:2011 - Selected illustrations of gauge repeatability and reproducibility studies.
Published standard (2011). http://www.iso.org/iso/catalogue_detail.htm?csnumber=52899

78. ISO TC69/SC7–Applications of statistical and related techniques for the implementation of Six
Sigma: ISO 17258:2015 - Statistical methods – Six Sigma – Basic criteria underlying benchmarking
for Six Sigma in organisations. Published standard (2015). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=59489

79. ISO TC69/SC8–Application of statistical and related methodology for new technology and product
development: ISO 16336:2014 - Applications of statistical and related methods to new technology and
product development process – Robust parameter design (RPD). Published standard (2014). http://
www.iso.org/iso/catalogue_detail.htm?csnumber=56183

80. ISO TC69/SCS–Secretariat: ISO 11453:1996/Cor 1:1999. Published standard (1999). http://www.iso.
org/iso/catalogue_detail.htm?csnumber=32469

81. ISO TC69/SCS–Secretariat: ISO/TR 18532:2009 - Guidance on the application of statistical methods
to quality and to industrial standardization. Published standard (2009). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=51651

82. ISO TC69/SCS–Secretariat: ISO 16269-4:2010 - Statistical interpretation of data – Part 4: Detection
and treatment of outliers. Published standard (2010). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=44396

83. ISO TC69/SCS–Secretariat: ISO 11453:1996 - Statistical interpretation of data – Tests and confidence
intervals relating to proportions. Published standard (2012). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=19405

84. ISO TC69/SCS–Secretariat: ISO 16269-7:2001 - Statistical interpretation of data – Part 7: Median –
Estimation and confidence intervals. Published standard (2012). http://www.iso.org/iso/catalogue_
detail.htm?csnumber=30709

85. ISO TC69/SCS–Secretariat: ISO 5479:1997 - Statistical interpretation of data – Tests for departure
from the normal distribution. Published standard (2012). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=22506

86. ISO TC69/SCS–Secretariat: ISO/TR 13519:2012 - Guidance on the development and use of ISO
statistical publications supported by software. Published standard (2012). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=53977

87. ISO TC69/SCS–Secretariat: ISO 16269-8:2004 - Statistical interpretation of data – Part 8:

determination of prediction intervals. Published standard (2013). http://www.iso.org/iso/catalogue_
detail.htm?csnumber=38154

88. ISO TC69/SCS–Secretariat: ISO 16269-6:2014 - Statistical interpretation of data – Part 6:

determination of statistical tolerance intervals. Published standard (2014). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=57191

89. ISO TC69/SCS–Secretariat: ISO 2602:1980 - Statistical interpretation of test results – Estimation of
the mean – Confidence interval. Published standard (2015). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=7585
90. ISO TC69/SCS–Secretariat: ISO 2854:1976 - Statistical interpretation of data – Techniques of
estimation and tests relating to means and variances. Published standard (2015). http://www.iso.org/
iso/catalogue_detail.htm?csnumber=7854

91. ISO TC69/SCS–Secretariat: ISO 28640:2010 - Random variate generation methods. Published
standard (2015). http://www.iso.org/iso/catalogue_detail.htm?csnumber=42333

92. ISO TC69/SCS–Secretariat: ISO 3301:1975 - Statistical interpretation of data – Comparison of two
means in the case of paired observations. Published standard (2015). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=8540

93. ISO TC69/SCS–Secretariat: ISO 3494:1976 - Statistical interpretation of data – Power of tests relating
to means and variances. Published standard (2015). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=8845

94. ISO/IEC: ISO/IEC Directives – Part 2:2011. Principles to structure and draft documents intended to
become International Standards, Technical Specifications or Publicly Available Specifications.
ISO/IEC Directives (2011). http://www.iec.ch/members_experts/refdocs/

95. ISO/IEC: ISO/IEC Directives – Part 1:2014. Official procedures to be followed when developing and
maintaining an International Standard and procedures specific to ISO. ISO/IEC Directives (2014).
http://www.iec.ch/members_experts/refdocs/

96. ISO/IEC JTC 1/SC 22 - Programming languages, their environments and system software interfaces:
ISO/IEC.9899:2011 Information technology – Programming languages – C. Published standard
(2011). http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=57853

97. ISO/IEC JTC 1/SC 22 - Programming languages, their environments and system software interfaces:
ISO/IEC 14882:2014 Information technology – Programming languages – C++. Published standard
(2014). http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=64029

98. Juran, J., Defeo, J.: Juran’s Quality Handbook: The Complete Guide to Performance Excellence.
McGraw-Hill, New York (2010)

99. Matsumoto, M., Nishimura, T.: Mersenne twister: a 623-dimensionally equidistributed uniform
pseudo-random number generator. ACM Trans. Modell. Comput. Simul. 8, 3–30 (1998)
[MATH][CrossRef]

100. Moen, R.D., Nolan, T.W., Provost, L.P.: Quality Improvement Through Planned Experimentation, 3rd
edn. McGraw-Hill, New York (2012)

101. Pathak, M.: Beginning Data Science with R. SpringerLink: Bücher. Springer, New York (2014)
[CrossRef]

102. Taguchi, G., Chowdhury, S., Wu, Y.: Taguchi’s Quality Engineering Handbook. Wiley, Hoboken
(2005)
[MATH]

Footnotes
1 Check the up-to-date list at http://www.iso.org/iso/home/about/iso_members.htm.

2 Frank Kaplan, see http://asq.org/about-asq/who-we-are/bio_box.html.

3 http://dictionary.cambridge.org/dictionary/british/quality.

4 http://en.wikipedia.org/wiki/Quality_(business).

5 For example, ISO 14000—Environmental management.

6 http://www.iso.org/iso/home/standards.htm.

7 Sometimes also in Russian.

8 Check http://www.iso.org/iso/home/store/catalogue_tc/catalogue_tc_browse.htm?commid=560992&
development=on.

9 http://www.iso.org/iso/home/store/catalogue_tc/catalogue_tc_browse.htm?commid=45202&published=
on.

10 There is an RSS channel for each standard, but the abstract is not there.

11 The grep function uses regular expressions to match text info.

12 The syntax package::function avoids conflict with other packages’ functions with the same name.

Part II
Statistics for Quality Control
 This part includes two chapters about basic Statistics, needed to perform
statistical quality control. The techniques explained throughout the book make
use of basic concepts such as mean, variance, sample, population, probability
distribution, etc. A practical approach is followed, without details about the
mathematical theory. Chapter 5 reviews descriptive statistics, probability, and
statistical inference with quality control examples illustrated with the R
software. Chapter 6 tackles the crucial task of appropriately taking samples from
a population, using intuitive examples and R capabilities.
© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_5

5. Modelling Quality with R

Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
This chapter provides the necessary background to understand the fundamental
ideas of descriptive and inferential statistics. In particular, the basic ideas and
tools used in the description both graphical and numerical, of the inherent
variability always present in real world are described. Additionally, some of the
most usual statistical distributions used in quality control, for both the discrete
and the continuous domains are introduced. Finally, the very important topic of
statistical inference contains many examples of specific applications of R to
solve these problems. The chapter also summarizes a selection of the ISO
standards available to help users in the practice of descriptive and inferential
statistic problems.

5.1 The Description of Variability

5.1.1 Background
The data we want to analyze are measurements or observations of a
characteristic in a set of items. These items can be people, products, parts, or
some individual unit in a set where we can observe the characteristic under
study. The characteristic we are observing is a variable as it is expected not to be
constant among the items. In terms of probability theory, a characteristic is a
random variable with a given probability distribution. It is important to keep in
mind that statistical analyses are based on the underlying probability distribution
of the data.
There are two main types of variables: quantitative variables and qualitative
variables. When the observed characteristic can be measured using some scale,
we have a quantitative variable. When the observed characteristic is a
description or a categorization of an item, we have a qualitative variable.
Furthermore, quantitative variables can be classified in continuous and discrete
variables. Continuous variables can take any value within an interval, for
example, temperature or length. Discrete variables can take a countable number
of values (finite or infinite); for example, the number of defective units within a
lot of finished products can only have the values 1, 2, 3, …, n where “n” is the
total number of units in the lot.
It is rather complex to describe variability. Scientists have developed two
general approaches that greatly help in this job: graphical representations (a
series of different charts) and numerical descriptions (numbers that condense the
information obtained from the measurements). From the point of view of quality
control, the description of variability is vital since a change in variability
behavior is the flag that indicates that something has changed in the process
under study.
There is a huge amount of literature regarding descriptive statistics,
probability, and inference, the topics of this chapter. Just to mention some
advisable ones, [13] is an easy-to-read one, while [12] and [3] contain deeper
explanations and theoretical background.
This chapter provides the necessary background to understand the
fundamental ideas of descriptive and inferential statistics. Sect. 5.1. develops the
basic ideas and tools used in the description both graphical and numerical, of the
inherent variability always present in real world. Sect. 5.2. introduces some of
the most usual statistical distributions used in quality control, for both the
discrete and the continuous domains. Sect. 5.3. develops the very important topic
of statistical inference. Sects. 5.1 to 5.3. contain many examples of specific
applications of R to solve these problems. Finally, Sect. 5.4 provides a selection
of the ISO standards available to help users in the practice of descriptive and
inferential statistic problems.

5.1.2 Graphical Description of Variation

There is a large number of charts that can be used for describing
variability/variation. Sometimes we simply want to condense the information
regardless of its behavior with respect to time (static vision); in other occasions,
the factor time is essential for understanding variability (dynamic vision). It is
very common that for a certain analysis more than a single chart has to be used;
each of them will help us in understanding a specific aspect of variation.

Histogram
The histogram, briefly introduced in Chapter 3, is one of the most popular charts
used in statistics; it is simple to construct and simple to understand. A histogram
is a bar chart used for describing continuous variables. This bar chart shows the
distribution of the measurements of variables. On the x-axis, each bar’s width
represents an interval of the possible values of a variable. The height of the bars
(that is, the y-axis) represents the frequency (relative or absolute) of the
measurements within each interval. The rule is that the area of the bars should be
proportional to the frequencies.
The histogram does not give us information about the behavior of the
measurements with respect to time; it is used to find the distribution of a
variable, that is:
Is the variable centered or asymmetric?
What is the variation like? Are the observations close to the central values,
or is it a spread distribution?
Is there any pattern that would prompt further analysis?
Is it a normal distribution?
To make a histogram of our data, we first determine the number of bins
(bars) that we are going to plot. Then, we decide on the width of the intervals
(usually the same for all intervals) and count the number of measurements within
each interval. Finally, we plot the bars. For intervals with equal widths, the
height of the bars will be equal to the frequencies.
The simplest way to create a histogram with R is by means of the standard
graphics capabilities, i.e., using the hist function. A simple call to this function
with the vector of data as argument plots the histogram. More elaborate charts
can be made using the lattice and ggplot2 packages, see Chapter 2 They are
specially useful if we need to visualize several histograms in the same chart as
we will see in the following example. Remember that histograms are one of the
seven basic quality control tools, see Chapter 3, you can see there more
examples.
Example 5.1.
Metal plates thickness. Histogram.
Table 5.1 contains two sets of 12 measurements, each one corresponding to
the thickness of a certain steel plate produced in 2 successive days. Nominal
thickness of this product is 0.75 in. Production equipment was readjusted after
Day 1 because the engineer in charge of the production line concluded that the
product was thicker than required. The following code creates a data frame in the
R workspace1:

Table 5.1 Thickness of a certain steel plate

Thickness Day Thickness Day
0.821 Day1 0.678 Day2
0.846 Day1 0.742 Day2
0.892 Day1 0.684 Day2
0.750 Day1 0.766 Day2
0.773 Day1 0.721 Day2
0.786 Day1 0.785 Day2
0.956 Day1 0.759 Day2
0.840 Day1 0.708 Day2
0.913 Day1 0.789 Day2
0.737 Day1 0.732 Day2
0.793 Day1 0.804 Day2
0.872 Day1 0.758 Day2

Although in the case of the data corresponding to Day1 it may seem clear
that the numbers are larger than 0.75 in (only two data points out of twelve are
not larger than that value) a histogram will put the situation even clearer. Fig. 5.1
shows the histogram generated with the following code:
Fig. 5.1 Histogram. A histogram provides an idea of the data distribution. The x axis represents the
magnitude we are measuring. The y axis is for the frequency of data. Thus, hints about the range, central
values and underlying probability distribution can be got

As was shown in Chapter 2, even though R can produce charts with very
little information (for example, in the above code only the first argument would
be needed), we can add options to get the desired result. Try the function with
and without the added options and see the differences. When doing so, we
explain the arguments used in the code. In this case:

main Sets the plot title

xlab Sets the label for the x axis
las Sets the axis labels orientation
col Sets the histogram bars fill color
border Sets the histogram bars border color

There are more graphical options, you can always check the documentation
of the plotting function (in this case, hist) and the graphical parameters options
(par).
Having this chart will give the engineer a good argument to backup his
feeling about Day1 data. A good idea would be to represent data before and after
the adjustment of the equipment, in the search for an evidence of the
improvement. A straightforward way to display charts for different groups in R
is using functions of the lattice package. The following code produces the chart
in Fig. 5.2.

Fig. 5.2 Histograms by groups. When different groups are in a data set, visualization by those groups is a
powerful tool

This is what the code does:

Load the lattice package;
Set graphical parameters:
– A black and white theme for monochrome printing;
– A gray background for the whole canvas;
– A white background for the panels (where plots are actually drawn);
Plot a histogram of thickness for each level of the day factor, variables
 which are in the plates data frame. Title and labels are added similarly to
the standard graphics histogram, and the type argument is to state that the y
axis is for counts instead of the default value (percent). Note that lattice
uses a formula expression to decide how to display plots in panels.
The histogram tells us in a very direct way that process has been successfully
adjusted in Day2 around a value close to the goal of 0.75 in. □ 

Run Chart
A run chart is a bidimensional chart where the x-axis represents a time line and
on the y-axis is plotted the variable that we want to monitor. These types of
charts are also called time-series charts when we have a time scale on the x-axis.
The scale of the x-axis may not necessarily be equally temporally shifted (for
example, the volume of some recipients whose production is sequential). Thus,
we will have a number of subgroups where a characteristic is measured, and we
have the order of the subgroups (notice that a subgroup may contain only one
element). Usually a centered line is plotted in a run chart. It may represent a
target, the mean of the data, or any other value. Run charts allow us to detect
patterns that can be indicative of changes in a process. Changes entail variability
and, thus less quality. In particular, if we detect cycles, trends, or shifts, we
should review our process. If we use the median as center line, then half of the
points are below the center line, and have of the points are above the center line.
If a process is in control, the position of each point with respect to the center line
is random. However, if the process is not in control, then non-random variation
appears in the run chart. In addition to the apparent patterns detected visualizing
the chart, additional numerical tests can be run to detect such non-random
variation.
Simple run charts can be plotted using R standard graphics simply plotting a
vector of data with the plot function, and then adding a center line using the
abline function. The qichart package provides a simple interface to plot run
charts and get run tests.

Example 5.2.
Metal plates thickness (cont.) Run chart.
The run chart corresponding to our example of plate thickness in Fig. 5.3 is
the simplest version of a run chart we can get with just the following two
expressions:
Fig. 5.3 Simple run chart. The run chart provides insights about changes in the process

type Sets the type of representation for each data point (lines, points, both)
pch Sets the symbol to be plotted at each point
h Sets the value in the vertical axis at which a horizontal line will be plotted
lwd Sets the line width

The run chart produced by the qicharts package in Fig. 5.4 provides further
information. In particular, the number of observations, the longest run, and the
crossings.
Fig. 5.4 Run chart with tests. Numerical tests can be run to detect process shifts

freeze Point in the x axis that divides the data set in two subsets
pre.text Text to annotate the pre-freeze period
post.text Text to annotate the post-freeze period
runvals (logical) whether to print the statistics from runs analysis

Note that the median has been computed only with day 1 data, as we have
frozen those data as in-control data. The numbers in brackets for the longest run
and the crossings are the limits to consider that the process is in control. In this
case, both criteria indicate that the distribution around the median cannot be
considered random. Details about the foundations of these tests can be found in
[2] and [15].
The sequential representation of the same data provides a substantially
different vision that what the histogram does. Now it seems to be evident that a
real change occurred after observation number 12 (corresponding to the
adjustment of the process after Day1)
If we analyze in detail the information provided by the histogram and the run
chart, we can conclude that none is better, both are complementary and
necessary to understand the whole picture of the situation. □ 

Other Important Charts in Quality Control

Tier Chart
A tier chart is similar to a run chart. We use tier charts when we have more
than one observation in each run (e.g., batches, days, etc.). With the tier chart we
can see short-term variation and long-term variation jointly in a single chart.
Short-term variation is the variation within each subgroup, whereas long-term
variation is the variation among all the groups.
To create a tier chart, we plot vertical lines at the position of each run from
the higher to the lower value. Then, the single values are plotted as a point or as
a vertical segment.
There is not a specific function to plot tier charts in R (neither in all
statistical software). Nevertheless, we can tune up a dot plot in order to get the
typical tier chart. The dotplot function in the lattice package does the trick
using the appropriate panel functions.

Example 5.3.
Metal plates thickness (cont.) Tier chart.
Now let us suppose that our plate thickness data for each day can be divided
into two groups of equal size, each corresponding to a different working shift.
The following code adds this information to the data frame:

We can plot a dot plot using the dotplot function in the lattice package,
and then add lines to get a typical tier chart. The following code plots the tier
chart in Fig. 5.5.
Fig. 5.5 Tier chart by shifts. A tier chart displays variability within and between groups

cex Magnifying factor for text and symbols

lty Type of line to be plotted (1 = solid)
panel.superpose Function to plot different elements by groups within a panel
subscripts indices of the data in the original data source to be used
panel.groups the name of the function to be used when superposing

Note that the panel argument in lattice plots allows to add a lot of
sophistication to multivariate plots. The value for this argument is an anonymous
function in which the elements in the panel are drawn by calling panel.*
functions. Usually, the first element is the counterpart of the container function,
in our example, panel.dotplot. See [14] for details about lattice graphics. □ 

Box-and-Whisker Plot
The box-and-whisker plot is also known as the box plot. It graphically
summarizes the distribution of a continuous variable. The sides of the box are
the first and third quartiles (25th and 75th percentile, respectively).2 Thus, inside
the box we have the middle 50 % of the data. The median is plotted as a line that
crosses the box. The extreme whisker values can be the maximum and minimum
of the data or other limits beyond which the data are considered outliers. The
limits are usually taken as:

where Q 1 and Q 3 are the first and third quartiles, respectively, and IQR is the
interquartile range (Q 3 − Q 1). Quantiles and IQR will be explained in detail in
Sect. 5.1.3.2. We can replace 1.5 with any value in the boxplot function of R.
The outliers are plotted beyond the whiskers as isolated points and can be
labeled to identify the index of the outliers. The box plot tells us if the
distribution is centered or biased (the position of the median with respect to the
rest of the data), if there are outliers (points outside the whiskers), or if the data
are close to the center values (small whiskers or boxes). This chart is especially
useful when we want to compare groups and check if there are differences
among them.
To create a boxplot with R we can use the boxplot standard function or other
packages functions like bwplot in the lattice package.

Example 5.4.
Metal plates thickness (cont.) Box plot.
For the example of the metal plates, we obtain the boxplot in Fig. 5.6 for the
whole data set using the following code:
Fig. 5.6 Box plot for all data. A box plot shows the distribution of the data

To compare groups using boxplots, we can add a formula to the function, see
Fig. 5.7:

Fig. 5.7 Box plot by groups. Displaying boxplots for different groups is a powerful visualization tool

We can even make comparisons using more than one grouping variable. The
lattice package is more appropriate for this task. In Fig. 5.8, we create a panel
for each day using the following expression:
Fig. 5.8 Lattice box plots. A deeper stratified analysis can be made using lattice graphics

 □ 

5.1.3 Numerical Description of Variation

Central Tendency
The central values of a data set are the simplest way to summarize the data.
A central measure value is a number around which the data vary. Three
important central tendency measures are mainly used in any statistical analysis:

1. The sample mean is the average value. This is the most widely used measure
due to its mathematical properties. The main inconvenience is that it is
sensitive to outliers (values far from the central values):

To calculate the mean with R we use the mean function.

Example 5.5.
Metal plates thickness (cont.) Mean of the metal plates thickness.
The mean function over a vector returns the average of the values in the
vector. If there are missing values (NA), the returned value is NA, unless we set the
na.rm argument to TRUE, see Chapter 2 for details.
 To make any computation by groups in R, we can use the tapply function, in
our example:

 □ 

2. The median is the value that divides the data into two halves: one containing
the higher values, the other containing the lower values. It is not influenced
by outliers. If we have an even number of data, the average value of the two
central values is taken.

The median function computes the median of a numeric vector in R.

Example 5.6.
Metal plates thickness (cont.) Median of the metal plates thickness.

 □ 

3. The mode is the most frequent value (or range of values in a continuous
variable). In a frequency table, it is the value that has the maximum
frequency. Even though the mean and the median are unique values, a data set
might have more than one mode. Many times this means that the sample data
come from merged populations, and we should measure categorical variables
in order to make an appropriate stratification.

There is not a function to calculate the mode with R. Instead, we need to
check the value or range of values that has the maximum frequency.

Example 5.7.
Metal plates thickness (cont.). Mode of the metal plates thickness.
For the example of the metal plates none of the values are repeated, so there
is no value that can be considered as the mode. However, we can check which
interval has the highest frequency. This is the modal interval. We can follow two
approaches. One is to divide the range into a number of intervals and check
which interval is the modal one, for example:

In the code above, we first divide the range of the variable into (rounded)
intervals with the cut function, where n is the number of observations. This is
quite a common rule to start, but there are other, for example to build a
histogram the Sturges’ formula [16] is used, type ?nclass.Sturges to see other.
Then we create an object with the frequency table, and finally the name of the
class whose frequency is maximal is showed.
The second approach is to use the intervals produced to construct a
histogram, which are taken with a rounded interval width. To do that, we simple
save the result of the hist function instead of plotting it, and then access to the
object in which that information was saved. We could also take the mid value of
the interval in order to provide a single value for the mode.
 In the object saved, the element mid contains the mid points of each interval,
and the counts element contains the frequencies.
Note that both approaches provide a similar result, which is definitely
enough for a descriptive analysis. Putting all together, if we compare the central
tendency measures and the graphical tools (histogram and box plot), we can see
how the asymmetry shown by the histogram, with a slight positive skew (longest
right tail) results in a mean higher than the median and the mode. Fig. 5.9 shows
the measures in the own histogram, produced with the following code.

Fig. 5.9 Histogram with central tendency measures. The mean of a sample data departs from the median to
the longest tail
  □ 

In the code above, we have:

1. saved the three central tendency measures in a list;

2. plotted a histogram of the thickness data;

3. plot a vertical line for each central tendency measure;

4. added a legend to the plot.

Variability
Variability is statistics’ reason for being [1]. In this section we will see how to
measure such variability. The variance is the most important measure of
variability due to its mathematical properties. It is the average squared distance
from the mean, and we will represent it by σ 2:

Another estimator for the variance of a population with better mathematical

properties than σ 2 is the sample variance, computed in a slightly different way
and represented by s 2:

To calculate the sample variance with R we use the var function.

Example 5.8.
Metal plates thickness (cont.) Variance.

 □ 

The variance is in square units compared with the mean. Hence, the standard
deviation is the most commonly used variability measure:

To calculate the standard deviation with R we use the sd function.

The sample standard deviation is not an unbiased estimator of the population
standard deviation. This is the reason why in some control charts described in
Chapter 9 one of the following estimates of the standard deviation is used:

where c 4 and d 2 are tabulated constants for a given n sample size, and R is
the sample range (see below).

Example 5.9.
Metal plates thickness (cont.) Standard deviation.
  □ 

The range (R) is the difference between the maximum value and the minimum
value, but it is strongly influenced by extreme values. Nevertheless, when we
have few data, it is used as a robust method to estimate the variability as outlined
above, see [12] for details:

To calculate the range with R, we need to get the maximum and minimum
using the range function, and then get the difference.

Example 5.10.
Metal plates thickness (cont.) Sample range.

Note how we can add any customized function on the fly to make any
computation by groups. The third argument of the tapply function in the above
code is a so-called anonymous function, always with an x as argument that is
then used within the function body. This strategy can also be used in vectorized
functions such as lapply and sapply, type ?lapply in the console and check the
documentation and examples on the topic.
  □ 

If we want to measure the variability around the median, the appropriate

measure is the median absolute deviation (MAD), that is:

Similarly to the median, we can compute the quartiles. These are the values
that divide the data into four parts. Thus the median is the second quartile (Q 2).
The first quartile (Q 1) is the value that has 25 % of data below it, and the third
quartile (Q 3) is the value above which 25 % of data remain. The interquartile
range (IQR) is a measure of variability that avoids the influence of outliers. This
range contains the middle 50 % of the data:

To calculate the quartiles with R we can use the quantile function. The
summary function over a numeric vector returns the five numbers summary (max,
min, Median, Q 1 and Q 3) and the mean. The IQR function computes the
interquartile range. The mad function returns the median absolute deviation.

Example 5.11.
Metal plates thickness (cont.) MAD, Quartiles and interquartile range.

 □ 
Frequency Tables
Usually, raw data becomes as difficult to interpret as the number of observations
increases. Thus, many times the first operation we do with the data is to build a
frequency table. For a discrete variable, the frequency of a value of the variable
is the number of times this particular value appears. For a discrete variable the
relative frequency is the fraction of times the value appears.
For continuous variables we need to arrange the data into classes. For
example, if in our example of plate thickness we want to count the number of
values above and below the nominal value of 0.75 in, we first create a new factor
variable.
We use the R function table to get frequency tables. To discretize a
numerical variable we can use the cut function or just combine logical
expressions and assignments.

Example 5.12.
Metal plates thickness (cont.) Frequency tables.

breaks Points where breaking the data

labels Labels for the resulting factor levels
include.lowest whether to make the first interval closed

 □ 

5.2 Probability Distributions

Characteristics of processes can be modeled as random variables. A random
variable is a variable quantity measured over a population for which we can
compute probabilities. In general, a random variable is determined by the set of
values that can take, and the probabilities associated with such values. Even
though “random variable” concept is theoretical, we can compute its expectation
(average or expected value) and variance, usually denoted by μ and σ 2 ,
respectively. Most processes can be modeled by means of known probability
distribution models whose properties have been previously studied. As a
consequence, probabilities and characteristics are usually predetermined and can
be easily computed. In this section, the most important probability distribution
models, both discrete and continuous, are explained.
In inferential statistics, a usual path to follow is, after getting sample data or
study the process, guess the most appropriate probability distribution model.
Then, given that theoretical model, test hypothesis, computations, predictions,
and decisions can be made . Such predictions usually involve the estimation of μ
and σ 2 at some extent. In what follows, X represents a random variable, x a
possible value for such value, and the probability that the variable
takes the value x. The probability distribution function is denoted by F(x) = P(X 
≤ x).

5.2.1 Discrete Distributions

In certain occasions the random variable can only take on certain values, e.g.,
“defective” and “non defective” (0, 1), “number of events in a period” (0, 1, 2,
…). In these cases the probability distribution is called a discrete distribution,
and a certain probability of occurrence is assigned to each of the possible states
of the variable.

Hypergeometric Distribution
The hypergeometric distribution is used when, within a finite population of N
items, there is a certain number D of them that belong to a certain category, e.g.,
defectives. The problem consists in calculating the probability of obtaining x
items of that special category if a random sample of n items is taken from the
population without replacement. In this case, we say that X follows a
hypergeometric distribution and we denote it by X ∼ H(N, n, p), where .
That probability can be calculated by
where, in general, is the binomial coefficient:

and n! is the factorial of n computed as .

A key idea regarding the hypergeometric distribution is that the probability
of obtaining an item of the special category is not constant with the sample
number. In other words, the successive extractions are not independent events.
This results comes from the fact that the population is finite. The mean and
variance of a hypergeometric random variable H(N, n, p) are:

To calculate the probability function of the hypergeometric distribution with

R we use the dhyper function. There are always four functions associated with a
probability distribution. For the hypergeometric distribution:
dhyper(x, m, n, k, …): Density of the probability distribution. For discrete
distributions, it is the probability that the random variable is equal to x, P(X 
= x);
phyper(q, m, n, k, …): Distribution function. It is the probability that the
random variable is less or equal to q, P(X ≤ q);
qhyper(p, m, n, k, …): Quantile function. It is the inverse of the
distribution function, i.e., the quantile x at which the probability that the
random variable is less than x equals p, .
rhyper(nn, m, n, k): random generation. Generates nn random variates of
a hypergeometric distribution.
The function to use depends on the question we want to answer, and
sometimes we need to use the properties of probability to get the appropriate
result, e.g., the probability of the complementary event. In addition to the main
function argument described above, the parameters of the probability distribution
model are to be specified within the function. In case of the hypergeometric
distribution, those parameters are: (1) m, corresponding to D in the formulae
above; (2) n, corresponding to N − D; and (3) k, corresponding to n in the
mathematical formulae. By default, functions phyper and qhyper return the
distribution function and its inverse, respectively, i.e., P(X ≤ x) and
 . Changing the argument lower.tail to FALSE, we get P(X > x) and
its inverse . Note that .

Example 5.13.
Metal plates thickness (cont.)
In our example, considering the population of the 24 plates produced in the 2
days, there is a total of seven plates thinner than the nominal value. If we take a
random sample of five plates, which is the probability of obtaining all of them
thicker than the nominal value?
If X is the random variable number of plates thinner than the nominal value,
the question to answer is: P(X = 0), and therefore we use the dhyper function. For
this particular random variable, N = 24, D = 7, and n = 5. Thus, we get the sought
probability as follows:

This is a relatively low probability (14.56 %).

 □ 

Binomial Distribution
The binomial distribution is the appropriate distribution to deal with proportions.
It is defined as the total number of successes in n independent trials. By
independent trial we mean the so-called Bernoulli trial, whose outcome can be
success or failure, with p being the probability of success assumed as constant.
The binomial distribution is absolutely determined by the parameters p and n,
and its probability function is

where and n! (n-factorial) is calculated as

In this case, we say that X follows a Binomial distribution and we denote it

by X ∼ B(n, p). The mean and variance of a binomial random variable B(n, p)
are:
 To compute probabilities of the binomial distribution with R we use the
*binom family of functions, being * one of d, p, q, or r for density (probability in
discrete distributions), distribution function, quantile, or random generation,
respectively.

Example 5.14.
Metal plates thickness (cont.) Binomial distribution.
In our example, in the second day just 6 out of the 12 plates (50 %) were
thinner than the nominal value. If we suppose that this rate remains constant (a
necessary hypothesis to consider the binomial distribution as applicable), it
could be possible to calculate the probability of obtaining x=0 plates thinner that
the nominal values in the next 5 units. Let X be the binomial random variable
number of thinner-than-nominal plates in a 5-sized sample . The
question is to compute P(X = 0), which can be done with the following
expression:

This probability, just around 3 %, indicates that this would be a rare event.
Another example would be to get the probability of having more than 1
thinner-than-nominal plates. We could add up the densities from 2 to 5, or
simply use the pbinom function to get P(X > 1) as follows:

 □ 

If, instead of having a fixed number of Bernoulli trials n, we have a continuous

process in which the random variable X measures the number of failures until
getting r successes, then X follows a negative binomial distribution NB(r, p). A
particular case of the binomial negative distribution when r = 1 leads to the
geometric distribution G(p), see [1].

Poisson Distribution
The Poisson distribution is useful to describe random processes where the events
occur at random and at a per unit basis, e.g., defects per unit surface, or defective
units per hour. In this distribution the rate of occurrence is supposed to be
constant and, theoretically, the number of events may range from zero to infinity.
The Poisson probability of observing x events in a sample unit for which the
average number of defects were λ, would be calculated as:

In this case, we say that X follows a Poisson distribution and we denote it by

X ∼ P(λ). The mean and variance of a Poisson random variable are both λ.
To calculate the probability function of the binomial distribution with R we
use the *pois family of functions, being * one of d, p, q, or r for density
(probability in discrete distributions), distribution function, quantile, or random
generation respectively.

Example 5.15.
Metal plates thickness (cont.)
If in our production of metal plates the rate of a certain surface defect were
0.2 defects/unit, which would be the probability that the next unit has zero
defects? Being X the random variable number of defects per unit ∼ P o(λ = 0. 2),
we want to know P(X = 0), computed in R as follows:

This is a pretty high probability (81,87 %).

 □ 

5.2.2 Continuous Distributions

The discrete distributions above take a finite or countable number of values, i.e.,
between one value and the following one the probability of the intermediate
values equals zero. In the continuous case, the random variable can take all
values in a continuous scale, e.g. 12.071 etc., being the only limitation the
number of decimal places available due to the measuring device. In these cases
the probability distribution is called a continuous distribution, and a certain
probability of occurrence is assigned to intervals of variation of the variable. For
this type of distributions, we mainly use the distribution function F(x). Thus, as
F(x) = P(X ≤ x), , where a and b are constants such that
a < b. We can also compute P(a ≤ X ≤ b) using the probability density function
f(x), defined as:

so that P(a ≤ X ≤ b) = ∫ a b f(x)d x. 

Normal Distribution
The normal distribution, also known as the Gaussian distribution, is the most
important probability distribution for continuous variables. It is determined by
two parameters, the mean μ (which in this case coincides with the median and
the mode) and the variance σ 2, and has the following probability density
function

whose shape is the one in Fig. 5.10.

Fig. 5.10 Normal distribution. Between the mean and three standard deviations fall 99.7 % of the data.
Between the mean and two standard deviations fall 95.5 % of the data. Between the mean and one standard
deviation fall 68.3 % of the data

In this case, we say that X follows a Normal distribution and we denote it by

X ∼ N(μ, σ). The relevance of the normal distribution is due to the central limit
theorem, which states that the sum of n random variables (regardless of its mean,
variance, and distribution) approximates a normal distribution as n increases.
Normally, a process is the result of many other subprocesses, and therefore the
normal distribution appears in many real-world processes such as human
measurements (e.g., height and weight of people) or industrial processes.
To calculate values of the distribution function of the normal distribution
with R we use the pnorm function. Similarly to the discrete distributions above,
we can get the density, without a clear interpretation in continuous variables but
useful to represent a normal distribution as in Fig. 5.10, as well as the quantile
for a given value of the distribution function. Random generation of normal data
can be done using the rnorm function.
Example 5.16.
Metal plates thickness (cont.)Normal distribution.
It would be a useful information about the change that has taken place in the
production process between day 1 and day 2 if we knew, in the long term, which
fraction of each of these two populations is lower than the nominal value of 0.75 
in.
In Sect. 5.3.2.4 we will demonstrate that these two populations are normal.
Once this has been done, we obtain the values of the mean and standard
deviation for each group, namely:

Then, we have two random variables X 1 thickness in day 1

and X 2 thickness in day 2 . The
results of the distribution function corresponding to x = 0. 75 for both random
variables are computed as follows:
P(X 1 ≤ x):

P(X 2 ≤ x):

These results clearly show that a shift has taken place in the process, since
the fraction of values raised from 11.40 % to 56.03 %.
 □ 

Other Important Continuous Distributions

Although the normal distribution is by far the most important one in the field of
quality control, many processes follow other different distributions. Among
these so-called non-normal distributions we may mention the uniform
distribution, the exponential distribution (that appears when we measure the time
until an event occurs in a Poisson process), the lognormal distribution (when the
logarithm of the random variable follows a normal distribution), the Weibull
distribution (extensively used in reliability analysis), etc. Other important type of
probability distributions are those used for inference, namely: the Student’s t, the
F distribution, and the χ 2 distribution. For a brief description and R functions for
these and other important continuous distributions, see [1].
Check also the documentation of the topic “Distributions” (type ?Distri-
butions to see a list of distributions supported in R). Basically, they work as the
distributions explained above, i.e., four functions for each distribution (d*, p*,
q*, r*), for example pweibull for the distribution function of a Weibull random
variable, and so on. You can also find a number of distributions defined in ISO
3534-1 [4] with details about their parameters, distribution function, mean,
variance, and other features.

Data Transformation
In many situations we have to deal with non-normal distributions. Typical non-
normal distributions are skewed, a clearly different behavior to the one of the
symmetrical one exhibited by the normal distribution.

Example 5.17.
Metal plates thickness (cont.) Non-normal data.
That could be the case, in our example, after some kind of change in the
production process in days 3, 4, and 5. The histogram departs from the
symmetric normal distribution, being better represented by a probability
distribution skewed to the right. First, let us add the new data to the data frame:

The histogram in Fig. 5.11 clearly depicts the new situation. The mean is
now far from the median, and a normal density (solid line) does not fit at all with
the histogram shape.
Fig. 5.11 Histogram of non-normal density data. The histogram shows a non-normal distribution with a
positive skew

 □ 

Typical control charts can work well with non-normal data, since their basic goal
is to identify anomalous departures from a stable (normal) process. We will see
in detail control charts in Chapter 9. Nevertheless, we are going to illustrate here
a data transformation using control charts for individual values.

Example 5.18.
Metal plates thickness (cont.) Non-normal data control chart.
The control chart using the non-normal data would be the one in Fig. 5.12.
We use the following code to plot this control chart, see Chapter 9 for details on
the qcc package and function.
Fig. 5.12 Individuals control chart of non-normal density data. In principle, we detect two out-of-limits
points and a broken pattern rule

 □ 

But if from the knowledge of the process one could conclude that this new
behavior will represent the process in the near future, it could be advisable to
adjust the data in order to account for the non normality.
The idea is to transform the data with the help of an algorithm in such a way
that after the transformation they will look like a true normal population. Then
the control chart, ideally conceived to detect assignable causes in a normal
population would be even more effective.
The simplest algorithm used for this purpose is called the Box-Cox
transformation. It consists in taking the original data to the power λ (or logλ if λ 
= 0). There exists an optimum value of this parameter for which the transformed
data mostly shows a normal behavior.
There are several ways for performing Box-Cox data transformation with R.
One is the boxcox function in the MASS package, which provides a plot of the
possible values of λ against their log-Likelihood, including a confidence interval.
Another one is to use the powerTransform function in the car package, which
provides numerical results. We will combine both approaches in the following
example.

Example 5.19.
Metal plates thickness (cont.) Non-normal data transformation.
The application of the transformation for the data corresponding to day 3, 4,
and 5 yields the plot in Fig. 5.13, generated with the following code:

Fig. 5.13 Box-Cox transformation plot. A 95 % confidence interval contains

The boxcox function expects a model in order to perform the transformation

on the response. To transform a single vector, a formula with the vector in the
left-hand side and the number 1 in the right-hand side should be provided as first
argument. The argument lambda is a vector of possible values of λ (by default a
sequence of numbers between − 2 and 2 in steps of 0.1). Sometimes it is
necessary to change this default values in order to see the whole confidence
interval in the plot, as it is the case due to the fact that the confidence lower limit
is lower than − 2. An exact point estimator for λ can be computed, but usually,
any value within the confidence interval would work, and a rounded value is
desirable. In this case, it seems that − 2 would be a good value of λ for
transforming our data to fit a normal distribution.
The following code provides more details using the powerTransform function
in the car package:
 The result of the powerTransform function is an object of class
powerTransform, containing a list of 13 elements that can be accessed as usual.
Generic functions summary and coef return details and the optimal value of λ
(rounded when argument round is set to TRUE), respectively. Thus, the estimated
optimal value for λ is − 3. 1506, but as the confidence interval contains the value
-2, this is considered to be the best value for the transformation.
Now we can proceed plotting the control chart for the transformed data, see
Fig. 5.14. First we save the transformed data using the bcPower function in the
car package.
Fig. 5.14 Control chart of transformed data. After the transformation into normal data no points are out of
the control limits

The fact that working with the transformed data yields no data points out of
the control limits is a clear indication that, actually, this process was statistically
in control.
 □ 

5.3 Inference About Distribution Parameters

Statistical inference is the branch of statistics whereby we arrive at conclusions
about a whole population through a sample of the population. We can make
inferences concerning several issues related to the data, for example, the
parameters of the probability distribution, the parameters of a given model that
explains the relationship between variables, goodness of fit to a probability
distribution, and differences between groups (e.g., regarding the mean or the
variance).
In this section, some basic statistical inference tools and techniques are
reviewed. In Sect. 5.3.1 confidence intervals and point estimation are explained.
Hypothesis-testing concepts are very important for every inference analysis. You
can read about them in Sect. 5.3.2.
5.3.1 Confidence Intervals

Sampling Distribution and Point Estimation

Through point estimation, one or more parameters of a population’s probability
distribution can be inferred using a sample. A function over the values of the
sample is called a statistic. For example, the sample mean is a statistic. When we
make inferences about the parameters of a population’s probability distribution,
we use statistics. These statistics have, in turn, a probability distribution. That is,
for every sample extracted from a given population, we have a value for the
statistic. In this way, we may build a new population of values (of the statistic)
that follows some probability distribution.
The probability distribution of a statistic is a sampling distribution, and the
properties of this distribution allow us to know if the statistic is a good estimator
of the parameter under study. To determine if a statistic is a good estimator, some
important properties are to be studied. A desirable property is unbiasedness. An
estimator is unbiased if its expectation equals the real value of the parameter.
To distinguish the actual value of a parameter from its estimator, a hat ( ) is
placed over the symbol of the estimator, e.g., is an estimator for the
variance σ 2. We will not explain in detail the properties of the statistics or
explain how to study sampling distributions, see [3] for in-depth explanations.
Instead, we will introduce some of the most important statistics for estimating
proportions, means, and variances.
For binomial distributions, the sample proportion is an unbiased estimator:

That is, the number of events (x) in n Bernoulli experiments over the number
of experiments.
For normal distributions the sample mean is an unbiased estimator of the
population mean

The sampling distribution of this mean is, in turn, a normal distribution with
the following parameters:

For the variance the unbiased estimator is the sample variance, defined as
 We have obtained an estimation for the parameter of interest for our process.
However, any estimation is linked to some uncertainty, and therefore we will
have to deal with some error level. To quantify this uncertainty, we use interval
estimation. Interval estimation consists in giving bounds for our estimation (LL
and UL , lower and upper limits). These limits are calculated so that we have
confidence in the fact that the real value of the parameter is contained within
them. This fact is stated as a confidence level and expressed as a percentage. The
confidence level reflects the percentage of times that the real value of the
parameter is assumed to be in the interval when repeating the sampling. Usually
the confidence level is represented by 100 × (1 −α) %, with α the confidence
coefficient. The confidence coefficient is a measure of the error in our
estimation. Common values for the confidence level are 99, 95, or 90 %,
corresponding, respectively, to α = 0. 01, α = 0. 05, and α = 0. 1.
A confidence interval is expressed as an inequality. If θ is a parameter, then
[LL, UL] means

Proportion Confidence Interval

When we are dealing with proportions, the binomial distribution is the
appropriate probability distribution to model a process. The typical application
of this model is the calculation of the fraction of nonconforming items in a
process.
Due to the central limit theorem, we can construct a confidence interval for
the proportion using the following formula:

where is the quantile of the standard normal distribution3 that leaves a

probability of α⁄2 on the right-hand side. This is the classical way to construct a
confidence interval for the proportion when the sample size n is large and is
not small (under these circumstances, the normal distribution can be used to
approximate the binomial distribution).
The R function prop.test uses another approximation approach (non-
parametric). Nevertheless, the binom.test function provides an exact confidence
interval for the probability of success, though this exact test might be too
conservative. We can compare methods and results using the binconf function in
the Hmisc package.

Example 5.20.
Metal plates thickness (cont.) Confidence interval for a proportion.
In Example 5.14 we estimated the proportion of plates thinner than the
nominal value as:

That was a point estimator, we can obtain a confidence interval with the
following code:

Thus, a confidence interval for the proportion of plates thinner than the
nominal value is [0.2538, 0.7462].
An exact test returns a wider (more conservative) interval:
 Other methods are used and compared with the following code, see the
documentation of the binconf function to find out more.

 □ 

Mean Confidence Interval

Thanks to the central limit theorem, for large sample sizes we can construct
confidence intervals for the mean of any distribution using the following
formula:

Usually, σ is unknown. In this case, instead of σ and the normal quantile z, we

must use the sample standard deviation s and Student’s t quantile with n − 1
degrees of freedom . A thorough explanation of this important concept is
outside the scope of this book. The degrees of freedom can be thought of as the
number of data minus the number of constraints used to estimate the parameter
under study. Therefore, the confidence interval takes the following form:
Example 5.21.
Metal plates thickness (cont.) Confidence interval for the mean.
Let’s calculate with R the 95 % confidence intervals for the mean of the
populations corresponding to Day1 and Day2. We use the function t.test, but
let us save the result in an object to focus just in the confidence interval, which is
the element conf.int of the returned list.

We can see that the two confidence intervals do not overlap, since the upper
limit of Day2 (0,7696) is smaller than the lower limit of Day1 (0,7885). We may
anticipate that a situation like this is a clear indication that the two means are
different. □ 

Variance Confidence Interval

Sometimes we need to find out if the variance of a process is within a given
range. Confidence intervals are a fast way to verify this issue. There are two
important differences between mean and variance confidence intervals:
Methods for variance are more sensitive to the normality assumption. Thus,
a normality test is advisable for validating results;
The statistic used to construct the confidence interval, χ 2 (chi square), for
the variance is not symmetric, unlike z or t. Therefore, the limits are not
symmetric with respect to the point estimator.
The formulae to construct the confidence interval are as follows:

Example 5.22.
Metal plates thickness (cont.) Confidence interval for the mean.
 Let’s calculate with R the 95 % confidence intervals for the variance of the
populations corresponding to Day1 and Day2. The function var.test does not
return a confidence interval for the variance of a population given a sample, but
for the ratio between variances. Nevertheless, we can easily construct the
confidence interval computing the limits in the formulae above.

 □ 

5.3.2 Hypothesis Testing

In statistical inference, hypothesis testing is intended to confirm or validate
some conjectures about the process we are analyzing. Importantly, these
hypotheses are related to the parameters of the probability distribution of the
data.
Hypothesis testing tries to find evidence about the refutability of the null
hypothesis H 0 using probability theory. We want to check if a new situation
represented by the alternative hypothesis (H 1) is arising. Subsequently, we will
reject the null hypothesis (H 1) if the data do not support it with “enough
evidence.” The threshold for enough evidence is decided by the analyst, and it is
expressed as a significance level α, similarly to the confidence intervals
explained in Sect. 5.3.1. A 0.05 significance level is a widely accepted value in
most cases, although other typical values are 0.01 or 0.1.
To verify whether the data support the alternative hypothesis, a statistic
(related to the underlying probability distribution given H 0) is calculated. If the
value of the statistic is within the rejection region, then the null hypothesis is
rejected. If the statistic is outside the rejection region, then we say that we do not
have enough evidence to accept the alternative hypothesis (perhaps it is true, but
the data do not support it).
Usually the refutability of the null hypothesis is assessed through the p-value
stemmed from the hypothesis test. If the p-value is larger than α, then H 0 should
not be rejected, otherwise H 0 must be rejected. The p-value is sometimes
interpreted as the probability that the null hypothesis is true. This interpretation
is not correct. The p-value is the probability of finding a more extreme sample
(in the sense of rejecting H 0) than the one that we are currently using to perform
the hypothesis test. So if the p-value is small, the probability of finding a more
extreme sample is small, and therefore the null hypothesis should be rejected.
Otherwise, if the p-value is large, the null hypothesis should not be rejected. The
concept of “large” is determined by the significance level α. In practice, if p < α,
then H 0 is rejected. Otherwise, H 0 is not rejected.
Thus, for instance, if the confidence level is 95 % (α = 0. 05) and the p-value
is smaller than 0.05, we do not accept the null hypothesis taking into account
empirical evidence provided by the sample at hand.
There are some functions in R to perform hypothesis tests, for example,
t.test for means, prop.test for proportions, var.test and bartlett.test for
variances, chisq.test for contingency table tests and goodness-of-fit tests,
poisson.test for Poisson distributions, binom.test for binomial distributions,
shapiro.test for normality tests. Usually, these functions also provide a
confidence interval for the parameter tested.

Means
This test can be stated in various ways; we could, for example, test the null
hypothesis that the mean of one population is equal to the mean of another
population; these kinds of tests are called two-sided tests. For example,

On the other hand, a one-sided test would look like

The purpose of this last test is to demonstrate that the mean of the second sample
is smaller than the first one.

Example 5.23.
Metal plates thickness (cont.) Hypothesis tests for the mean.
The application of this last kind of test to our data corresponding to Day1
and Day2 yields the following result:
 Since the p-value is very small (even lower than 0.01) we reject the null
hypothesis and consider that the Day1 values mean is larger than the Day2
mean. □ 

Variances
The null hypothesis of this test is usually stated as the ratio of the two variances
to be compared being equal to one. This is

On the other hand, a one-sided test would look like

The purpose of this last test is to demonstrate that the variance of the second
sample is smaller than the first one.

Example 5.24.
Metal plates thickness (cont.) Hypothesis tests for the variance.
The application of the two-sided test to Day1 and Day2 data yields the
following result.
 Since p-value is large (even larger than 0.1), we cannot reject the null
hypothesis, thus accepting that the two variances are equal. □ 

Proportions
The test to decide if two proportions, p 1 and p 2, differ in a hypothesized value,
p 0, is similar to the test of means, this is;

Example 5.25.
Metal plates thickness (cont.) Hypothesis tests for proportions.
In our example, in the second day just 6 out of the 12 plates were thinner
than the nominal value of 0.75, while in the first day only 1 out of the 12 plates
was. The question is, are these two proportions equal? The test of proportions
gives the following result:
 Since the p-value is greater than the 0.05 criterion, we cannot reject the null
hypothesis at a 95 % confidence level, thus accepting that the two proportions
are equal. The p-value is close to α = 0. 05, though. A larger sample could be
drawn to perform again the test if we suspect that there is a real difference
between proportions.
 □ 

Normality
In many situations it is necessary to check if the data under analysis follow a
normal distribution. The reason for this is that many tests have been developed
under the hypothesis that the data are normal; therefore, if this requirement is not
fulfilled by the data, the results of the test could be misleading.

Example 5.26.
Metal plates thickness (cont.) Hypothesis tests for normality.
There are several statistical tests to check normality, the most known is
called the Shapiro-Wilks test. The hypothesis are as follows:
H 0: The data are normally distributed
H 1: The data are not normally distributed
Let’s use this test to check normality for the 12 data points of Day1:
 Day1 data is normal. A graphical tool can also be used to check normality, it
is called a Quantile-Quantile plot (or Q-Q plot). In this plot, if data come from a
normal distribution, the points lie approximately along a straight line, see
Fig. 5.15, which has been obtained with the following expressions:

Fig. 5.15 Quantile-Quantile plot. The points are approximately in a straight line

If we try with Day2 we get a very similar result. But what happens with data
from days 3, 4, and 5? Let us give it a try:

These data are clearly non-normal, and we reject the null hypothesis with a
high confidence (the p-value is very small). We guessed that in Sect. 5.2.2.3 and
transformed the data using the Box-Cox transformation. In this case, non-
normality was clear just looking at the histogram in Fig. 5.11, but sometimes this
is not evident and we can confirm it with this simple test and a Quantile-Quantile
plot like the one in Fig. 5.16.

Fig. 5.16 Quantile-Quantile plot (non normal). The points clearly depart from the straight line

 □ 

5.4 ISO Standards for Quality Modeling with R

The complete list of Standards related to the topics addressed in this chapter can
be found from ISO/TC 69. The most relevant of them for the scope of this
chapter are in the following ones.
ISO 2602:1980 Statistical interpretation of test results—Estimation of
the mean—Confidence interval [8] . This International Standard describes
the estimation of the mean of a normal population on the basis of a series of
tests applied to a random sample of individuals drawn from this population
when the variance of the population is unknown and calculation of the
confidence interval for the population mean therefrom and from the
standard deviation.
ISO 2854:1976 Statistical interpretation of data—Techniques of
estimation and tests relating to means and variances [9] . This
International Standard describes the comparison of a variance with a given
value, estimation of a variance, comparison of two variances, estimation of
the ratio of two variances, and the same procedures for a mean with known
or unknown variance are dealt with.
ISO 3301:1975 Statistical interpretation of data—Comparison of two
means in the case of paired observations [10] . This International
Standard specifies a method for comparing the mean of a population of
differences with zero or any other preassigned value.
ISO 3494:1976 Statistical interpretation of data—Power of tests
relating to means and variances [11] . This International Standard deals
with comparison of a mean with a given value (variance known or
unknown), of two means (variance known or unknown), of a variance with
a given value, and of two variances, and gives sets of curves for these type
II risk for a given alternative and given size of sample and to determine the
size of sample to be selected for a given alternative and a given values of
type II risk.
ISO 5479:1997 Statistical interpretation of data—Tests for departure
from the normal distribution [7] . This International Standard gives
guidance on methods and tests for use in deciding whether or not the
hypothesis of a normal distribution should be rejected, assuming that the
observations are independent.
ISO 11453:1996 (and ISO 11453:1996/Cor 1:1999) Statistical
interpretation of data—Tests and confidence intervals relating to
proportions [6] . This International Standard Describes specific statistical
methods for the interpretation of data and for determining the two-sided
confidence limits for a desired confidence level.
ISO 3534-1:2006 - Statistics—Vocabulary and symbols—Part 1:
General statistical terms and terms used in probability [4] . This
Standard defines general statistical terms and terms used in probability,
including the description of a number of probability distributions, numerical
and graphical tools.
ISO 16269-4:2010 - Statistical interpretation of data—Part 4: Detection
and treatment of outliers [5] . This standard provides methods for the
detection and accommodation of outlier(s), including sound statistical
testing procedures and graphical data analysis methods for detecting
outliers in data obtained from measurement processes. Box-and-whiskers
plots are explained in detail in this standard.
References
1. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. In: Statistical Engineering for Process
Improvement. Use R!, vol. 36. Springer, New York (2012). http://www.springer.com/statistics/book/
978-1-4614-3651-5

2. Chen, Z.: A note on the runs test. Model Assist. Stat. Appl. 5, 73–77 (2010)

3. Hsu, H.: Shaum’s Outline of Probability, Random Variables and Random Processes. Shaum’s Outline
Series, 2nd edn. McGraw-Hill, New York (2010)

4. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). http://www.
iso.org/iso/catalogue_detail.htm?csnumber=40145

5. ISO TC69/SCS–Secretariat: ISO 16269-4:2010 - Statistical interpretation of data – Part 4: Detection

and treatment of outliers. Published standard (2010). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=44396

6. ISO TC69/SCS–Secretariat: ISO 11453:1996 - Statistical interpretation of data – Tests and confidence
intervals relating to proportions. Published standard (2012). http://www.iso.org/iso/catalogue_detail.
htm?csnumber=19405

7. ISO TC69/SCS–Secretariat: ISO 5479:1997 - Statistical interpretation of data – Tests for departure
from the normal distribution. Published standard (2012). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=22506

8. ISO TC69/SCS–Secretariat: ISO 2602:1980 - Statistical interpretation of test results – Estimation of the
mean – Confidence interval. Published standard (2015). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=7585

9. ISO TC69/SCS–Secretariat: ISO 2854:1976 - Statistical interpretation of data – Techniques of

estimation and tests relating to means and variances. Published standard (2015). http://www.iso.org/iso/
catalogue_detail.htm?csnumber=7854

10. ISO TC69/SCS–Secretariat: ISO 3301:1975 - Statistical interpretation of data – Comparison of two
means in the case of paired observations. Published standard (2015). http://www.iso.org/iso/catalogue_
detail.htm?csnumber=8540

11. ISO TC69/SCS–Secretariat: ISO 3494:1976 - Statistical interpretation of data – Power of tests relating
to means and variances. Published standard (2015). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=8845

12. Montgomery, D.: Statistical Quality Control, 7th edn. Wiley, New York (2012)

13. Rumsey, D.: Statistics For Dummies. Wiley, New York (2011)

14. Sarkar, D.: Lattice: Multivariate Data Visualization with R. Springer, New York (2008). http://lmdvr.r-
forge.r-project.org. ISBN 978-0-387-75968-5

15. Schilling, M.F.: The surprising predictability of long runs. Math. Mag. 85, 141–149 (2012)
[MATH][MathSciNet][CrossRef]
16. Sturges, H.A.: The choice of a class interval. J. Am. Stat. Assoc. 21, 65–66 (1926)
[CrossRef]

Footnotes
1 The data frame is also available in the SixSigma package.

2 Actually, a version of those quartiles called hinches, see [5] and?boxplot.stats.

3 A normal distribution with μ = 0 and σ = 1.

© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_6

6. Data Sampling for Quality Control with

R
Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
Statistical Quality Control tries to predict the behavior of a given process
through the collection of a subset of data coming from the performance of the
process. This chapter showcases the importance of sampling and describes the
most important techniques used to draw representative samples. An example
using R on how to plot Operating Characteristic (OC) curves and its application
to determine the sample size of groups within a sampling process is shown.
Finally, the ISO Standards related to sampling are summarized.

6.1 The Importance of Sampling

Process’ owner main responsibility is to assure that their process remains
under control, thus leading to products that comply with design specifications.
Among the several tasks required to fulfill this responsibility, one of the most
important consists in the observation of the process. By “observing the process”
we understand measuring it. There are different things that can be measured in a
process: finished product, product in an intermediate production stage, process
parameters, etc. Although all these things are very different to each other, all of
them have something in common: it is rarely possible to gather all the
information that is generated in the process. There are several reasons why this is
the case in general. In some cases, the cost of measuring an item is very high or
it takes a long time, in other cases the population is very large thus making it
impractical to measure thousands of items (no matter if the individual cost of
measuring were very low). Finally, in other situations, the measuring process is
destructive, which obviously forces the reduction in the number of observations.
Therefore, process’ owner have to take decisions based on limited pieces of
information obtained from the process. This is what we call a sample. A first
broad distinction can be made with regard to the purpose of sampling. Samples
can be taken to: (a) make a decision (normally accept/reject) about a lot of items;
or (b) make a decision about the state of control of a process. The first case will
be dealt in detail in Chapter 7, while the second one will be dealt in Chapter 9 in
the context of control charts.
Typically, lot populations are finite (composed of a limited number of items)
while process populations are infinite (very large number of items or even
theoretically infinite). The previous paragraph depicts the situation to which
process’ owner have to face every day; in order to make decisions about a certain
population of items, sampling is an inevitable tool they have to be aware of.
Sampling has a number of advantages over a complete—if possible— measuring
of the population: lower cost, quicker reaction time, etc. But sampling has one
major weakness; there is always an inherent error of such an observation
strategy. It could be understood as the price to be paid in order to get the
aforementioned advantages. Fortunately, this error can be estimated and bounds
can be set on it.

6.2 Different Kinds of Sampling

Depending on the nature of the population to be measured by means of a
sampling procedure, there may be a number of difficulties. An example will
illustrate this idea.

Example 6.1.
Pool Liquid Density.
Let us suppose we have to determine the average density of the liquid
contained in a large pool. Let us also suppose this liquid contains a certain solid
compound dissolved in the base liquid; as long as the solid material will slowly
tend to fall downwards forced by gravity, density will not be uniform at different
depths in the pool.
 If, based on ease of collection, we took samples from the surface of the pool,
the resulting average density so calculated would underestimate the real density
in the entire pool. In this case we can say that these samples do not adequately
represent the population parameter. □ 

The key element in a sampling procedure is to guarantee that the sample is

representative of the population. Then, any previous available information about
the population’s nature should be taken into account to develop the sampling
procedure.
In Example 6.1, the total number of observations should be distributed at
different depths in the pool. If there is no information about the population’s
nature, a simple random sampling procedure would proceed. Let us see this and
other sampling procedures and learn when to use all them.

6.2.1 Simple Random Sampling

In this kind of sampling every item in the population has the same probability of
being chosen for the sample. In order to select the sample items from the
population, random numbers are commonly used. In Chapter 5 we saw how to
generate random values for a random variable given its probability distribution,
e.g. normal, Poisson, etc. In general, uniform random numbers can be generated
between 0 and 1. In this way, the probability of an interval only depends on its
width. Taking the appropriate number of digits, random numbers in a given
range can be easily obtained. In practice, software packages select random
samples of a set using this strategy transparently for the user. Actually, random
variate generation is based on the fact that a uniform random variate is a sample
of a probability, and thus it can be used to sample values of a random variable
just looking for the quantile where the distribution function equals a uniform
random variate. The following simple example will illustrate how R will help
determine the sample.

Example 6.2.
Complex Bills.
A transactional process generates complex bills, consisting of many data
fields that have to be filled by the clerks. Thirty-two bills were produced
yesterday, and the supervisor wishes to check eight of them in detail. Which ones
should he choose? Table 6.1 shows all the population of bills. The data in
Table 6.1 is in the ss.data.bills data frame of the SixSigma package and it is
available when loading the package:
Table 6.1 Complex bills population
Bill no Clerk Errors Bill no Clerk Errors Bill no Clerk Errors
1 Mary 2 9 John 0 17 John 1
2 Mary 2 10 John 1 18 John 0
3 John 0 11 John 2 19 John 0
4 John 1 12 Mary 1 20 John 0
5 John 2 13 Mary 1 21 John 0
6 John 0 14 Mary 1 22 John 0
7 John 0 15 John 0 23 Mary 1
8 John 0 16 John 1 24 Mary 1

Thus, we have a data frame with 32 observation and three variables: nbill
for the bill identification; clerk for the clerk name and errors for the count of
errors in the bill.
We have to select eight random numbers between 1 and 32 and choose the
bills with the selected identifiers as sample elements. In other words, we need to
take a random sample of the nbill variable in the ss.data.bills data frame.
To do that with R, we use the sample function. It has three important arguments:
the vector that contains the population to be sampled, the sample size, and
whether the sample is with or without replacement. Replacement means that a
member of the population can be selected more than once. In this case, the
population is formed by the bill’s identifiers, the size is equal to eight, and the
sample is without replacement.

Note that in the above code we fix the seed using the set.seed function for
the sake of reproducibility of the example. In this way, anyone who runs the
code will get the same results. This is due to the fact that random numbers
generated with computers are actually pseudo-random because they are based on
an initial seed. In a production environment, the seed is rarely set, except in
specific conditions such as simulation experiments that should be verified by a
third party. Find out more about Random Number Generation (RNG) with R in
the documentation for the RNG topic (type?RNG in the R console). ISO 28640
Standard deals with random variate generation methods, see [8].
The result is that the supervisor has to select bills No. 27, 23, 29, 3, 2, 16, 11,
and 13. We can save the sample in a new data frame as a subset of the population
as follows:

Based on this sample, the average number of defects in the population should
be estimated (see Chapter 5) as 1 defect per bill:

 □ 

6.2.2 Stratified Sampling

If we analyze the sample that resulted from the simple random procedure
followed in Example 6.2 we see that bills No 2, 13, 23, and 27 correspond to
clerk Mary (50 % of the sample) while the four others correspond to clerk John
(the remaining 50 %). But in the total population of bills clerk Mary only
produced 8 bills out of 32 (25 %) while John produced 24 of 32 (75 %). If the
probability of introducing an error in a bill depended on the clerk, then the
sampling approach followed would be misleading. This a priori information—or
suspicion—could be made a part of the sampling procedure in the form of a
stratified strategy.
In this strategy, the population is divided into a number of strata and items
are selected from each stratum in the corresponding proportion. Note that we are
actually applying one of the seven quality control tools, see Chapter 3

Example 6.3.
Complex Bills (Cont.) Stratified sampling.
We can get in R the proportions of each clerk both in the population and in
the sample with the following code:

Thus, in order to stratify the sample, a 25 % of the sample, namely 2 bills,
will be taken from Mary’s production and a 75 % of the sample, namely 6 bills,
will be taken from John’s production. In R, we can first extract the bills from
each stratum:

and then draw a sample from each stratum of the appropriate size:

and finally save the sample into a new data frame:

 Thus, with the aid of R, we have selected two random items from Mary’s
stratum (1, 2, 12, 13, 14, 23, 24, and 27). The result is that the supervisor has to
select bills No. 24 and 14. Similarly, we have selected six random items from
John’s stratum (3, 4, 5, 6, 7, 8, 9, 10, 11, 15, 16, 17, 18, 19, 20, 21, 22, 25, 26,
28, 29, 30, 31, and 32). The result is that the supervisor has to select bills No. 32,
4, 31, 18, 10, and 15. Therefore, the number of errors in this sample of bills is:

Based on this sample, the average number of defects in the population should
be estimated as a weighted mean:

This can be computed in R using the weighted.mean function, which accepts

the values to be averaged as first argument, and the weights as the second
argument. In this case, the means and proportions for each clerk:
 This estimation is closer to population’s real average value (0.719). This
result is the expected one as long as the means are clearly different between the
two strata and the final weighting takes into account this difference in the final
sample value calculation. □ 

6.2.3 Cluster Sampling

In occasions, population data are grouped in clusters whose variability is
representative of the whole population variability. Then, it will only be necessary
to sample some of these clusters to get a reasonable idea of the population.

Example 6.4.
Complex Bills (Cont.) Cluster sampling.
Going back to the example of the bills, the clusters could be the different
customers to whom bills are made for. Measuring the number of defects for the
bills corresponding to one or two customers a good result could be obtained at a
much lower cost. □ 

6.2.4 Systematic Sampling

Sometimes it is easier to choose sample items at a constant interval period. This
is especially common in production lines where a stream of items are processed.

Example 6.5.
Complex Bills (Cont.) Systematic sampling.
In our example of the bills it was decided to take a sample of 8 items, so an
item must be selected every 32/8=4 bills. We only have to decide, at random,
which of the four first bills will be selected as the first one in the sample (let this
number be n) and then continue selecting (n + 4), (n + 8), etc. □ 

6.3 Sample Size, Test Power, and OC Curves with R

A control chart is, in its essence, nothing but a hypothesis test that is
performed online, sample after sample. See the foundations of hypothesis testing
as inference tool in Chapter 5. In any hypothesis test there exist two possibilities
of error:
1. The null hypothesis is true and is rejected (Error type I);

2. The null hypothesis is false and is not rejected (Error type II).

Fig. 6.1 illustrates these two possibilities for a typical control chart that keeps
track of sample average value, i.e., the X-bar chart, see Chapter 9. In this chart,
the null and alternative hypotheses are, respectively

Fig. 6.1 Error types. Different error types for an x-bar chart

If H 0 were true (left part of the figure), a sample A could fall outside of the
control limits thus leading us to reject H 0. On the other hand, if H 0 were false
(right part of the figure), a sample B could fall within the control limits thus
leading us to accept H 0.
The maximum probabilities for these situations to occur are denoted as α for
error type I and β for error type II, and they are set up in the design stage of any
hypothesis test. In particular, in Chapter 5 we showed that usually α is typically
set to 0.01, 0.05, or 0.1. It can be proved that there exists a specific relationship
among α, β, δ, and n (sample size) for every hypothesis test.
 For the case of control charts it is very important to know what the capability
of the chart will be for detecting a certain change in the process, e.g., in the
process mean. This capability of detecting a change of a certain magnitude is
called the “power” of the chart. It can be shown that

It is common practice to plot β as a function of δ for different sample sizes.

This plot is called the “operating characteristic (OC) curve.” Let’s show how to
construct these OC curves for the case of the X-bar control chart. Going back to
Fig. 6.1, β is the probability of a sample mean to fall within the control limits in
the case the population mean has shifted δ units from the original value.
Mathematically:

where NCD stands for “normal cumulative distribution.” Since X-bar

approaches a normal distribution with mean μ 0 and variance σ 2⁄n 1, and the
control limits are UCL = and LCL = , we have:

If we express δ in terms of σ, e.g., δ = γ σ we finally arrive at

We can easily plot OC curves for quality control with R. The function
oc.curves in the qcc package plots the operating characteristic curves for a ‘qcc’
object. We explain in detail objects whose class is qcc in Chapter 9. To illustrate
OC curves in this chapter, let us consider the example in Chapter 1

Example 6.6.
Pellets Density.
In this example, a set of 24 measurements for the density of a given material
are available, see Table 6.2. In order to plot OC curves for an X-bar chart, we
need the data organized in rational subgroups. Let us assume that every four
measurements make up a group. Therefore, there are six samples whose size is
four. With this information, we can create a qcc object as mentioned above. First,
we need to create the data and the qcc.groups object as follows:
Table 6.2 Pellets density data

10.6817 10.6040 10.5709 10.7858 10.7668 10.8101

10.6905 10.6079 10.5724 10.7736 11.0921 11.1023
11.0934 10.8530 10.6774 10.6712 10.6935 10.5669
10.8002 10.7607 10.5470 10.5555 10.5705 10.7723

Now we can create the qcc object, and plot the OC curves for that specific
control chart (see Fig. 6.2):

Fig. 6.2 OC curves. Each curve represents a function of the error type II probability as a function of the
deviation from the process mean that the control chart will be able to detect for different sample sizes

Fig 6.2 shows the representation of β for different sample sizes. This figure is
very useful as it is the basis for determining the sample size required for
detecting a given process shift with a desired probability. Furthermore, if we
save the result of the oc.curves function in an R object, we can explore the
complete set of data and look for the best sampling strategy. The first rows of the
matrix created are as follows:

and we can check the type II error for each sample size for a given deviation
from the current process mean. For example, if we want to detect a 1.5 standard
deviations depart from the mean:

With the current sample size (n = 4), the probability of false negatives β, i.e.,
being the process out of control the chart does not show a signal, is near 50 %.
We need groups of 10 to have this value around 0.04, i.e., a power of at least 95 
%. Note that we can choose the samples sizes to plot through the n argument of
the oc.curves function. On the other hand, the function also provides OC curves
for attributes control charts (see Chapter 9). □ 

6.4 ISO Standards for Sampling with R

These are the most relevant ISO Standards in relation to the topic addressed in
this chapter:
ISO 24153:2009 Random sampling and randomization procedures [7].
This International Standard defines procedures for random sampling and
randomization. Several methods are provided, including older approaches
based on mechanical devices, random numbers, etc. as well as more modern
ones based on algorithms for random numbers generations. Different
sampling strategies included random, stratified and cluster sampling are
described.
ISO 28640:2010 Random variate generation methods [8]. This
International Standard specifies typical algorithms by which it is possible to
generate numerical sequences as if they were real random variates. Two
annexes contain relevant information regarding random numbers tables and
several algorithms that can be used to generate pseudo-random numbers
with the aid of a computer.
ISO 3534-4:2014 Statistics—Vocabulary and symbols—Part 4: Survey
sampling [4] . This standard defines the terms used in the field of survey
sampling, but it is not constrained to surveys to the use of questionnaires.
Other standards related to the topics covered in this chapter are ISO 11462-2
[5] (SPC, Statistical Process Control) , ISO 7870-2 [6] (Shewhart control charts)
, and parts 1 and 2 of ISO 3534 (Vocabulary and symbols) [2, 3] .
There are also some books worth to reading, or just having them as
reference. Cochran [1] is a classic on sampling techniques; a more recent book is
the one by Lohr [9]; Montgomery [10] is cited in ISO 11462-2 [5] for sample
sizes calculation.

References
1. Cochran, W.: Sampling Techniques. Wiley Series in Probability and Mathematical Statistics: Applied
Probability and Statistics. Wiley, New York (1977)

2. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). http://www.
iso.org/iso/catalogue_detail.htm?csnumber=40145

3. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
csnumber=40147

4. ISO TC69/SC1–Terminology and Symbols: ISO 3534-4:2014 - Statistics – Vocabulary and symbols –
Part 4: Survey sampling. Published standard (2014). http://www.iso.org/iso/catalogue_detail.htm?
 csnumber=56154

5. ISO TC69/SC4–Applications of statistical methods in process management: ISO 11462-1:2010 -

Guidelines for implementation of statistical process control (SPC) – Part 2: Catalogue of tools and
techniques. Published standard (2010). http://www.iso.org/iso/home/store/catalogue_tc/catalogue_
detail.htm?csnumber=42719

6. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 - Control

charts – Part 2: Shewhart control charts. Published standard (2013). http://www.iso.org/iso/catalogue_
detail.htm?csnumber=40174

7. ISO TC69/SC5–Acceptance sampling: ISO 24153:2009 - Random sampling and randomization

procedures. Published standard (2015). http://www.iso.org/iso/catalogue_detail.htm?csnumber=42039

8. ISO TC69/SCS–Secretariat: ISO 28640:2010 - Random variate generation methods. Published standard
(2015). http://www.iso.org/iso/catalogue_detail.htm?csnumber=42333

9. Lohr, S.: Sampling: Design and Analysis. Advanced (Cengage Learning). Cengage Learning, Boston
(2009)

10. Montgomery, D.: Statistical Quality Control, 7th edn. Wiley Global Education, Hoboken (2012)

Footnotes
1 See the concept of sampling distribution in Chapter 5.

Part III
Delimiting and Assessing Quality
 This part includes two chapters covering how to compare the quality standards
with the process. In Chapter 7, acceptance sampling techniques are reviewed.
Sampling plans are obtained in order to fulfill requirements pertaining to
producer’s risk and consumer’s risk. The sampled items are assessed against an
attribute (defective, non defective) or a variable (a given continuous quality
characteristic). Chapter 8 starts establishing the quality specifications, i.e., the
voice of the customer (VoC), in order to compare with the voice of the process
(VoP) through Capability Analysis. Then, examples using R illustrate the
methods, and the ISO Standards related to these topics are discussed.
© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_7

7. Acceptance Sampling with R

Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
Undoubtedly, an effective but expensive way of providing conforming items to a
customer is making a complete inspection of all items before shipping. In an
ideal situation, a process designed to assure zero defects would not need
inspection at all. In practice, a compromise between these two extremes is
attained, and acceptance sampling is the quality control technique that allows
reducing the level of inspection according to the process performance. This
chapter shows how to apply acceptance sampling using R and the related ISO
standards.

7.1 Introduction
The basic problem associated with acceptance sampling is as follows:
whenever a company receives a shipment of products (typically raw material)
from a supplier a decision has to be made about the acceptance or rejection of
the product. In order to make such a decision, the company selects a sample out
of the lot, measures a specified quality characteristic and, based on the results of
the inspection decides among:
Accepting the lot (and send it to the production line);
 Rejecting the lot (and send it back to the supplier);
Take another sample before deciding (if results are not conclusive).
Sampling plans can be classified in attribute and variables. The attribute case
corresponds to the situation where the inspection simply determines if the item is
“good” or “bad,” this means it complies or not with a certain specification . This
kind of inspection is cheaper, but larger sample sizes are required. The variable
case, on the other hand, corresponds to the situation where the quality
characteristic is measured, thus allowing the inspector to decide based on the
value obtained. This kind of inspection is more expensive, but smaller sample
sizes are required.
In practice, the procedure to be followed is very simple; whenever the
company receives a shipment of N units, a random sample of n units is taken
from the lot and if d or less units happen to be considered as defective then the
lot is accepted. The procedure described corresponds to the case of attribute
inspection; the variable case is somewhat more sophisticated but conceptually
equivalent. For details on sampling methods, see Chapter 6
As any other hypothesis test, acceptance sampling is not a perfect tool but
just a useful one. There always exists the possibility of accepting a lot containing
too many defective items, as well as rejecting another one with very few
defectives. Fortunately, an upper bound for these two probabilities can be set up
in all cases by adequately selecting the parameters n and d.
This chapter provides the necessary background to understand the
fundamental ideas of acceptance sampling plans. Section 7.1 describes the
philosophy of the acceptance sampling problem. Sections 7.2 and 7.3,
respectively, develop the basic computational methods for attribute and variable
acceptance sampling as well as the way to implement them with R. Finally,
Sect.7.4 provides a selection of the ISO standards available to help users in the
practice of acceptance sampling.

7.2 Sampling Plans for Attributes

As it was stated in Sect. 7.1 a sample plan for attributes is defined by means
of the following three parameters:

N lot size;
n sample size (taken at random from the lot);
d maximum number of defective units in the sample for acceptance.
 The result of the inspection of the sample is:

x number of defectives found in the sample

The decision rule is:

1. Accept the lot if x ≤ d;

2. Reject the lot if x > d.

This kind of sampling plans, the simplest ones, are called single sampling
plans because the lot’s fate is decided based on the results of a unique sample.
There exist other kinds of sampling plans where two values of d are established;
the lot is accepted if ; rejected if x ≥ d upper; and a second sample taken if
. This kind of sampling plans are called double sampling plans.
The performance of a determined sampling plan is described by its operating
characteristic (OC) curve. This curve is a graphical representation of the
probability of accepting the lot as a function of the lot’s defective fraction. This
probability can be computed by means of the binomial probability distribution
(see Chapter 5), as long as the lot size be much larger than the sample size (n⁄N 
< 0. 1):

(7.1)
where P a stands for the probability of accepting the lot and p stands for the
lot’s fraction defective.

Example 7.1.
single sampling plan.
If we assume that n = 100 and d = 5, the resulting OC curve should look like
Figure 7.1, which has been produced with the following R code:
Fig. 7.1 OC curve for a simple sampling plan. The parameters of this OC curve are n = 100, d = 5

 □ 

There is a specific OC curve for every different sample plan; this means that if
we change either n or d, the curve will also change. But the general behavior
of all the curves will be similar; they start at P a  = 1 for p = 0, decrease more or
less rapidly as p increases, and finish at P a  = 0 for p = 1.
Two points in the OC curve are of special interest. The point of view of the
producer is that he requires a sampling plan having a high probability of
acceptance for a lot with a low (agreed) defective fraction. This low defective
fraction is called “acceptable quality level” (AQL), and the probability of such a
good quality lot being rejected is called “producer’s risk” (α). On the other hand,
the point of view of the customer is that he requires a sampling plan having a
high probability of rejection for a lot with a high (agreed) defective fraction.
This high defective fraction is called “lot tolerance percent defective” (LTPD),
and the probability of such a low quality lot being accepted is called
“consumer’s risk” (β). Figure 7.2 illustrates these two probabilities for a typical
OC curve.

Fig. 7.2 OC curve risks illustration. The fraction defective values AQL and LTPD are agreed. A
sampling plan yields then a producer’s risk α and a consumer’s risk β

The problem with acceptance sampling plans is then to choose n and d in

such a way that reasonable values for α and β are achieved for AQL and LTPD.
In mathematical terms, the problem is equivalent to solving the following system
of equations, where the unknowns are n and d:

The solution to this system is not easy and even not feasible all the times, so
that in general an acceptable solution will be as far as we could go. For
“acceptable” solution we understand a sampling plan that leads to actual α and β
values close enough to target values. Traditionally, nomographs (also called
nomograms) have been used to get approximate values of n and d given α and β
with paper and pencil, see, for example, [15]. Computational methods can be
used, though. R and a simple iterative method will greatly help in finding such
an acceptable solution. The iterative method we suggest is as follows:
 Step 1) Choose your target α and β values.
Step 2) Start with a sampling plan like n = 10 and d = 1. Calculate α and β
values with R. Normally, such an initial plan will give α actual close to α
target and .
Step 3) There are two possibilities: If , then change d to d + 1.
Calculate α and β values with R and repeat Step 3.
Or
If , then change n to n +δ n . Calculate α and β values
with R and repeat Step 3.
Normally, δ n should range between 10 to 50 depending on how large
be the difference between α actual and α target. Larger δ n correspond to
larger differences.
Step 4) If the solution happens to be feasible, final values obtained for α actual
and β actual will be close to their target values. If not, judgement will
have to be used in order to decide the best values for n and d.

Example 7.2.
Iterative method to select a sampling plan.
An example will illustrate this method. Let us suppose we need a sampling
plan that will provide us with α target = 0. 05 for AQL = 5 % and β target = 0. 10 for
LTPD = 16 %. The following R code runs the method above getting the result
in Table 7.1.
Table 7.1 Iterative sampling plan selection method
Iteration n d α β Decision
1 10 1 0.09 0.51 Increase n
2 25 1 0.36 0.07 Increase d
3 25 2 0.13 0.21 Increase d
4 25 3 0.03 0.42 Increase n
5 40 3 0.14 0.10 Increase d
6 40 4 0.05 0.21 Increase n
7 55 4 0.14 0.05 Increase d
8 55 5 0.06 0.11 Increase d
9 55 6 0.02 0.20 Increase n
10 70 6 0.06 0.05 …
 Note that we fix 10 iterations and make a decision depending on which risk
is farther away from the target. A customized function can be programmed
taking into account the specific problem at hand. In the eighth iteration we get a
plan that yields producer’s and customer’s risks very close to the targets.
 □ 

In addition to the iterative method described above, we can use the

AcceptanceSampling R package [14] . Function OC2c plots OC curves for attribute
acceptance sampling plans, and function find.plan finds a simple sampling plan
with smallest sample size.

Example 7.3.
OC curve and acceptance sampling plan with the AcceptanceSampling R
package.
The following code gets the OC curve for the sampling plan in Example 7.1,
i.e., with n = 100 and d = 5. The result is in Fig. 7.3.
Fig. 7.3 OC curve with the AcceptanceSampling package. Graphical parameters can be added to
customize the plot

Now let us compute the sampling plan proposed by the find.plan function
for the requirements in Example 7.2, i.e., α = 0. 05 for AQL = 5 % and β = 0. 10
for LTPD = 16 %. The arguments of the find.plan function are the producer risk
point (PRP) and consumer risk point (CRP). Each argument should be a vector
of two numbers, the first number being AQL or LTPD, and the second one being
the corresponding probability of acceptance, i.e., 1 −α and β, respectively.
 Thus, the proposed plan is drawing samples of size 64 and reject the lot if
there are seven or more defectives. We can create an object of class OC2c for this
plan in order to plot the OC curve (see Fig. 7.4) and assess its performance. The
assess function returns the plan and its probabilities of acceptance, and
compares them with the required ones.

Fig. 7.4 OC curve for the found plan. The found plan can be plotted and assessed
 Note that the result is slightly different to the one obtained in Example 7.2.
Both are close to risk targets and probably acceptable approximations for both
parts.
 □ 

Throughout this section, the assumption was made that the binomial probability
distribution could be used for the purpose of calculating the probabilities
associated with the sampling process. As it was stated before, this assumption
holds as long as the sample size be small in comparison with the lot size (n < 
< N). This will therefore guarantee that the probability of finding a defect in one
sampled item will remain approximately constant. But in the general case this
assumption is not true, and the more accurate hypergeometric distribution, which
was described in Chapter 5 should be employed instead. Nevertheless, the
methods are the same, just changing probability functions to the appropriate
distribution. As for the AcceptanceSampling package, functions OC2c and
find.plan accept a type argument whose possible values are binomial,
hypergeom, poisson, and normal (the latter for sampling plans for variables in the
next section).

7.3 Sampling Plans for Variables

A variable sampling plan corresponds to the situation where a certain quality
characteristic is measured in a continuous scale for every item selected from the
lot. The distinction between a “good” and a “bad” individual value results from
its comparison with the specified limit. Technical specifications may incorporate
a lower (LSL) or an upper (USL) specification limit. In some cases two
simultaneous limits may exist. But in variable sampling we are not specially
interested in individual values. What is done is to compute the mean of the
measured values and calculate the statistic

(7.2)
where USL is the Upper Specification Limit, is the sample mean, and σ is
the process standard deviation. This case corresponds to the situation where only
the USL exists, and the standard deviation of the distribution of the individual
values is assumed to be known. If the so calculated Z USL value is larger than k (a
value known as “acceptability constant”), then the lot may be accepted. Fig. 7.5
illustrates this concept.

Fig. 7.5 Variables acceptance sampling illustration. Maximum allowable defective fraction and
acceptability constant

In a way equivalent to what is done for attribute sampling, OC curves are

generated for variable sampling. The two elements that constitute a variable
sampling plan, namely: n, the sample size, and k, the acceptability constant,
are calculated to assure that:

a) For a lot with a low (agreed) fraction defective (AQL), the probability of
rejection (producer’s risk) is equal to α;

b) For a lot with a high (agreed) fraction defective (LTPD), the probability of
acceptance (consumer’s risk) is equal to β.

Conceptually, the situation is illustrated in Figures 7.6 and 7.7. Figure 7.6
corresponds to the situation where the population has a defective fraction p 1
equal to AQL, whereas Figure 7.7 corresponds to the situation where the
population has a defective fraction p 2 equal to LTPD.

Fig. 7.6 Probability of acceptance when p=AQL. Probability of acceptance for a population with
defective fraction equal to AQL
Fig. 7.7 Probability of acceptance when p=LPTD. Probability of acceptance for a population with
defective fraction equal to LPTD

Sample size and the acceptability constant are chosen in such a way that the
probability of acceptance approximately corresponds to (1 −α) for Figure 7.6 and
(β) for Figure 7.7. Note that sample size has a clear effect on sample mean
distribution variance, as long as

The resulting formulae corresponding to the case when there is a single

specification limit and the standard deviation is known are:

where:

and:

Z 1 is the (1 − p 1) × 100 percentile of the standard normal distribution;

p 1 is the AQL;
Z 2 is the (1 − p 2) × 100 percentile of the standard normal distribution;
p 2 is the LTPD;
Z α is the (1 −α) × 100 percentile of the standard normal distribution;
 Z β is the (1 −β) × 100 percentile of the standard normal distribution.

Example 7.4.
Variable acceptance sampling. Known standard deviation.
A simple example will illustrate how these formulae are implemented in R.
Let us suppose we wish to develop a variable sample plan where:

AQL p 1 = 1 %
LTPD p 2 = 5 %
producer’s risk α = 5 %
consumer’s risk β = 10 %
σ assumed known

To find the sampling plan for these requirements, we use again the find.plan
function. In this case, we need to add a new argument to the function call,
namely type, in order to get the sampling plan for continuous variables.

Thus, the sampling plan is n = 19, k = 1. 949. □ 

In general, for a quality characteristic with only an upper specification limit

(USL), we would proceed with the implementation of an acceptance plan as
follows:

1. Take random samples of n items from each lot;

2.
Compute the sample mean ;

3. Compute the Z USL value in Eq. (7.2);

4. Compare Z USL with k;

5. Decide whether to accept (Z USL > k) or reject (Z USL ≤ k) the lot.

Example 7.5.
Variable acceptance sampling. Implementation for the metal plates thickness
example.
A numerical example will illustrate the procedure. Let us simulate the
process described in Example 5.1 of Chapter 5. The quality characteristic was
the thickness of a certain steel plate produced in a manufacturing plant. Nominal
thickness of this product was 0.75 in. Let us assume that the standard deviation
is known and equal to 0.05, and the USL is 1 in. A simulated sample of this
process can be obtained with the following code:

Now we compute the sample mean and the Z USL value as follows:

As Z USL > k, this lot must be accepted. We suggest the reader to run this
simulation for different values of the mean and standard deviation and see how
lots are rejected as mean shifts or increase in variation occur. The following
convenient function helps automate this decision process1:

Thus, if a new sample whose mean is 0.92 is drawn, then the lot should be
rejected.
 □ 

In the example above, we assumed that the standard deviation of the population
was known. If this is not the case, the sampling plan must be more conservative
as we have less knowledge about the process. The resulting formulae
corresponding to the case when there is a single specification limit and the
standard deviation is unknown are:

Example 7.6.
Variable acceptance sampling (cont). Unknown standard deviation.
If the standard deviation in Example 7.4 is unknown, then the sampling plan
corresponding with the conditions:

AQL p 1 = 1 %
LTPD p 2 = 5 %
producer’s risk α = 5 %
consumer’s risk β = 10 %
 σ assumed unknown

is obtained with the following code:

Notice that we only have to change the s.type argument in the find.plan
function (by default "known"). Now we need much more items to be sampled in
order to achieve the objectives. We can simulate a new sample from our
production process, but now we need to estimate σ in Eq. (7.2) through the
sample standard deviation s.

 □ 

In this chapter we have assumed a smaller-the-better quality characteristic. In the

case when the quality characteristic is a larger-the-better one, we have only a
lower specification limit (LSL), and the procedure is the same we have
explained so far just using instead of Eq. (7.2). When both limits
exist (nominal-is-best characteristic), both Z USL and Z LSL must be larger than k
in order to accept the lot. The computation of k for different situations may vary
depending on the software used and the specific model that applies. Some of
these models can be found in the corresponding ISO standards (see the following
section) and all of them can be implemented with R similarly to what we have
done in this chapter. In addition to numerical computations, ISO standards
provide a set of tabulated sampling plans given the most common values for
producer and customer risks, AQL, and LTPD. Moreover, different rules to
change from normal to reduced and rigorous sampling can also be applied in
sequential plans.
We have focused on simple sampling plans for both attributes and variables.
There exist more complex sampling plans which are out of the scope of this
book, such as double, multiple, or sequential plans. Double and multiple
sampling plans for attributes can be created and assessed with the
AcceptanceSampling R package just providing sample sizes n i and maximum
number of defects d i for each i stage as vectors to the OC2c function.

7.4 ISO Standards for Acceptance Sampling and R

The complete list of Standards related to the topic addressed in this chapter
can be found from Subcommittee SC5, ISO/TC 69/SC 5—acceptance sampling.
The most relevant of them are in the following.
ISO 2859-1:1999 Sampling procedures for inspection by attributes –
Part 1: Sampling schemes indexed by acceptance quality limit (AQL)
for lot-by-lot inspection [8] . This International Standard specifies an
acceptance sampling system for inspection by attributes. It is indexed in
terms of the AQL. Its purpose is to induce a supplier through the economic
and psychological pressure of lot non-acceptance to maintain a process
average at least as good as the specified AQL, while at the same time
providing an upper limit for the risk to the consumer of accepting the
occasional poor lot.
ISO 2859-3:2005 Sampling procedures for inspection by attributes –
Part 3: Skip-lot sampling procedures [9] .
This International Standard specifies generic skip-lot sampling
procedures for acceptance inspection by attributes. The purpose of these
procedures is to provide a way of reducing the inspection effort on products
of high quality submitted by a supplier who has a satisfactory quality
assurance system and effective quality controls. The reduction in inspection
effort is achieved by determining at random, with a specified probability,
whether a lot presented for inspection will be accepted without inspection.
ISO 2859-5:2005 Sampling procedures for inspection by attributes –
Part 5: System of sequential sampling plans indexed by AQL for lot-by-
lot inspection [10] .
This International Standard contains sequential sampling schemes that
supplement the ISO 2859-1 acceptance sampling system for inspection by
attributes, whereby a supplier, through the economic and psychological
pressure of lot non-acceptance, can maintain a process average at least as
good as the specified AQL, while at the same time provide an upper limit
for the risk to the consumer of accepting the occasional poor lot.
ISO 3951-1:2013 Sampling procedures for inspection by variables –
Part 1: Specification for single sampling plans indexed by AQL for lot-
by-lot inspection for a single quality characteristic and a single AQL [6]
.
This International Standard specifies an acceptance sampling system of
single sampling plans for inspection by variables. It is indexed in terms of
the AQL and is designed for users who have simple requirements.
ISO 3951-2:2013 Sampling procedures for inspection by variables –
Part 2: General specification for single sampling plans indexed by AQL
for lot-by-lot inspection of independent quality characteristics [7] .
This International Standard specifies an acceptance sampling system of
single sampling plans for inspection by variables. It is indexed in terms of
the AQL and is of a technical nature, aimed at users who are already
familiar with sampling by variables or who have complicated requirements.
ISO 3951-3:2007 Sampling procedures for inspection by variables –
Part 3: Double sampling schemes indexed by AQL for lot-by-lot
inspection [5] .
This International Standard specifies an acceptance sampling system of
double sampling schemes for inspection by variables for percent
nonconforming. It is indexed in terms of the AQL.
ISO 3951-5:2006 Sampling procedures for inspection by variables –
Part 5: Sequential sampling plans indexed by AQL for inspection by
variables (known standard deviation) [11] .
This International Standard specifies a system of sequential sampling
plans (schemes) for lot-by-lot inspection by variables. The schemes are
 indexed in terms of a preferred series of AQL values, ranging from 0.01 to
10, which are defined in terms of percent nonconforming items. The
schemes are designed to be applied to a continuing series of lots.
Other standards useful for acceptance sampling are ISO 24153 [12] (Random
sampling and randomization procedures), ISO 3534-4 [4] (vocabulary and
symbols about sampling) and parts 1 and 2 of ISO 3534 (Vocabulary and
symbols about Statistics, Probability, and Applied Statistics) [2, 3] .
Acceptance Sampling can be usually found in any SPC book, see, for
example, the ones by Juran [13], Ishikawa [1], or Montgomery [15]. A more
complete book, devoted entirely to Acceptance Sampling, is the one by Schilling
[16] where you can find more details about acceptance sampling techniques.

References
1. Ishikawa, K.: Guide to Quality Control. Asian Productivity Organisation, Tokyo (1991)

2. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). url http://
www.iso.org/iso/catalogue_detail.htm?csnumber=40145

3. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=40147

4. ISO TC69/SC1–Terminology and Symbols: ISO 3534-4:2014 - Statistics – Vocabulary and symbols –
Part 4: Survey sampling. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=56154

5. ISO TC69/SC5–Acceptance sampling: ISO 3951-3:2007 - Sampling procedures for inspection by

variables – Part 3: Double sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot
inspection. Published standard (2010). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=
40556

6. ISO TC69/SC5–Acceptance sampling: ISO 3951-1:2013 - Sampling procedures for inspection by

variables – Part 1: Specification for single sampling plans indexed by acceptance quality limit (AQL)
for lot-by-lot inspection for a single quality characteristic and a single AQL. Published standard (2013).
url http://www.iso.org/iso/catalogue_detail.htm?csnumber=57490

7. ISO TC69/SC5–Acceptance sampling: ISO 3951-2:2013 - Sampling procedures for inspection by

variables – Part 2: General specification for single sampling plans indexed by acceptance quality limit
(AQL) for lot-by-lot inspection of independent quality characteristics. Published standard (2013).
url http://www.iso.org/iso/catalogue_detail.htm?csnumber=57491

8. ISO TC69/SC5–Acceptance sampling: ISO 2859-1:1999 - Sampling procedures for inspection by

attributes – Part 1: Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-lot
inspection. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=1141
9. ISO TC69/SC5–Acceptance sampling: ISO 2859-3:2005 - Sampling procedures for inspection by
attributes – Part 3: Skip-lot sampling procedures. Published standard (2014). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=34684

10. ISO TC69/SC5–Acceptance sampling: ISO 2859-5:2005 - Sampling procedures for inspection by
attributes – Part 5: System of sequential sampling plans indexed by acceptance quality limit (AQL) for
lot-by-lot inspection. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=39295

11. ISO TC69/SC5–Acceptance sampling: ISO 3951-5:2006 - Sampling procedures for inspection by
variables – Part 5: Sequential sampling plans indexed by acceptance quality limit (AQL) for inspection
by variables (known standard deviation). Published standard (2014). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=39294

12. ISO TC69/SC5–Acceptance sampling: ISO 24153:2009 - Random sampling and randomization
procedures. Published standard (2015). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=
42039

13. Juran, J., Gryna, F.: Juran’s Quality Control Handbook. Industrial Engineering Series. McGraw-Hill,
New York (1988)

14. Kiermeier, A.: Visualizing and assessing acceptance sampling plans: the R package
AcceptanceSampling. J. Stat. Softw. 26(6), 1–20 (2008). url http://www.jstatsoft.org/v26/i06/

15. Montgomery, D.: Statistical Quality Control, 7th edn. Wiley Global Education, New York (2012)

16. Schilling, E., Neubauer, D.: Acceptance Sampling in Quality Control. Statistics: A Series of Textbooks
and Monographs, 2nd edn. Taylor & Francis, Boca Raton (2009)

Footnotes
1 It is relatively easy to implement this in an on-line process via an R interface like, Shiny (http://www.
shiny.rstudio.com), possibly using automatically recorded measurements.

© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_8

8. Quality Specifications and Process

Capability Analysis with R
Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
In order to assess quality, specification limits are to be established. In this
chapter a method to set specification limits taking into account customers’ and
producer’s loss is presented. Furthermore, the specification limits are the voice
of the customer, and quality can be assessed by comparing it with the voice of
the process, that is, its natural limits. Capability indices and the study of long-
and short-term variability do the job.

8.1 Introduction
In Chapter 4 we reviewed some definitions of Quality from several standpoints.
The fulfillment of some specifications seemed to be a generally accepted criteria
to assess Quality. In this chapter we provide some guidelines and resources to
establish such specifications, and how to measure and analyze the capability of a
process to fulfill them. The idea is to fix some specification limits, Upper and/or
Lower (USL and LSL respectively), and compare them with the natural limits of
the process. These natural limits are normally the same used as Upper and Lower
Control Limits (UCL and LCL respectively) in the Control Charts explained in
Chapter 9 This is the first caution a Quality analyst must take: making clear the
difference between Specification Limits and Control Limits. Specification Limits
are the voice of the customer1 (VoC). Natural limits (or Control Limits) are the
voice of the process (VoP). Thus, the capability of a process is a way of
assessing how the VoP is taking care of the VoC. Capability analysis quantifies
this fact through graphical tools, capability indices, and other metrics, thereby
measuring the Quality of our process.

8.2 Tolerance Limits and Specifications Design

In this section, we focus on tolerance from the point of view of customer
specifications. Note that specification limits are independent of the process.

8.2.1 The Voice of the Customer

The setting of the specification limits is a crucial task in Quality Control. Such
specifications should stem from the customer during the design phase of a
product or service, involving not only engineering departments, but also other
customer-related teams, especially from marketing. In this chapter we assume
that the specification limits have already been set, either by the client or during
the design phase. We refer to them as upper specification limit (USL) and lower
specification limit (LSL). Processes can be classified into three categories
according to their specification limits, namely:
Smaller-the-better, when the process has only a USL;
Larger-the-better, when the process has only an LSL;
Nominal-is-best, when the process has both USL and LSL.
Modern Quality Control techniques for the establishment of appropriate
specifications include quality function deployment (QFD) and robust parameter
design (RPD). Details on RPD can be consulted in ISO 16336 [11] . ISO
technical committee TC69 SC8 is also developing a standard on QFD, at the
time this is written in DIS stage.2 It is also a usual tool in design for Six Sigma
(DFSS) methodologies, see, for example, See, [17]

8.2.2 Process Tolerance

Customer specification limits can be taken directly as specification limits in a
production environment.3 However, an economic approach can be followed in
order to take into account the producer’s and the customer’s perception of loss
due to poor quality. The basis is the Taguchi Loss Function. Details about Loss
Function analysis can be found in Chapter 4 of [1] or in the free on-line
textbook [12]. According to Taguchi method, if there is a target value Y 0 for a
given quality characteristic Y, departures from this target produce non-perfect
products and, therefore, there is a loss for the society due to poor quality. This
loss can be modeled as a quadratic function of the form:

where Y is the quality characteristic and Y 0 is the target value. For a specific
process, k is obtained as:

where Δ c is the tolerance for the characteristic Y from the point of view of the
customer, and L c is the loss for the customer when the characteristic is just out
of specification, i.e., Y 0 ±Δ c . Notice that the loss is zero when the process is
exactly at the target, and increases as the value of the characteristic departs from
this target, see Fig. 8.1.

Fig. 8.1 Taguchi’s loss function and specification design. The farther the target, the higher the loss. The
function is determined by the cost for the customer at the specification limits. The manufacturing limits are
then fixed at the point in which the function equals the cost for the producer

The cost of poor quality is usually higher for the customer than for the
producer. The reason is that a product or service that is delivered to a customer is
usually compound of a series of items or components. Thus, if a component of
the product or service is defective, then the whole thing must be repaired or
replaced. Not to mention installation, transport, and collateral costs. However, a
defective item in the house of the producer would produce a lesser cost, just for
reworking or dismissing the individual component. Let L m be the loss for the
producer. Then we can find a value for the product characteristic Y 0 ±Δ m for
which the loss function equals that loss. Thus, as it is clearly shown in Fig. 8.1,
the manufacturing specification limit is lower than the customer specification
limit. The distance between both of them depends on the difference between the
customer’s loss and the producer’s loss, as the manufacturing tolerance can be
computed as:

Example 8.1.
Metal plates thickness.
We use the example described in Chapter 5 The quality characteristic under
study is the thickness of a certain steel plate. Nominal thickness of this product
is Y 0 = 0. 75 in, with a standard deviation of 0.05 in. The production is organized
in two shifts, seven days a week, and a sample of n = 6 units is drawn from each
shift. The data frame ss.data.thickness2 in the SixSigma package contains the
thickness measurements for two given weeks.
The structure of the data frame and a sample of its first rows are:
 A visualization of all the data is in Fig. 8.2 by means of a dot plot with the
lattice package [15] using the following code:

Fig. 8.2 Thickness example: one week data dot plot. Each point represents the thickness of one metal
plate

The layout argument is a two-element integer vector for the number of

columns and rows in which the matrix of plots is organized, in this case we have
one panel for each day and in this way we sequentially display the whole week
in one row.
Let us assume that a tolerance Δ c  = ±0. 05 in is allowed by design in order to
consider the product acceptable by 50 % of the customers4 and that the customer
loss at that point is L c  = 2. 5 USD. Then, the expression of the loss function is:

If the cost for the producer at the specification limit L m  = 1 USD, then the
manufacturing tolerance is:

and the manufacturing specification limits are 0. 75 ± 0. 032. □ 

In this section, we have focused on a nominal-is-best characteristic, but there are

models for larger-the-better and smaller-the-better characteristics loss functions,
see, for example, [1] or [16].

8.3 Capability Analysis

In this section, the VoC will be compared with the VoP. Firstly, we need to
quantify the VoP.

8.3.1 The Voice of the Process

In the previous section, we saw that the VoC is determined by the specification
limits. Now, the reference limits take the stall in name of the VoP. Reference
limits are usually named natural limits. They are also used as control limits in
control charts, so sometimes the three terms are used as synonyms. Reference
limits are defined in ISO 22514-4 [8] as the 0.135 % and the 99.865 % quantiles
of the distribution that describe the process characteristic. This means that the
interval within the reference limits includes 99.73 % of the observations. In a
Normal distribution, this is equivalent to a distance of three standard deviations
from the mean of the process, see Fig. 8.3. The probability of being out of the
natural limits (0.0027) is used as the α value in hypothesis tests, see Chapter 5

Fig. 8.3 Reference limits in a Normal distribution. Within the reference limits fall 99.7 % of the data

At this point, we need data in order to listen to the VoP. On the one hand, we
need to estimate the reference limits, so we need a sample of an in-control
process and make inference about the probability distribution. If we can accept
that data come from a normally distributed process, then the reference limits are
just μ ± 3σ. We estimate μ and σ, and we are done, see Chapter 5 for inference
and estimation. On the other hand, samples are to be taken in order to
compare items’ actual measurements with specification limits. Thus, the easiest
way of assessing our quality is to count the number of items in the sample that
are correct. The proportion of correct units is the yield of the sample. The yield
of the process may be calculated taking into account rework (first time yield)
and several linked processes (rolled throughput yield). The proportion of defects
is the complementary of process yield. Defects per unit and defects per million
opportunities (DPMO) are other usual metrics. Either way, if the sample is
representative of the population, then we can estimate the yield of the process
through the sample proportion of correct items.

Example 8.2.
Metal plates thickness (cont.) Process yield.
For the sake of clarity, we use the customer specification limits, i.e., 0. 75 ±
0. 05. Thus, we can count the items out of specification in the sample as follows:

Therefore, the yield of the sample is 78/84 = 92.86% and the proportion of
defects in the sample is 7.14%. □ 

The number of defects in a sample is useful for accounting purposes or for

acceptance sampling procedures (see Chapter 7). However, in Statistical Quality
Control we are interested in the long-term performance of a process. To find out
about that, we identify the probability distribution of the data and estimate its
parameters. Then, the proportion of defects is estimated as the probability that
the random variable defining the process exceeds the specification limits.

Example 8.3.
Metal plates thickness (cont.) Proportion of defects.
The first question would be: is the random variable that characterizes the
quality characteristic of our process normally distributed? The first tool we can
use is a histogram. Fig. 8.4 shows the histogram for all the samples of the week
in our example. Even though it seems normal, we can perform a hypothesis test
and see if we should reject normality:

Fig. 8.4 Histogram of metal plates thickness. The histogram provides clues about normality.
A normality test proves to be helpful when in doubt

We use the Anderson-Darling normality test, which is recommended in ISO

22514-4 . As the p-value is large (even larger than 0.1), we cannot reject the
normality hypothesis (see Chapter 5). Then let us estimate the parameters of our
population:

And now we can estimate the likely proportion of defects of our process:
 More than 8.6 % of the items will be, in the long term, out of specifications. 
□ 

8.3.2 Process Performance Indices

For the sake of simplicity in the exposition, in what follows we assume normally
distributed characteristics. For non-normal characteristics the estimation
methods are slightly different, but the logic is the same, see [8] . The process
performance can be evaluated at any point by comparing the reference limits
with the specification limits. Thus, the process performance index is defined as:

where σ LT represents long-term (LT) variation . Note that this quotient is the
number of times the reference limits (natural variation) fits into the specification
limits. The lower the index, the greater the proportion of non-conforming items.
However, if the process is not centered in the target, we have a different
performance for larger values and for lower values. To overcome this situation,
the upper and lower process performance indices are defined as:

and the minimum process performance index reflects better the performance
of the process:

An important interpretation of those indices is that if P p ≠ P p k , then the

process is not centered at the specification tolerance, and the process might need
to be adjusted.
 As we have a sample of our process, we need to estimate the performance
indices. To do so, we need to estimate the mean and standard deviation of the
population, i.e., μ and σ LT respectively. The sample mean and the sample
(overall) standard deviation are usually used to estimate performance indices:

where:

The above reasoning applies to the capability indices in the next section.
Fig. 8.5 shows the interpretation of the index depending on its value. If an index
is equal to 1, it means that the reference limits and the specification limits are of
equal width, and therefore we will get approximately (α × 100) % defects in the
long term. If the index is greater than one, then the process is “capable” of
fulfilling the specifications, whilst if the index is lower than 1, then we have a
poor quality and the proportion of defects is greater than α.

Fig. 8.5 Specification limits vs. reference limits. The larger the distance from the reference limit to the
specification limits, the better the performance. The plots correrspond, left to right, to performance (or
capability) indices equal, greater, and lower than 1

Example 8.4.
Metal plates thickness. Process performance.
The sample mean standard deviation were computed in the previous
example. Moreover, the producer’s specification limits are 0. 75 ± 0. 032, i.e.,
USL = 0.782 and LSL = 0.718. Then we can easily calculate the performance
indices as follows:
 We will see in the following section how to compute performance indices
using the qcc R package.
 □ 

8.3.3 Capability Indices

Performance indices in the previous section can be used even when the process
is out of control. Moreover, performance indices measure long-term (LT)
variability. Once the process is in control, capability indices can be obtained. In
order to compute capability indices, we need data in m rational subgroups of size
n. The aim of a capability index is the same as in performance indices, i.e., to
measure the ratio between the specification tolerance and the reference interval:

The difference is in the data used and in the fact that short term (ST)
variability σ S T is used for an in-control process instead of the LT variability.
Again, the lower the index, the greater the proportion of non-conforming
items. We also define upper and lower process capability indices as:

and the minimum process capability index reflects better the performance of
the process:

If the process is not centered in the specification tolerance, then we have

different values for C p and C p k . Likewise with performance indices, we need to
estimate the capability indices. The mean of the process μ is estimated also
through the sample mean . However, we use the following estimator for the ST
standard deviation σ ST:

where s j is the standard deviation of each subgroup. If the process is

monitored using a Range control chart or a Standard Deviation control chart, the
following estimators can be used respectively:

where and are the average range and standard deviation of the
subgroups respectively, and d 2 and c 4 are tabulated constants that only depend
on the sample size n. See Chapter 9 to find out more about control charts. Then,
the appropriate point estimators for the capability indices are:

Finally, let us consider a process in which the target T is not centered in the
specification interval, i.e., . In this situation, we can use the so-
called Taguchi index defined as:

which can be calculated also from the C p k as follows:

Example 8.5.
Metal plates thickness. Capability indices.
In order to perform a capability analysis using the qcc package we need to
create a qcc object as if we wanted to plot a control chart (see Chapter 9) with
subgroups. Assuming that each shift is a subgroup:
 Now we can get the capability indices and a graphical representation of the
process using the process.capability function, see Fig. 8.6:

Fig. 8.6 Capability analysis for the thickness example. A histogram is shown and compared with the
specification limits and target, along with the computed indices
 Notice that, in addition to point estimators, a confidence interval is provided
which is very useful for the monitoring of the capability. It is apparent that this
illustrative process is not capable at all. □ 

A final remark on Performance and Capability indices. Performance indices

measure the actual capability of the process, using the variability in the long
term, i.e., the overall variability. Capability indices measure the potential level
of performance that could be attained if all special causes of variation were
eliminated. Indeed, capability will be normally lower than performance indices
as they measure variability in the short term, i.e., the within groups variability.
This can be clearly seen in Fig. 8.2 where each group individually has
approximately the same variation, but if we take all the measurements there is a
larger variation. On the other hand, we can find in the literature tables of
recommended values for the indices. From our view, all processes are different,
and the important thing is that the expert in the subject matter understands
capability indices and monitors processes consciously. Just to have some
numbers in mind, values for C p k equal to 1.33 and 1.67 could be for good and
outstanding processes respectively. A value of C p k  = 2 corresponds with a
Sigma score of 6, i.e., a Six Sigma quality process, see [1].
We have provided formulae for estimating capability indices through point
estimators. These point estimators have a sampling distribution (see Chapter 5),
and confidence intervals can also be obtained, as in the above example. More
details about capability indices estimation can be found in [14] and [13].
The R packages SixSigma and qualityTools also compute capability indices.
For the sake of space we do not include examples of them, check their
documentation for details.

8.4 ISO Standards for Capability Analysis and R

As it was shown in Sect. 8.2, ISO technical committee TC69/SC8 (Application
of statistical and related methodology for new technology and product
development) is developing standards about new products design. Several
standards are under development at the time this is written, and one standard is
already published:
ISO 16336:2014 Applications of statistical and related methods to new
technology and product development process – Robust parameter
design (RPD) [11] . This standard gives guidelines for applying the
optimization method of RPD, an effective methodology for optimization
based on Taguchi Methods, to achieve robust products.
ISO technical committee TC69/SC4 (Applications of statistical methods in
process management) is, in turn, responsible for Capability analysis standards.
ISO 22514 Series (Statistical methods in process management – Capability and
performance) is a comprehensive set of standards covering the topics addressed
in this chapter, among others. The Series is composed of the following eight
parts:
Part 1: General principles and concepts [9] . This standard describes the
fundamental principles of capability and performance of manufacturing
processes. It has been prepared to provide guidance about circumstances
where a capability study is demanded or necessary to determine if the
output from a manufacturing process or the production equipment (a
production machine) is acceptable according to appropriate criteria. Such
circumstances are common in quality control when the purpose for the
study is part of some kind of production acceptance. These studies can also
be used when diagnosis is required concerning a production output or as
part of a problem solving effort. The methods are very versatile and have
been applied for many situations;
Part 2: Process capability and performance of time-dependent process
models [6] . This standard describes a procedure for the determination of
statistics for estimating the quality capability or performance of product and
process characteristics. The process results of these quality characteristics
are categorized into eight possible distribution types. Calculation formulae
for the statistical measures are placed with every distribution;
Part 3: Machine performance studies for measured data on discrete
parts[4] . This standard prescribes the steps to be taken in conducting
short-term performance studies that are typically performed on machines
where parts produced consecutively under repeatability conditions are
considered. The number of observations to be analyzed will vary according
to the patterns the data produce, or if the runs (the rate at which items are
produced) on the machine are low in quantity;
Part 4: Process capability estimates and performance measures [8] .
This standard is being technically revised at the time this is written. This
standard describes process capability and performance measures that are
commonly used;
Part 5: Process capability estimates and performance for attributive
characteristics. This standard is under development at the time this is
written;
Part 6: Process capability statistics for characteristics following a
multivariate normal distribution [7] . This standard provides methods for
calculating performance and capability statistics for process or product
quantities where it is necessary or beneficial to consider a family of singular
quantities in relation to each other;
Part 7: Capability of measurement processes [5] . This standard defines a
procedure to validate measuring systems and a measurement process in
order to state whether a given measurement process can satisfy the
requirements for a specific measurement task with a recommendation of
acceptance criteria. The acceptance criteria are defined as a capability
figure or a capability ratio;
Part 8: Machine performance of a multi-state production process [10] .
The aim of this standard is to define the evaluation method to quantify the
short-term capability of a production process (capacity of the production
 tool, widely termed capability), i.e. the machine performance index, to
ensure compliance to a toleranced measurable product characteristic, when
said process does not feature any kind of sorting system.
Finally, parts 1 and 2 of ISO 3534 (Vocabulary and symbols about Statistics,
Probability, and Applied Statistics) [2, 3] are also useful for the scope of
Capability Analysis.

References
1. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. Statistical Engineering for Process
Improvement, Use R!, vol. 36. Springer, New York (2012). url http://www.springer.com/statistics/book/
978-1-4614-3651-5

2. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard (2010). url http://
www.iso.org/iso/catalogue_detail.htm?csnumber=40145

3. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=40147

4. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-3:2008 -

Statistical methods in process management – Capability and performance – Part 3: Machine
performance studies for measured data on discrete parts. Published standard (2011). url http://www.iso.
org/iso/catalogue_detail.htm?csnumber=46531

5. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-7:2012 -

Statistical methods in process management – Capability and performance – Part 7: Capability of
measurement processes. Published standard (2012). url http://www.iso.org/iso/catalogue_detail.htm?
csnumber=54077

6. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-2:2013 -

Statistical methods in process management – Capability and performance – Part 2: Process capability
and performance of time-dependent process models. Published standard (2013). url http://www.iso.org/
iso/catalogue_detail.htm?csnumber=46530

7. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-6:2013 -

Statistical methods in process management – Capability and performance – Part 6: Process capability
statistics for characteristics following a multivariate normal distribution. Published standard (2013).
url http://www.iso.org/iso/catalogue_detail.htm?csnumber=52962

8. ISO TC69/SC4–Applications of statistical methods in process management: ISO/TR 22514-4:2007 -

Statistical methods in process management – Capability and performance – Part 4: Process capability
estimates and performance measures. Published standard (2013). url http://www.iso.org/iso/catalogue_
detail.htm?csnumber=46532

9. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-1:2014 -

Statistical methods in process management – Capability and performance – Part 1: General principles
 and concepts. Published standard (2014). url http://www.iso.org/iso/catalogue_detail.htm?csnumber=
64135

10. ISO TC69/SC4–Applications of statistical methods in process management: ISO 22514-8:2014 -

Statistical methods in process management – Capability and performance – Part 8: Machine
performance of a multi-state production process. Published standard (2014). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=61630

11. ISO TC69/SC8–Application of statistical and related methodology for new technology and product
development: ISO 16336:2014 - Applications of statistical and related methods to new technology and
product development process – Robust parameter design (RPD). Published standard (2014). url http://
www.iso.org/iso/catalogue_detail.htm?csnumber=56183

12. Knight, E., Russell, M., Sawalka, D., Yendell, S.: Taguchi quality loss function and specification
tolerance design. Wiki (2007). url https://controls.engin.umich.edu/wiki/index.php/Taguchi_quality_
loss_function_and_specification_tolerance_design. In Michigan chemical process dynamics and
controls open textbook. Accessed 23 June 2015

13. Montgomery, D.: Statistical Quality Control, 7th edn. Wiley Global Education, New York (2012)

14. Pearn, W., Kotz, S.: In: Encyclopedia and Handbook of Process Capability Indices: A Comprehensive
Exposition of Quality Control Measures. Series on Quality, Reliability & Engineering Statistics. World
Scientific, Singapore (2006)

15. Sarkar, D.: Lattice: Multivariate Data Visualization with R. Springer, New York (2008). url http://
lmdvr.r-forge.r-project.org. ISBN 978-0-387-75968-5

16. Taguchi, G., Chowdhury, S., Wu, Y.: Taguchi’s Quality Engineering Handbook. Wiley, Hoboken (2005)

Footnotes
1 A current trend is to use voice of stakeholders (VoS) instead of VoC.

2 see Chapter 4 for details on ISO standards development stages.

3 Production is applicable to products and services for the scope of this chapter.

4 This way of fixing specifications is called functional tolerance in Taguchi’s method terminology.

Part IV
Control Charts
 This Part contains two chapters dealing with the monitoring of processes. In
Chapter 9, the most important tool in statistical process control is explained:
control charts. Several types of control charts are shown in order to detect if a
process is out of control. By controlling the stability of the process, we may
anticipate future problems before products/services are received by the customer.
It is also a powerful improvement tool, as the investigation of special causes of
variation may result on better procedures to avoid the root cause of the out-of-
control situation. Chapter 10 presents a methodology to monitor processes where
a nonlinear function characterizes the quality characteristic. Thus, confidence
bands are computed for the so-called nonlinear profiles, allowing the monitoring
of processes under a similar methodology to the control charts approach.
© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_9

9. Control Charts with R

Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
Control charts constitute a basic tool in statistical process control. This chapter
develops the fundamentals of the most commonly applied control charts.
Although the general basic ideas of control charts are common, two main
different classes are to be considered: control charts for variables, where
continuous characteristics are monitored; and control charts for attributes, where
discrete variables are monitored. In addition, as a special type of control charts,
time weighed charts are also outlined in the chapter. Finally, to guide users in the
practice of control charts, a selection of the available ISO standards is provided.

9.1 Introduction
In Chapter 1 we introduced quality control with an intuitive example based on
the use of a control chart. In fact, control charts are one of the most important
tools in Statistical Process Control (SPC). The underlying idea of control charts
is to build some natural limits for a given summary statistic of a quality
characteristic. Under the presence of common (natural) causes of variation, this
summary statistic is expected to remain within these limits. However, if the
statistic falls out of the natural limits, it is very unlikely that only natural
variability is present, and an investigation should be carried out in order to look
for possible assignable causes of variation, which should be eliminated [18]. In
practice, the natural limits will be estimated according to the sampling
distribution of the statistic to be monitored, and we will refer to the estimated
limits as “control limits.” In fact, every point in a control chart leads to a
hypothesis test: a point out of the control limits may imply an abnormal
performance of the process under study and, as a consequence, the process may
be considered to be out of control. On the contrary, if all points remain within the
control limits, the process may be considered to be in control. See Chapter 5 for
details about statistics, sampling distributions, and hypothesis tests.
This chapter develops the fundamentals of the most commonly applied
control charts, the basic tool used in SPC. The remaining of this section depicts
the basic ideas of control charts; Sect. 9.2 describes the control charts for
variables as well as the special (time weighed) charts; Sect 9.3, describes the
control charts for attributes. Finally, Sect. 9.5 provides a selection of the ISO
standards available to help users in the practice of control chart.

9.1.1 The Elements of a Control Chart

A control chart is a two-dimensional chart whose y-axis represents the variable
we are monitoring. In general, a summary statistic for the variable is computed
for each sample j = 1, … m of the process (see Chapter 5), and plotted
sequentially in the order in which they have arisen. The x-axis of the chart is an
identification of such j sample. Sometimes this sample is an individual value x j ,
and sometimes the sample is a group of values x i j , i = 1, …, n j . If all the groups
have the same size, then n j  = n. The values are plotted as points and linked with
straight lines to identify patterns that show significant changes in the process
performance. Along with the sequence of points, three important lines are
plotted:

1. Center line (CL): This is the central value the statistic should vary around.
For example, the mean of the process;

2. Lower control limit (LCL). This is the value below which it is very unlikely
for the statistic to occur when the process is in control;

3.
Upper control limit (UCL). This is the counterpart of the LCL on the upper side
of the CL. The LCL and UCL are symmetric if the probability distribution of the
statistic to be monitored is symmetric (e.g., normal).

The control limits are completely different from the specification limits, that
is, the limits beyond which the process will not be accepted by the customer
(see Chap. 8). The control limits are computed as a confidence interval (see
Chap. 5) that comprises a high proportion of the values. Typical control limits
are those between the mean and three standard deviations (μ ± 3σ). For a normal
probability distribution these limits include 99.73 % of the data. Thus, if nothing
abnormal is taking place in the process, there will only be a probability of 0.0027
for an individual observation to be outside the control limits. Moreover, a control
chart adds information about the variation of the process. Figure 9.1 shows how
both types of information are related.

Fig. 9.1 Control charts vs. probability distribution. The control chart shows the sequence of the
observations. The variation around the central line provides an idea of the probability distribution

9.1.2 Control Chart Design

The main aspect to consider in a control chart is how to set the control limits.
In general, it is a two-phase process. In Phase I, reliable control limits are
estimated using a preliminary set of samples. In order to find appropriate limits,
the process should be in control during Phase I. Thus, if special causes of
variation are identified, those points should be removed from the data set.
Moreover, as a general rule the limits should be computed with at least 25–30
samples. From that point on, in Phase II, the subsequent samples are plotted in a
chart with the former control limits. When the individual observations of the
statistic that is being monitored are within the control limits, the process is
considered to be statistically in control. It should be noted that sometimes Phase
I limits are assumed to be fixed, for example because the nominal values are
accepted for the current process. In this case, that should be confirmed with the
appropriate hypothesis tests. In any case, if the process changes, e.g., a reduction
of variability is attained, then the control limits should be revised. Otherwise the
control chart losses effectiveness. For the sake of simplicity, in what follows we
will plot the charts in a single phase, although in practice it should be done in
two phases.
The sampling strategy is the other main concern when designing a control
chart. The first feature of this strategy is whether we can sample rational
subgroups or we can only sample individual items. When possible, it is always
preferable to have groups to monitor the mean instead of individual values, as
the methods are more robust against deviations of the data from the normal
distribution. Of course the sampling must be random, and the order of the
samples must be known. The crucial feature of a sample is that it must be
representative of the process under normal conditions. Next, the size of the
group samples is to be determined. In Chapter 6 we presented a method to get
Operating Characteristic (OC) curves as a tool to determine the sample size for
an chart. We will see more examples in this chapter. Finally, we could be
interested in deciding the frequency of sampling. To answer this question, we
look at the Average Run Length (ARL). The ARL is the number of samples, on
average, that will be drawn before detecting a change in the process. This
number follows a Geometric probability distribution, whose unique parameter p
is the probability of a point falling out the control limits., i.e., 0.0027 as we
showed in Sect. 9.1.1. The mean value of a Geometric distribution is
and, therefore, in an in-control process, we will get, on average, an out-of-
control false alarm every 370 samples as:

Having said that, we are interested in learning how many samples will be
needed to detect a given change in the process. In this case, the probability of
detection is the power of the control chart, i.e., 1 −β, where β is obtained from
the OC curve mentioned above, see Chapter 6. Thus, to detect a mean shift
corresponding to a given β, the ARL is:

The ARL indicates the number of samples we need to detect the change.
Then, we should check the sampling frequency depending on our preference
regarding the time we are willing to wait before detecting a change. We leave the
illustrative example for Sect. 9.2 in order to first introduce other concepts.

9.1.3 Reading a Control Chart

Process natural variation is due to common causes, whereas variation outside the
control limits is due to special causes. Common causes arise from randomness
and all we can do is try to reduce it in order to improve the process, for example
via Design of Experiments (DoE), see [3]. Special causes prompt variability that
is not a consequence of randomness. Thus, when a point is outside the control
limits, the (special) cause must be identified, analyzed, and eradicated (Fig. 9.2).

Fig. 9.2 Identifying special causes through individual points. When an individual point is out of the control
limits, a research on the cause should be started, in order to eliminate the root of the problem

Special causes can also generate other persistent problems in a process. They
can be identified through patterns in the chart. Three important patterns that we
can detect are trends, shifts, and seasonality (Fig. 9.3).
Fig. 9.3 Patterns in control charts. We can identify several patterns in a control chart, namely: Recurring
cycles (seasonality) (a), Shifts (b), or Trends (c)
 The following signals may help identify the above out-of-control situations.
Basically, the occurrence of any of the following circumstances is highly
unlikely, thus leading us to the conclusion that an assignable cause may be
present in the process.
Points out of control limits;
Seven consecutive points at the same side of the center line;
Six consecutive points either increasing or decreasing;
Fourteen consecutive points alternating up and down;
Any other unusual pattern.
In addition to control limits we may look inside them in order to anticipate
possible problems. Three zones can be defined comprised by the control limits
(Fig. 9.4).

1. Zone C: ranges between the central line and one standard deviation;

2. Zone B: ranges between one and two standard deviations from the central
line;

3. Zone A: ranges between two and three standard deviations from the central
line.

Fig. 9.4 Control chart zones. The distribution of the observations in the three zones can convey out-of-
control situations

After the definition of these three zones, some other unusual patterns may
arise, namely:
Two out of three consecutive points in Zone A or above;
Four out of five consecutive points in Zone B or above;
Fifteen consecutive points in Zone B.

9.2 Control Charts for Variables

9.2.1 Introduction
When the characteristic to be measured is a continuous numerical variable we
must use control charts for variables. In this type of process control we have to
control both the central values of the variable and its variability, this is the reason
why control charts for variables are usually used two by two: one chart for the
central values (or the individual values if there are no groups) and another one
for the variability.
For the calculation of the control limits we have to take into account that we
work with samples from our processes. Then we will have to estimate the
standard deviation of the process with the aid of these samples. The theoretical
details involved in this estimation are out of the scope of this book (see [15]); we
will simply point out that some Shewhart charts use the distribution of the range
from normal samples. The coefficients required to determine the standard
deviation may be found in Appendix A, and also in [15] and in [11].

9.2.2 Estimation of σ for Control Charts

The sample standard deviation is not an unbiased estimator of the population
standard deviation. Therefore, in SPC, the two following unbiased estimators are
usually employed: Coefficients c 4 and d 2 may be found in Appendix A,
and also in [15] and in ISO 7870-2 [11]. They can also be computed in R using
the ss.cc.getc4 and ss.cc.getd2 functions in the SixSigma package.

9.2.3 Control Charts for Grouped Data

In order to construct this kind of charts we have to define rational subgroups,
ideally of the same size. If the groups have different sizes, then the control limits
will not be constant, and will have an uneven aspect what makes them more
difficult to interpret. This is the reason why it is advisable that the subgroup sizes
be as similar as possible. By the way, in the range chart the central line would
also be variable.
As already mentioned, control charts should be plotted by pairs. The most
famous pair of control charts is the one made up by the X-bar chart and the
Range chart. The reason to use the range as a measure of variability is that we
normally work with small samples and the range is easy to compute and a good
measure of variability in that case. To estimate the standard deviation, and the
control limits thereafter, Shewhart’s formulae are used. The calculations are
easily performed by R, as will be shown soon.

Mean Chart (X-Bar Chart)

The statistic to be monitored is the mean of each sample :

The center line (CL) is the so-called grand average , i.e.:

To compute the control limits (CL ± 3σ) we need to estimate σ. In this case,
as we are monitoring means, we need the standard deviation of those means in
order to compute the limits. From the sample distribution of the sample mean
(see Chapter 5), we know that its standard deviation is . From the central limit
theorem, we can also use this result, even for non-normal distributions. Thus, the
formulae for the central line and the control limits of the mean chart are as
follows:

where is the average range, i.e.:

and d 2 is a constant that only depends on n. Note that, if groups have

different sample size n j , then the limits are to be calculated for each sample j.
For simplicity when computing control limits, the factor is also tabulated
as A 2, see Appendix A.

Example 9.1.
Metal plates thickness. X-bar chart.
In this chapter, we will use the metal plates thickness example in Chapter 8.
We recall here that we have measurements of metal plates thickness made up of
m = 14 samples of size n = 6, corresponding to 7 days, one sample for each of the
two shifts in which the production is organized. The data frame
ss.data.thickness2 is in the SixSigma package. The points of the X-bar control
are the following ones:

We could make the computations for the control limits using the formulae
above and then plot the control chart using R graphical capabilities. This might
be needed at some point, but in general it is more convenient to use contributed
packages that do all the work. Even though there are several packages that can
plot control charts, we will focus just on the qcc package [17], which is widely
used in both academy and industry. Before plotting the X-bar chart, we show the
main features of the functions in the package and the workflow to use it.
The use of the qcc package is simple. The main function is also named qcc,
and it returns a special object of class qcc. Even though the three entities have
the same name, they are not the same, check Chapter 2 to find out more about
packages, functions, and objects. The qcc function only needs two or three
arguments to create the object: one for the data, one for the type of chart, and
another for the sample sizes. The latter is only needed for certain types of charts,
as we will see later. The data argument can be one of the following: (1) a vector
of individual values; or (2) a matrix or a data frame containing one sample in
each row. In our case, we do not have the data structured in this way, but all the
measurements are in a column of the data frame, and the groups are identified in
another column of the data frame. We could transform the data using standard R
functions, but the qcc package includes a convenient function that does the job:
qcc.groups. This function requires as arguments two vectors of the same length:
data containing all the observations, and sample containing the sample
identifiers for each data value. Thus, we first transform our original data into a
matrix in the appropriate format for the qcc package:

Next, we can create the qcc object for an X-bar chart:

Finally, we can obtain the numerical or graphical results by using the generic
functions summary and plot, see Fig. 9.5:
Fig. 9.5 X-bar chart example (basic options). To get a control chart, just the data and the type of chart are
needed

The summary function returns: the call to the function; the title of the chart; a
five-number summary (Q 1, Q 3, median, maximum and minimum) plus the
mean of the statistic monitored; the sample size n; the number of groups m; the
center value of the statistic and its standard deviation; and the control limits. All
this information can be accessed in the object of class qcc, as it is actually a list
with the following elements:

This information can be used for further purposes. Notice that some
information is not shown by the summary function, namely: type, nsigmas, and
violations.

The violations item is, in turn, a list of two elements: (1) the indices of the
samples out of the control limits; and (2) the indices of the samples that violate
some of the rules listed above, e.g., too many points at the same side of the
center line. In our example, only an out-of-control-signal is shown: sample
number three is out of the control limits. Special causes of variation should be
investigated.
We can add options to the qcc function to customize our control chart. The
following is a brief description of the available options:
center: Phase I fixed center value;
std.dev: a fixed value for the standard deviation (or a method to estimate
it);
limits: Phase I fixed limits;
data.name: a character string, just for the plots;
labels: labels for the samples;
 newdata: if provided, the data in the argument data is used as Phase I data,
i.e., to compute limits. A vertical line is plotted to separate Phase I and
Phase II data;
newsizes; sample sizes for the Phase II new data;
newlabels; labels for the new samples;
nsigmas: The number of standard deviations to compute the control limits
(by default 3);
confidence.level: if provided, control limits are computed as quantiles. For
example, a confidence level of 0.9973 is equivalent to 3 sigmas;
rules: rules for out-of-control signals. Experienced R users can add new
rules adapted to their processes;
plot: whether to plot the control chart or not.

On the other hand, the qcc.options function allows to set global options for
the current session, check the function documentation (type ?qcc.options) to
find out more. Also the call to the plot function over a qcc object allows to
customize parts of the chart. The following code sets options and adds arguments
to our object of class qcc, see Fig. 9.6.

Fig. 9.6 X-bar chart example (basic options). Options can be added to the qcc function and globally to the
qcc.options
 Let us finish this example getting the OC Curves and ARL explained in
Sect. 9.1.2. Now that we have an object of class qcc, we can use the oc.curves
function to get values of β for some values of n, including n = 6 in the example,
and plot the OC curve, see Figure 9.7:

Fig. 9.7 OC curve for the X-bar control chart. The values of β can be stored and, afterwards, we can
calculate the ARL

The oc.curves function, in addition to plotting the OC curve, returns a

matrix with the values of β for each sample size n (columns) and process shift
(rows). Thus, the last expression in the above code calculates the ARL for a one-
standard-deviation process shift. From the calculated ARL = 3.43, we conclude
that, in average, more than three samples are needed to detect an out-of-control
situation. In our example, if we need to detect such a change within 1 day, then
the solution would be drawing two samples per shift, or increasing the sample
size of the groups and see the new ARL.
 □ 

In this section, we have explained in detail the use of the qcc package. The main
ideas are the same for all types of control charts. Therefore, in the following we
provide less details, explaining just the features that are differential among
charts. Likewise, for the sake of space, OC curves and ARL for each specific
control chart are not explained, see [15] for details.

Range Chart (R Chart)

This chart monitors the stability of the process variability by means of the
sample ranges. Thus, the statistic is the range, and the control limits are:

(9.1)

(9.2)

(9.3)
In this case, we estimate σ R as , see [15] for details. Usually the formulae are
simplified as:

where:

whose values are tabulated for certain values of n, see Appendix A.

Example 9.2.
Metal plates thickness (cont.). Range chart.
As we already have the matrix with the samples in the object samples.thick,
we just create the range chart in Fig. 9.8 just changing the type of control chart
as follows:
Fig. 9.8 Range chart for metal plates thickness. The range charts monitor variability

Apparently, even though we detected an out-of-control sample for the mean,

the variation of the process is in control. □ 

Standard Deviation Chart (S Chart)

Range charts are easy to compute and interpret. However, as sample size
increases, it is more appropriate to monitor variability using the standard
deviation. In general for n ≥ 8 (10 according to some authors), we should use the
standard deviation chart (S Chart). Obviously, the statistic monitored for each
sample is the sample standard deviation s j :

We estimate the standard deviation of the s statistic as:

and therefore the central line and the control limits are the following:
 New constants B 3 and B 4 are then defined to simplify the formulae as
follows:

where:

Example 9.3.
Metal plates thickness (cont.) Standard deviation chart.
Figure 9.9 shows the S chart created with the following expression, again
using the matrix of samples:

Fig. 9.9 S chart for metal plates thickness. The S chart monitors variability through the standard deviation
of the samples

As expected, there are not out-of-control samples for standard deviations.

  □ 

At the beginning of this section we pointed out that control charts for variables
should be shown in couples. Why is this so important? We have seen in the
examples above that we might have out-of-control samples in terms of the mean
values, while being the variability in control; and the opposite case may also
occur. This is the reason why if we only monitor mean values, we will not be
aware of such situations.

Example 9.4.
Metal plates thickness (cont.) X-bar & S chart.
To illustrate the importance of monitoring variability, let us simulate a new
shift sample with the following code:

In order to jointly plot the two control charts, we use the graphical parameter
mfrow to divide the graphics device in two rows. We add the new sample as
Phase II data:

As it is clearly shown in Fig. 9.10, the last point is in control from the point
of view of the mean, but it is out of control from the point of view of variability.
In fact, one of the six new values is even out of the specification limits. □ 
Fig. 9.10 X-bar and S chart for metal plates thickness. It is important to monitor both mean values and
variability

9.2.4 Control Charts for Non-grouped Data

Individual Values Chart and Moving Range Chart
The simplest control chart we may create is the individual values chart (I Chart).
When it is not possible to create rational subgroups we may monitor data
individually. In this case, we estimate the global standard deviation as MR⁄d 2
taking d 2 for n = 2.
The control lines for the I Chart are:

where is the moving range, computed as:

 The most adequate chart to accompany the individual values chart in order to
control process variability is the moving range chart. Actually, what we do in
such a case is to assume that every two successive data points constitute a group,
and in this way we can determine a range equivalence to the difference between
two consecutive observations. This is how Shewhart’s principles for sample
sizes n = 2 can be applied to individual values. Therefore, the center line and
control limits are the same as in Eq. (9.1) for the R chart, taking the constants d 3
and d 2 for n = 2. In practice, the lower limit is always zero since the formula
returns a negative value and a range cannot be negative by definition.

Example 9.5.
Metal plates thickness (cont.) I & MR control charts.
For illustrative purposes, let us use the first 24 values in the ss.data.
thickness2 data frame to plot the individuals control chart. In this case, a vector
with the data is required, hence we do not need any transformation. On the other
hand, to plot the moving range control chart, we can create a matrix with two
artificial samples: one with the first original 23 values, and another one with the
last original 23 values. The following code plots the I & MR control charts in
Fig. 9.11.
Fig. 9.11 Individual and moving range charts for metal plates thickness. The moving range assumes
sample size n = 2 to compute limits

In this case, an out-of-control signal is produced for both charts in

measurement 12. In addition, a violating run is also produced in sample 19, as
there are too many points at the same side of the center line.
 □ 

9.2.5 Special Control Charts

The Shewhart control charts for variables explained so far are very powerful to
detect significant out-of-control situations under the assumption of independent
samples. In this section we outline two more types of charts that are useful for
time-dependent processes. Their computations and interpretation are less
intuitive than in Shewhart charts. We provide the formulae and illustrate with
examples, without going into the details, that may be consulted, for example, in
[15].

CUSUM Chart
The CUSUM chart (cumulative sums) controls the process by means of the
difference of accumulated sums with respect to a target value (usually the mean).
This chart may be used with either grouped or individual values. For each
sample two statistics are monitored, the so-called cusum coefficients: one for the
positive deviations C j + and another one for the negative deviations C j −, which
are calculated as follows:

where K represents the amount of shift we want to detect in terms of standard

errors. A typical value for K is 1. Note that for samples of n ≥ 2 observations,
should be used instead of x j . For an in-control process, coefficients will
randomly vary around 0. Thus, the center line is at 0. The control limits of the
CUSUM control chart are fixed at ± H, where H is related to the number of
standard errors. A typical value for H is 5. See [15] for a detailed design of the
CUSUM control chart.

Example 9.6.
Metal plates thickness (cont.) CUSUM chart.
The following expression plots the CUSUM chart in Fig. 9.12 using the
default settings, check the documentation of the function to learn more about
how to change them:
Fig. 9.12 CUSUM chart for metal plates thickness. Process shifts are detected sooner than with Shewhart
charts

Note that an out-of-control signal is detected in sample 18. The number of

points at the same side of the central line indicates when the shift took place. □ 

EWMA Chart
EWMA is the acronym for Exponentially Weighted Moving Average. This chart
permits the identification of small deviations. It is said that the EWMA chart has
memory, as each monitored value takes into account the information from
previously monitored values. It is specially appropriate when data significantly
deviates from the normal distribution. The statistic to be monitored is a weighted
moving average z j computed as:

The center line is at and the control limits are:

Therefore, we need two parameters to design the EWMA chart: the

smoothing parameter λ and the number of sigmas L determining the width of the
control limits. Typical values for these parameters are 0.2 and 3, respectively.

Example 9.7.
Metal plates thickness (cont.) EWMA chart.
The following expression plots the EWMA chart in Fig. 9.13 using the
default settings, check the documentation of the function to learn more about
how to change them:

Fig. 9.13 EWMA chart for metal plates thickness. The statistic plotted is in same scale that the variable
 Note that in addition to the points and lines of the statistic z j , the real value
of each sample or observation is plotted as a ‘+’ symbol. □ 

Finally, it is important to remark that CUSUM and EWMA charts should be

initialized when out-of-control signals appear.

9.3 Control Charts for Attributes

9.3.1 Introduction
It is not always possible to measure a quality characteristic numerically.
However it is easy to check certain attributes, for instance:
conforming/nonconforming. It is possible to perform a process control over this
kind of data by means of the so-called attribute control charts. Likewise the
control charts for variables, this kind of control charts detect highly unlikely
situations according to the inherent probability distribution of the process. In this
case, discrete probabilities like the binomial or the Poisson distributions are used
instead of the normal distribution, see Chapter 5 for details on probability
distributions.

9.3.2 Attributes Control Charts for Groups

In attribute control charts, we may have j samples of data with a given size n j ,
in which we count the items D j that fall in a given category, e.g., “defective,”
and calculate the proportion of items in the sample . Then we can
monitor the process with the p and n p control charts. The underlying probability
distribution used in these charts is the binomial B(n, p).

The p Chart
This chart is used to control proportions within groups of a certain size, such as
lots, orders in a day, etc. The statistic to be monitored is the sample proportion p
j , whose standard deviation is:

The center line is the total proportion of defects , i.e.:

and the limits are calculated as follows:

Note that if we have different sample sizes the control limits are not constant.

The n p Chart
The n p control chart is used to monitor the number of elements D j with the
characteristic to be controlled, not the proportion. Nevertheless, the type of data
is the same as in the p chart, i.e., groups with a given size n, mandatory all of the
same size for this chart. The center line is the average number of items with the
characteristic per sample, i.e., . In this case the control limits are calculated
as:

Example 9.8.
Metal plates thickness (cont.) p and n p charts.
Suppose we want to monitor the proportion and the number of items whose
thickness is larger than the midpoint between the nominal value and the
specification limit, i.e., 0.775 in. We first need a vector with the proportions for
each sample, that can be calculated with the following expression:

The p control chart in Fig. 9.14 can now be obtained with the following call
to the qcc function:
Fig. 9.14 p chart for metal plates thickness. The statistic monitored is the proportion of items in each
sample

The n p control chart in Fig. 9.15 shows the same pattern. The election of one
or another is most of the times a matter of what is easier of interpret by the team:
proportions or counts. Moreover, the n p chart only can be used when all the
samples have the same size n.
Fig. 9.15 n p chart for metal plates thickness. The statistic monitored is the number of items within a
category in each sample

 □ 

9.3.3 Control Charts for Events

Unlike in the previous charts, sometimes we count events, e.g., nonconformities,
not within a finite sample, but in a fixed interval of time or space.
We can monitor these type of processes with the c and u control charts. The
underlying probability distribution used in these charts is the Poisson P o(λ).

The c Chart
The c control chart is used to control the total number of events for a given
process in an errors per interval basis, this is a process that follows a Poisson
distribution in which there could theoretically be an infinite number of possible
events. In this kind of processes we do not have a sample size from which a
proportion could be calculated, as in the p and n p charts. The most common
application of this chart is to control the total number of nonconformities
measured in a series of m samples of the same extension, either temporal or
spatial.
Examples could be the number of unattended calls per hour, the number of
nonconformities per day, etc. It could also be used to monitor the number of
events of physical samples, e.g., when samples form a continuous material are
taken (fabric, surfaces (ft 2), liquid (l), etc.) and the average number of events per
sample are measured. The statistic of each sample is the count of events c j . The
center line is the average number of events per sample . As in a Poisson
distribution the variance is the parameter λ, then an estimator of the standard
deviation is , and therefore the control limits are:

The u Chart
When in the previous situation we have n j items of different size within each
sample j in which we count the total number of events x j in all the elements
within the sample, it may be interesting to monitor the average number of
defects per item. In such a situation the u chart should be used. The statistic to be
monitored is u j , the number of defects per unit in the sample j, i.e.:

and the center line is the average number of defects per unit in all the
samples:

The limits are calculated as follows:

Example 9.9.
Metal plates thickness (cont.) c and u charts.
In addition to the thickness measurement, some metal plates (1, 2, or 3) are
inspected to find flaws in the surface. The inspector counts the number of flaws
in each inspected metal plate, and this information is in the column flaws of the
ss.data.thickness2 data frame.
To plot the c chart for all the metal plates in Fig. 9.16 we need the vector
with just the inspected items, i.e., removing the NA values:
Fig. 9.16 c chart for metal plates thickness. The statistic monitored is the count of flaws in each individual
metal plate

Finally, if we want to monitor the average flaws per metal plate in each shift,
then we need the u chart in Fig. 9.17 that is the result of the following code:

Fig. 9.17 u chart for metal plates thickness. Note that, as we have a different number of inspected items in
each shift, the limits are not constant
  □ 

9.4 Control Chart Selection

In this chapter, we have reviewed the main types of control charts used. A
decision tree summarizing how to choose the appropriate control chart is shown
in Fig. 9.18.
Fig. 9.18 Decision tree for basic process control charts. First, check your variable type. If it is a continuous
variable, the chart depends on the groups size. If it is an attribute variable, the chart depends on what you
want to monitor (proportion of defects or number of defects per unit)

We have focused on the use of the qcc R package [17]. Other packages can
be used to plot control charts, e.g., IQCC [2], qcr [4], spc [14], and qicharts[1].
Packages also useful for quality control charting in R are, for example, the
spcadjust package [5], with functions for the calibration of control charts; and
the edcc package [20], specific for economic design of control charts.
Nevertheless, with the formulae provided for the control lines (center, upper
limit, and lower limit) you are prepared to plot your own control charts with the
R graphical functions in the packages graphics, lattice [16] or ggplot2 [19],
just plotting points and lines and adding control lines (CL, UCL, LCL). In this
way, you can customize any feature of your control chart. Furthermore, for the
sake of completeness, Appendix A contains the constants used in the formulae.
9.5 ISO Standards for Control Charts
The following are the more relevant standards in the topic of control charts:
ISO 7870-1:2014 Control charts—Part 1: General guidelines [12] . This
Standard presents the key elements and philosophy of the control chart
approach, and identifies a wide variety of control charts, such as Shewhart
control chart and specialized control charts.
ISO 7870-2:2013 Control charts—Part 2: Shewhart control charts [11]
. This Standard establishes a guide to the use and understanding of the
Shewhart control chart approach to the methods for statistical control of a
process. It is limited to the treatment of SPC methods using only
Shewhart’s charts. Some supplementary material that is consistent with the
Shewhart approach, such as the use of warning limits, analysis of trend
patterns and process capability is briefly introduced.
ISO 7870-3:2012 Control charts—Part 3: Acceptance control charts
[10] . This Standard provides guidance on the uses of acceptance control
charts and establishes general procedures for determining sample sizes,
action limits, and decision criteria. This chart is typically used when the
process variable under study is normally distributed, however, it can be
applied to a non-normal distribution. Examples are included to illustrate a
variety of circumstances in which this technique has advantages and to
provide details of the determination of the sample size, the action limits,
and the decision criteria.
ISO 7870-4:2011 Control charts—Part 4: Cumulative sum charts [8] .
This Standard provides statistical procedures for setting up cumulative sum
(CUSUM) schemes for process and quality control using variables
(measured) and attribute data.
ISO 7870-5:2014 Control charts—Part 5: Specialized control charts
[13] . This Standard establishes a guide to the use of specialized control
charts in situations where commonly used Shewhart control chart approach
to the methods of statistical control of a process may either be not
applicable or less efficient in detecting unnatural patterns of variation of the
process.
ISO 11462-1:2001 Guidelines for implementation of statistical process
control (SPC)—Part 1: Elements of SPC [9] . This Standard provides
guidelines for the implementation of a SPC system, and a variety of
elements to guide an organization in planning, developing, executing,
 and/or evaluating a SPC system.
In addition, parts 1 and 2 of ISO 3534 (Vocabulary and symbols about
Statistics, Probability, and Applied Statistics) [6, 7] are also useful for the scope
of Control Charts.

References
1. Anhoej, J.: qicharts: quality improvement charts. http://CRAN.R-project.org/package=qicharts (2015).
R package version 0.2.0

2. Barbosa, E.P., Barros, F.M.M., de Jesus, E.G., Recchia, D.R.: IQCC: improved quality control charts.
http://CRAN.R-project.org/package=IQCC (2014). R package version 0.6

3. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six Sigma with R. Statistical Engineering for Process
Improvement. Use R!, vol. 36. Springer, New York (2012). http://www.springer.com/statistics/book/
978-1-4614-3651-5

4. Flores, M., Naya, S., Fernandez, R.: qcr: quality control and reliability. http://CRAN.R-project.org/
package=qcr (2014). R package version 0.1-18

5. Gandy, A., Kvaloy, J.T.: Guaranteed conditional performance of control charts via bootstrap methods.
Scand. J. Stat. 40, 647–668 (2013). doi:10.1002/sjos.12006
[MATH][MathSciNet][CrossRef]

6. ISO TC69/SC1–Terminology and Symbols: ISO 3534-1:2006 - Statistics – Vocabulary and symbols –
Part 1: General statistical terms and terms used in probability. Published standard. http://www.iso.org/
iso/catalogue_detail.htm?csnumber=40145 (2010)

7. ISO TC69/SC1–Terminology and Symbols: ISO 3534-2:2006 - Statistics – Vocabulary and symbols –
Part 2: Applied statistics. Published standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=
40147 (2014)

8. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-4:2011 - Control

charts – Part 4: Cumulative sum charts. Published standard. http://www.iso.org/iso/catalogue_detail.
htm?csnumber=40176 (2011)

9. ISO TC69/SC4–Applications of statistical methods in process management: ISO 11462-1:2001 -

Guidelines for implementation of statistical process control (SPC) – Part 1: Elements of SPC. Published
standard. http://www.iso.org/iso/catalogue_detail.htm?csnumber=33381 (2012)

10. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-3:2012 - Control
charts – Part 3: Acceptance control charts. Published standard. http://www.iso.org/iso/catalogue_detail.
htm?csnumber=40175 (2012)

11. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 - Control
charts – Part 2: Shewhart control charts. Published standard. http://www.iso.org/iso/catalogue_detail.
htm?csnumber=40174 (2013)

12. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-1:2014 - Control
 charts – Part 1: General guidelines. Published standard. http://www.iso.org/iso/catalogue_detail.htm?
csnumber=62649 (2014)

13. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-5:2014 - Control
charts – Part 5: Specialized control charts. Published standard. http://www.iso.org/iso/catalogue_detail.
htm?csnumber=40177 (2014)

14. Knoth, S.: spc: statistical process control—collection of some useful functions. http://CRAN.R-project.
org/package=spc (2015). R package version 0.5.1

15. Montgomery, D.: Statistical Quality Control, 7th edn. Wiley Global Education, New York (2012)

16. Sarkar, D.: Lattice: Multivariate Data Visualization with R. Springer, New York (2008). http://lmdvr.r-
forge.r-project.org. ISBN 978-0-387-75968-5

17. Scrucca, L.: qcc: an r package for quality control charting and statistical process control. R News 4/1,
11–17 (2004). http://CRAN.R-project.org/doc/Rnews/

18. Shewhart, W.: Economic Control of Quality in Manufactured Products. Van Nostrom, New York (1931)

19. Wickham, H.: ggplot2: Elegant Graphics for Data Analysis. Use R! Springer, New York (2009)

20. Zhu, W., Park, C.: edcc: An R package for the economic design of the control chart. J. Stat. Softw.
52(9), 1–24 (2013). http://www.jstatsoft.org/v52/i09/
© Springer International Publishing Switzerland 2015
Emilio L. Cano, Javier M. Moguerza and Mariano Prieto Corcoba, Quality Control with R, Use R!,
DOI 10.1007/978-3-319-24046-6_10

10. Nonlinear Profiles with R

Emilio L. Cano1, 2 , Javier M. Moguerza1 and
Mariano Prieto Corcoba3
(1) Department of Computer Science and Statistics, Rey Juan Carlos
University, Madrid, Spain
(2) Statistics Area, DHEP, The University of Castilla-La Mancha, Ciudad Real,
Spain
(3) ENUSA Industrias Avanzadas, Madrid, Spain

Abstract
In many situations, processes are often represented by a function that involves a
response variable and a number of predictive variables. In this chapter, we show
how to treat data whose relation between the predictive and response variables is
nonlinear and, as a consequence, cannot be adequately represented by a linear
model. This kind of data are known as nonlinear profiles. Our aim is to show
how to build nonlinear control limits and a baseline prototype using a set of
observed in-control profiles. Using R, we show how to afford situations in which
nonlinear profiles arise and how to plot easy-to-use nonlinear control charts.

10.1 Introduction
In Chapter 9 we presented control charts, considered to be the basic tools of
statistical process control (SPC). In particular, control charts are useful to test the
stability of a process when measuring one or more response variables
sequentially. However, in many situations, processes are often represented by a
function (called profile) that involves a response variable and a number of
predictive variables. The simplest profiles are those coming from a linear
relation between the response and the predictive variables. Nevertheless, in
many cases, a nonlinear relation exists among the variables under study and,
therefore, more complex models are demanded by the industry. In this chapter,
we show how to treat data whose relation between the predictive and response
variables is nonlinear and, as a consequence, cannot be adequately represented
by a linear model. This kind of data are known as nonlinear profiles.
One of the first approaches to nonlinear profiles in the industry can be
consulted in the document entitled “Advanced Quality System Tools” published
by Boeing in 1998 [1]. On page 91 of that publication, a location variability
chart of the flange-angle deviation from target at each location within a given
spar is shown, with the peculiarity that, in addition to the classical specification
limits, location averages and natural tolerance limits are included for each
location. These so-called natural tolerance limits are far from linear,
demonstrating that a nonlinear relation between the location and the flange-angle
deviation exists. A brief review on nonlinear profiles research can be consulted
in [12].

10.2 Nonlinear Profiles Basics

Consider a typical process where, given an item, a characteristic has to be
observed under different conditions or locations within the item.

Example 10.1.
Engineered woodboards.
To illustrate this chapter, we will use a data set in the SixSigma package [2].
It is a variation of the example introduced in [11]. Our data set is made up of
observations over 50 items, named P1, …, P50. Each item corresponds to an
engineered woodboard. For each woodboard, 500 observations (measurements)
of its density are taken at locations 0.001 in apart within the board. The
observations have been sequentially obtained. Each five observations correspond
to a sample from the same 4-h shift, that is, the first five observations correspond
to the first 4-h shift, and so on. The data are available in the SixSigma package as
data objects ss.data.wbx for the locations, and ss.data.wby for the density
measurements. Note that ss.data.wby is a matrix in which each column contains
the measurements corresponding to a woodboard. The column names allow to
easily identify the woodboard and the group it belongs to. Fig. 10.1 graphically
represents board P1, using standard R graphics functions. The 500 locations
where the measurements were taken are in the x-axis, while the y-axis is for the
density measurements at each location. We will refer to each set of 500
measurements taken over a given board as a profile.
Fig. 10.1 Single woodboard example. Plot of the first board in the sample (named P1). The density
measurements (y axis) are plotted against their location within the woodboard (x axis)

 □ 

From a mathematical point of view, consider a typical design of an in-control

process. Given an item, its associated profile is made up of n observations of a
characteristic Y of the item, y i , i = 1, …, n. We will refer to Y as the response
variable. Let X be the predictive variable. Let x i , i = 1, …, n be the
corresponding values of the predictive variable. For each item, the underlying in-
control model is:
(10.1)
where f is a smooth function, and ɛ i are error terms from an unknown
distribution. In this chapter, our aim is to show how to build confidence bands
and an estimation of f(x) using a set of observed in-control profiles. Such
confidence bands correspond to upper and lower nonlinear control limits around
function f(x) and will be used to detect out-of-control profiles. Although
rigorously speaking the term “confidence bands” may not be appropriate, we
will use it for the sake of clarity.

Example 10.2.
Engineered woodboards (cont.)
In the case at hand, n = 500, corresponding every x i to each one of the 500
locations taken 0.001 inches apart within the board, and every y i to each one of
the 500 measurements taken at the locations. The first 20 locations and
measurements corresponding to board P1 are:

Notice that the locations are common to all profiles.

In many applications, it is interesting to work with smoothed versions of the
profiles [3, 9]. We can calculate a smoothed version of profile P1. The
smoothing procedure can be made using different techniques. The function
smoothProfiles in the SixSigma package makes use of regularization theory in
order to smooth the profile. In particular, a support vector machine (SVM)
approach [10] is followed. In any case, this is transparent for the user and the
function acts as a black box, that is, a profile is given as input and the function
provides the smoothed version of the profile. For instance, profile P1 and its
smoothed version can be jointly plotted using the following code (see Fig. 10.2):
Fig. 10.2 Single woodboard example (smoothed). Profile P1 and its smoothed version

Notice that the smoothing procedure is working well as the smoothed version
seems to fit the original profile.
Function smoothProfiles accepts as first input argument a matrix containing
a set of profiles. In this case, as we are working only with profile P1, the input
matrix is the vector containing the measurements corresponding to profile P1
(ss.data.wby[, "P1"]). The second argument is the vector corresponding to the
locations where the measurements are taken (ss.data.wbx). The output of the
function is the smoothed version of profile P1, which is assigned to object
P1.smooth. The function allows the tuning of the SVM model inside. To this aim,
additional arguments may be provided to change the default settings (check the
function documentation for details by typing ?smoothProfiles). Note that this
function makes use of the e1071 library, and therefore it must be installed in R in
advance.
Function PlotProfiles plots a set of profiles in a matrix. In this case, we
have built a matrix with the smoothed version of profile P1 in the first column
and profile P1 itself in the second column using the cbind function. We may plot
the whole set of profiles with the following code (see Fig. 10.3):
Fig. 10.3 Woodboard example: whole set of profiles. Plot of the 50 profiles. The density measurements (y
axis) are plotted against their location within the woodboard (x axis)

Straightforwardly, we may plot the smoothed versions of the 50 profiles. We

firstly smooth the profiles:

And secondly, we plot the smoothed profiles (see Fig. 10.4):

Fig. 10.4 Woodboard example: whole set of smoothed profiles. Plot of the 50 smoothed profiles. The
smoothed density measurements (y axis) are plotted against their location within the woodboard (x axis)

 □ 

It is important to remark that we should always check that the smoothing

procedure is working well. Graphically, this can be easily checked by choosing
some profiles at random. For each profile chosen, we should plot jointly the
profile itself and its smoothed version, and in this way visually check if the
smoothing procedure is working appropriately, similarly to what we did in
Fig. 10.2. If the smoothing procedure seems to fail, the simplest way to proceed
is to use the original set of profiles without smoothing. More expert users may
try to change the default SVM parameters in the smoothProfiles function.

10.3 Phase I and Phase II Analysis

In the SPC methodology two phases are involved, Phase I and Phase II. In Phase
I, an in-control baseline subset of profiles is sought. The goal in this phase is to
test the stability of the process, in order to model the in-control process
performance. In Phase II, the goal is to monitor the process. To this aim, the
model adjusted in Phase I is used for the detection of out-of-control profiles over
a set of new profiles not previously analyzed.

10.3.1 Phase I
Example 10.3.
Engineered woodboards (cont.) Phase I analysis.
We will divide the set of 50 profiles into two subsets. A first subset will be
made up of the profiles obtained within the first seven 4-h shifts, that is, profiles
P1, …, P35. The second subset will be made up of the remaining profiles, i.e.,
profiles P36, …, P50. In Phase I, we will use the profiles in the first group to
seek for a baseline subset of in-control profiles and, with this baseline subset,
model the in-control process performance. We will refer to this first group as
Phase I group.
First, we create a matrix with the profiles in the Phase I group (columns 1–35
in matrix ss.data.wby):

Next, we calculate and plot confidence bands from the profiles in the Phase I
group and an estimation of function f(x) in (10.1), see Fig. 10.5. In the following,
we will refer to f(x) as the prototype profile.
Fig. 10.5 Woodboard example: Phase I. Plot of the 35 Phase I group profiles, confidence bands, and
estimation of f(x)

Given a set of profiles, function climProfiles calculates confidence bands

and an estimation of the prototype profile f(x). This function, by default,
calculates 99 % confidence bands (parameter alpha= 0.01 by default, see the
function documentation). In the case at hand, the calculations are done using
smoothed versions of the input profiles (parameter smoothprof = TRUE), and the
confidence bands have also been smoothed (parameter smoothlim = TRUE). The
functions allows to change these default settings at the user’s preference through
the function arguments argument.
Note that to plot the profiles and the confidence bands, we have used again
the plotProfiles function adding a new argument cLimits with the computed
confidence bands and the prototype profile.
Fig. 10.5 shows the profiles in the Phase I group (thin black lines), the
confidence bands calculated from these profiles (thick blue lines) and the
estimation of f(x) (thick green line). In the plot it is apparent that some profiles
are out of the inner region determined by the confidence bands, and, therefore,
may be considered out of control. The following code returns the list of out-of-
control profiles.

The function outProfiles returns a list of three vectors. The first vector
(labOut) contains the labels of the out-of-control profiles. The second vector
(idOut) contains the indexes of the out-of-control profiles. This vector is given
for completeness as in some cases the index may be preferable to the label. The
third vector contains the proportion of times that each profile remains out of the
confidence bands. By default, the function considers a profile to be out of
control if the proportion of times that this profile remains out of the confidence
bands is over 0.5. The user may change this default value by including the tol
argument with the desired value. For instance, the user may change the value of
the tol parameter to 0.80. In this case, only one out-of-control profile arises,
P28. After some investigation, the user may consider that profile P28 is out of
control, and should not belong to the in-control baseline subset of profiles. As a
consequence, it is removed and the confidence bands are calculated again
without this profile, following in this way a classical Phase I SPC strategy.
 With the following code the confidence bands are calculated again using 34
profiles, that is, without profile P28 in the Phase I group of profiles:

Fig. 10.6 shows the 34 profiles (thin black lines), the confidence bands
calculated from these 34 profiles (thick blue lines) and the estimation of f(x)
(thick green line). In the plot it is apparent that some profiles are close to the
confidence bands. The following code looks for the new list of out-of-control
profiles using the 0.8 tolerance previously defined by the user.
Fig. 10.6 Woodboard example: Phase I. Plot of the Phase I baseline in-control profiles, confidence bands,
and estimation of f(x)

In this case, the value NULL indicates that the vector is empty, and therefore,
no out-of-control profiles were detected. It is clear that now the process seems to
be in control. And these 34 profiles constitute the Phase I in-control baseline
group of profiles. □ 

10.3.2 Phase II
Once the confidence bands and the baseline profiles have been determined, we
will use these confidence bands to check if some not previously analyzed
profiles (typically new ones) are out of control.

Example 10.4.
Engineered woodboards (cont.) Phase II analysis.
In Phase II, we will check whether the profiles in the last three 4-h shifts,
profiles P36, …, P50, are out of control. Notice that these profiles were not used
to estimate the confidence bands in Phase I. Next, we create a matrix with these
profiles:

In order to check if the new profiles are out of control, we just use function
outProfiles over the new set of profiles and the control limits calculated in
Phase I.

Profiles P46, P47, and P48 are considered to be out of control. The
proportion of times that these profiles remain out of the confidence bands is 1,
0.95, and 1 respectively. As a consequence, the user may conclude that during
the last 4-h shift the process was not in-control and the causes should be
investigated.
We can plot the Phase II profiles, the confidence bands, the estimation of
f(x), and the out-of-control profiles. To do this, we use again the plotProfiles
function adding a new argument outControl with the labels or indexes of the
out-of-control profiles, see Fig. 10.7.
Fig. 10.7 Woodboard example: Phase II. Plot of the Phase II profiles, confidence bands, estimation of f(x),
and out-of-control profiles

Finally, for the sake of clarity, we can plot a graph only containing the out-
of-control profiles, the confidence bands, and the prototype, see Fig. 10.8. The
code to plot this chart is:

Fig. 10.8 Woodboard example: Phase II out of control. Plot of the Phase II out-of-control profiles,
confidence bands, estimation of f(x)
 In order to plot only the out-of-control profiles, we must change the default
value of the argument onlyout to TRUE.

10.4 A Simple Profiles Control Chart

In this section we show how to plot a very simple but useful profiles control
chart. The chart summarizes sequentially the results shown in the previous
sections. Since for each profile we can obtain the proportion of times that it
remains out of the confidence bands, we can plot this proportions jointly with the
tolerance fixed by the user so that sequential patterns may be detected, similarly
to the typical SPC strategy. Notice that in many nonlinear profiles real problems
the concept “out of control” depends on the user’s knowledge and requirements.
More experienced users may implement this using real-time recorded data via an
R interface.

Example 10.5.
Engineered woodboards (cont.) Profiles control chart.
Function outProfiles provides a vector which includes the proportion of
times that each profile is out of the confidence bands in the pOut element of the
output list. Notice that the pOut is independent of the tol parameter. To plot this
chart (see Fig. 10.9) we use the function plotControlProfiles as follows:
Fig. 10.9 Woodboard example: Profiles control chart. Sequential plot of the profiles out-of-control rate for
a given tolerance

 □ 

10.5 ISO Standards for Nonlinear Profiles and R

To the best of our knowledge, no standards have been published regarding the
monitoring of nonlinear profiles. Nevertheless, nonlinearity is a subject that is
taken into account in some ISO Standards. For example, part 5 of ISO 11843
series [6] includes a methodology for nonlinear calibration within the capability
of detection topic. On the other hand, the out-of-control rates obtained as a result
of a nonlinear profiles analysis can be monitored using a Shewhart p-chart, as
described in Chapter 9. This kind of charts can be consulted in ISO 7870-2 [4].
Regarding specialized control charts, ISO 7870-5 [5] contains several charts
appropriate for time-dependent and non-normal data.
Finally, although out of the scope of ISO technical committee TC69, SVM
are considered in several standards such as ISO 13179-1 [7] or ISO 15746-1 [8].

References
1. Boeing Commercial Airplane Group, M.D.P.Q.A.D.: Advanced Quality System Tools, AQS D1-9000-
1. Toolbox (1998). url http://www.boeingsuppliers.com/supplier/d1-9000-1.pdf
2. Cano, E.L., Moguerza, J.M., Redchuk, A.: Six sigma with R. Statistical Engineering for Process
Improvement, Use R!, vol. 36. Springer, New York (2012). url http://www.springer.com/statistics/book/
978-1-4614-3651-5

3. Cano, J., Moguerza, J.M., Psarakis, S., Yannacopoulos, A.N.: Using statistical shape theory for the
monitoring of nonlinear profiles. Appl. Stoch. Model. Bus. Ind. 31(2), 160–177 (2015). doi:10.1002/
asmb.2059. url http://dx.doi.org/10.1002/asmb.2059

4. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-2:2013 - Control

charts – Part 2: Shewhart control charts. Published standard (2013). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40174

5. ISO TC69/SC4–Applications of statistical methods in process management: ISO 7870-5:2014 - Control

charts – Part 5: Specialized control charts. Published standard (2014). url http://www.iso.org/iso/
catalogue_detail.htm?csnumber=40177

6. ISO TC69/SC6–Measurement methods and results: ISO 11843-5:2008 - Capability of detection – Part
5: Methodology in the linear and non-linear calibration cases. Published standard (2012). url http://
www.iso.org/iso/catalogue_detail.htm?csnumber=42000

7. ISO/TC 108, Mechanical vibration, shock and condition monitoring, Subcommittee SC 5, Condition
monitoring and diagnostics of machines: Condition monitoring and diagnostics of machines – Data
interpretation and diagnostics techniques – Part 1: General guidelines. Published standard (2012).
url http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=39836

8. ISO/TC 184, Automation systems and integration, Subcommittee SC 5, Interoperability, integration and
architectures of automation systems and applications: Automation systems and integration – Integration
of advanced process control and optimization capabilities for manufacturing systems – Part 1:
Framework and functional model. Published standard (2015). url http://www.iso.org/iso/catalogue_
detail.htm?csnumber=61131

9. Moguerza, J., Muñoz, A., Psarakis, S.: Monitoring nonlinear profiles using support vector machines. In:
Rueda, L., Mery, D., Kittler, J. (eds.) Progress in Pattern Recognition, Image Analysis and
Applications. Lecture Notes in Computer Science, vol. 4756, pp. 574–583. Springer, Heidelberg
(2007). doi:10.1007/978-3-540-76725-1_60. url http://dx.doi.org/10.1007/978-3-540-76725-1_60

10. Moguerza, J.M., Muñoz, A.: Support vector machines with applications. Stat. Sci. 21(3), 322–336
(2006)
[MATH][CrossRef]

11. Walker, E., Wright, W.: Comparing curves with additive models. J. Qual. Technol. 34(1), 118–129
(2002)

12. Woodall, W.H.: Current research in profile monitoring. Producão 17(3), 420–425 (2007). Invited paper
Appendix A: Shewhart Constants for Control Charts
The main Shewhart constants d 2 , d 3 , and c 4 can be obtained for any n using R
as shown in the following examples:

The rest of Shewhart constants that can be found at any textbook are
computed using those three basic constants. A full table of constants can also be
generated using R. Table A.1 shows the constants used in this book. There are
other constants not covered by this book which could also be computed just
using the appropriate formula. A data frame with the constants in Table A.1 can
be obtained with the following code:
Table A.1 Shewhart constants
n d2 d3 c4 A2 D3 D4 B3 B4
2 1.1284 0.8525 0.7979 1.8800 0.0000 3.2665 0.0000 3.2665
3 1.6926 0.8884 0.8862 1.0233 0.0000 2.5746 0.0000 2.5682
4 2.0588 0.8798 0.9213 0.7286 0.0000 2.2821 0.0000 2.2660
5 2.3259 0.8641 0.9400 0.5768 0.0000 2.1145 0.0000 2.0890
6 2.5344 0.8480 0.9515 0.4832 0.0000 2.0038 0.0304 1.9696
7 2.7044 0.8332 0.9594 0.4193 0.0757 1.9243 0.1177 1.8823
8 2.8472 0.8198 0.9650 0.3725 0.1362 1.8638 0.1851 1.8149
9 2.9700 0.8078 0.9693 0.3367 0.1840 1.8160 0.2391 1.7609
10 3.0775 0.7971 0.9727 0.3083 0.2230 1.7770 0.2837 1.7163
11 3.1729 0.7873 0.9754 0.2851 0.2556 1.7444 0.3213 1.6787
12 3.2585 0.7785 0.9776 0.2658 0.2833 1.7167 0.3535 1.6465
13 3.3360 0.7704 0.9794 0.2494 0.3072 1.6928 0.3816 1.6184
14 3.4068 0.7630 0.9810 0.2354 0.3281 1.6719 0.4062 1.5938
15 3.4718 0.7562 0.9823 0.2231 0.3466 1.6534 0.4282 1.5718
16 3.5320 0.7499 0.9835 0.2123 0.3630 1.6370 0.4479 1.5521
17 3.5879 0.7441 0.9845 0.2028 0.3779 1.6221 0.4657 1.5343
18 3.6401 0.7386 0.9854 0.1943 0.3913 1.6087 0.4818 1.5182
19 3.6890 0.7335 0.9862 0.1866 0.4035 1.5965 0.4966 1.5034
20 3.7349 0.7287 0.9869 0.1796 0.4147 1.5853 0.5102 1.4898
21 3.7783 0.7242 0.9876 0.1733 0.4250 1.5750 0.5228 1.4772
22 3.8194 0.7199 0.9882 0.1675 0.4345 1.5655 0.5344 1.4656
23 3.8583 0.7159 0.9887 0.1621 0.4434 1.5566 0.5452 1.4548
24 3.8953 0.7121 0.9892 0.1572 0.4516 1.5484 0.5553 1.4447
25 3.9306 0.7084 0.9896 0.1526 0.4593 1.5407 0.5648 1.4352

The table of constants is also available as a one-page pdf document through

one of the SixSigma package vignettes :

Appendix B: ISO Standards Published by the ISO/TC69:

Application of Statistical Methods
This appendix contains all the international standards and technical reports
published by the ISO TC69—Application of Statistical Methods, grouped by
subcommittees. Please note that ISO standards are continously evolving. All
references to standards in this appendix and throughout the book are specific for
a given point in time. In particular, this point in time is end of June 2015.
Therefore, some new standards may have appeared when you are reading this
book, or even other changes may have happen in ISO. For example, at the time
of publishing a subcommittee has changed its denomination! Keep updated in
the committee website: http://www.iso.org/iso/home/store/catalogue_tc/
catalogue_tc_browse.htm?commid=49742 .

TC69/SCS: Secretariat
ISO 11453:1996 Statistical interpretation of data—Tests and confidence
intervals relating to proportions.
ISO 11453:1996/Cor 1:1999 .
ISO 16269-4:2010 Statistical interpretation of data—Part 4: Detection and
treatment of outliers.
ISO 16269-6:2014 Statistical interpretation of data—Part 6: Determination
of statistical tolerance intervals.
ISO 16269-7:2001 Statistical interpretation of data—Part 7: Median—
Estimation and confidence intervals.
ISO 16269-8:2004 Statistical interpretation of data—Part 8: Determination
of prediction intervals.
ISO 2602:1980 Statistical interpretation of test results—Estimation of the
mean—Confidence interval.
ISO 2854:1976 Statistical interpretation of data—Techniques of estimation
and tests relating to means and variances.
ISO 28640:2010 Random variate generation methods.
ISO 3301:1975 Statistical interpretation of data—Comparison of two means
in the case of paired observations.
ISO 3494:1976 Statistical interpretation of data—Power of tests relating to
means and variances.
ISO 5479:1997 Statistical interpretation of data—Tests for departure from
the normal distribution.
ISO/TR 13519:2012 Guidance on the development and use of ISO statistical
publications supported by software.
ISO/TR 18532:2009 Guidance on the application of statistical methods to
quality and to industrial standardization.

TC69/SC1: Terminology and Symbols

 ISO 3534-1:2006 Statistics—Vocabulary and symbols—Part 1: General
statistical terms and terms used in probability.
ISO 3534-2:2006 Statistics—Vocabulary and symbols—Part 2: Applied
statistics.
ISO 3534-3:2013 Statistics—Vocabulary and symbols—Part 3: Design of
experiments.
ISO 3534-4:2014 Statistics—Vocabulary and symbols—Part 4: Survey
sampling.

TC69/SC4: Applications of Statistical Methods in

Process Management
ISO 11462-1:2001 Guidelines for implementation of statistical process
control (SPC)—Part 1: Elements of SPC.
ISO 11462-2:2010 Guidelines for implementation of statistical process
control (SPC)—Part 2: Catalogue of tools and techniques.
ISO 22514-1:2014 Statistical methods in process management—Capability
and performance—Part 1: General principles and concepts.
ISO 22514-2:2013 Statistical methods in process management—Capability
and performance—Part 2: Process capability and performance of time-
dependent process models.
ISO 22514-3:2008 Statistical methods in process management—Capability
and performance—Part 3: Machine performance studies for measured
data on discrete parts.
ISO 22514-6:2013 Statistical methods in process management—Capability
and performance—Part 6: Process capability statistics for characteristics
following a multivariate normal distribution.
ISO 22514-7:2012 Statistical methods in process management—Capability
and performance—Part 7: Capability of measurement processes.
ISO 22514-8:2014 Statistical methods in process management—Capability
and performance—Part 8: Machine performance of a multi-state
production process.
ISO 7870-1:2014 Control charts—Part 1: General guidelines.
ISO 7870-2:2013 Control charts—Part 2: Shewhart control charts.
ISO 7870-3:2012 Control charts—Part 3: Acceptance control charts.
 ISO 7870-4:2011 Control charts—Part 4: Cumulative sum charts.
ISO 7870-5:2014 Control charts—Part 5: Specialized control charts.
ISO/TR 22514-4:2007 Statistical methods in process management—
Capability and performance—Part 4: Process capability estimates and
performance measures.

TC69/SC5: Acceptance Sampling

ISO 13448-1:2005 Acceptance sampling procedures based on the allocation
of priorities principle (APP)—Part 1: Guidelines for the APP approach.
ISO 13448-2:2004 Acceptance sampling procedures based on the allocation
of priorities principle (APP)—Part 2: Coordinated single sampling plans
for acceptance sampling by attributes.
ISO 14560:2004 Acceptance sampling procedures by attributes—Specified
quality levels in nonconforming items per million.
ISO 18414:2006 Acceptance sampling procedures by attributes—Accept-
zero sampling system based on credit principle for controlling outgoing
quality.
ISO 21247:2005 Combined accept-zero sampling systems and process
control procedures for product acceptance.
ISO 24153:2009 Random sampling and randomization procedures.
ISO 2859-10:2006 Sampling procedures for inspection by attributes—Part
10: Introduction to the ISO 2859 series of standards for sampling for
inspection by attributes.
ISO 2859-1:1999 Sampling procedures for inspection by attributes—Part 1:
Sampling schemes indexed by acceptance quality limit (AQL) for lot-by-
lot inspection.
ISO 2859-1:1999/Amd 1:2011 .
ISO 2859-3:2005 Sampling procedures for inspection by attributes—Part 3:
Skip-lot sampling procedures.
ISO 2859-4:2002 Sampling procedures for inspection by attributes—Part 4:
Procedures for assessment of declared quality levels.
ISO 2859-5:2005 Sampling procedures for inspection by attributes—Part 5:
System of sequential sampling plans indexed by acceptance quality limit
(AQL) for lot-by-lot inspection.
 ISO 28801:2011 Double sampling plans by attributes with minimal sample
sizes, indexed by producer’s risk quality (PRQ) and consumer’s risk
quality (CRQ).
ISO 3951-1:2013 Sampling procedures for inspection by variables—Part 1:
Specification for single sampling plans indexed by acceptance quality
limit (AQL) for lot-by-lot inspection for a single quality characteristic
and a single AQL.
ISO 3951-2:2013 Sampling procedures for inspection by variables—Part 2:
General specification for single sampling plans indexed by acceptance
quality limit (AQL) for lot-by-lot inspection of independent quality
characteristics.
ISO 3951-3:2007 Sampling procedures for inspection by variables—Part 3:
Double sampling schemes indexed by acceptance quality limit (AQL)
for lot-by-lot inspection.
ISO 3951-4:2011 Sampling procedures for inspection by variables—Part 4:
Procedures for assessment of declared quality levels.
ISO 3951-5:2006 Sampling procedures for inspection by variables—Part 5:
Sequential sampling plans indexed by acceptance quality limit (AQL)
for inspection by variables (known standard deviation).
ISO 8422:2006 Sequential sampling plans for inspection by attributes.
ISO 8423:2008 Sequential sampling plans for inspection by variables for
percent nonconforming (known standard deviation).

TC69/SC6: Measurement Methods and Results

ISO 10576-1:2003 Statistical methods—Guidelines for the evaluation of
conformity with specified requirements—Part 1: General principles.
ISO 10725:2000 Acceptance sampling plans and procedures for the
inspection of bulk materials.
ISO 11095:1996 Linear calibration using reference materials.
ISO 11648-1:2003 Statistical aspects of sampling from bulk materials—Part
1: General principles.
ISO 11648-2:2001 Statistical aspects of sampling from bulk materials—Part
2: Sampling of particulate materials.
ISO 11843-1:1997 Capability of detection—Part 1: Terms and definitions.
ISO 11843-2:2000 Capability of detection—Part 2: Methodology in the
linear calibration case.
ISO 11843-3:2003 Capability of detection—Part 3: Methodology for
determination of the critical value for the response variable when no
calibration data are used.
ISO 11843-4:2003 Capability of detection—Part 4: Methodology for
comparing the minimum detectable value with a given value.
ISO 11843-5:2008 Capability of detection—Part 5: Methodology in the
linear and non-linear calibration cases.
ISO 11843-6:2013 Capability of detection—Part 6: Methodology for the
determination of the critical value and the minimum detectable value in
Poisson distributed measurements by normal approximations.
ISO 11843-7:2012 Capability of detection—Part 7: Methodology based on
stochastic properties of instrumental noise.
ISO 21748:2010 Guidance for the use of repeatability, reproducibility and
trueness estimates in measurement uncertainty estimation.
ISO 5725-1:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 1: General principles and definitions.
ISO 5725-2:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 2: Basic method for the determination of
repeatability and reproducibility of a standard measurement method.
ISO 5725-3:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 3: Intermediate measures of the precision of a
standard measurement method.
ISO 5725-4:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 4: Basic methods for the determination of the
trueness of a standard measurement method.
ISO 5725-5:1998 Accuracy (trueness and precision) of measurement
methods and results—Part 5: Alternative methods for the determination
of the precision of a standard measurement method.
ISO 5725-6:1994 Accuracy (trueness and precision) of measurement
methods and results—Part 6: Use in practice of accuracy values.
ISO/TR 13587:2012 Three statistical approaches for the assessment and
interpretation of measurement uncertainty.
ISO/TS 21749:2005 Measurement uncertainty for metrological applications
 —Repeated measurements and nested experiments.
ISO/TS 28037:2010 Determination and use of straight-line calibration
functions.

TC69/SC7: Applications of Statistical and Related

Techniques for the Implementation of Six Sigma
ISO 13053-1:2011 Quantitative methods in process improvement—Six
Sigma—Part 1: DMAIC methodology.
ISO 13053-2:2011 Quantitative methods in process improvement—Six
Sigma—Part 2: Tools and techniques.
ISO 17258:2015 Statistical methods—Six Sigma—Basic criteria underlying
benchmarking for Six Sigma in organisations.
ISO/TR 12845:2010 Selected illustrations of fractional factorial screening
experiments.
ISO/TR 12888:2011 Selected illustrations of gauge repeatability and
reproducibility studies.
ISO/TR 14468:2010 Selected illustrations of attribute agreement analysis.
ISO/TR 29901:2007 Selected illustrations of full factorial experiments with
four factors.
ISO/TR 29901:2007/Cor 1:2009 .

TC69/SC8: Application of Statistical and Related

Methodology for New Technology and Product
Development
ISO 16336:2014 Applications of statistical and related methods to new
technology and product development process—Robust parameter design
(RPD).

Appendix C: R Cheat Sheet for Quality Control

R Console
↑ Navigate expressions history
 CTRL+L Clear console
ESC Cancel current expression

RStudio
CTRL + number Go to panel:
1: Editor
2: Console
3: Help
4: History
5: Files
6: Plots
7: Packages
8: Environment
CTRL + MAYÚS + K knit current R Markdown report
CTRL + MAYÚS + I Compile R Sweave (LaTeX) current report
CTRL + S Save file
F1 Contextual help (upon the cursor position)
CTRL + F
Activates search (within different panels) 1
<console>

↑ Expressions history
CTRL+L Clear console
ESC Cancel current expression
<editor and console>
TAB Prompt menu:
Select objects in the workspace
Select function arguments (when in parenthesis)
Select list elements (after the $ character)
Select chunk options (when in chunk header)
Select files (when in quotes)
<editor>
CTRL + ENTER Run current line or selection
 CTRL + MAYÚS + S Source full script
CTRL + ALT + I Insert code chunk
CTRL + ALT + C Run current code chunk (within a chunk)
CTRL + MAYÚS + P Repeat las code run
CTRL + MAYÚS + C Comment current line or selection (add # at the beginning
of the line)
CTRL + D Delete current line

ALT + Move current line or selection up or down

ALT + MAYÚS + Copy current line or selection up or down

Help
?, help Help on a function

??, help.search Search help over a topic

apropos Show function containing a given string

General
; Separate expressions in the same line
<- Assignment operator
{ <code>} Code blocks within curly brackets
# Comment (ignores the remaining of the line)
‘ <string> ‘ (backtick) Allow using identifiers with special
characters and/or blank spaces
Math Operators
+, - /, *, Arithmetic

<, >, <=, >=, ==, !=, %in% Comparisons

&, &&, |, ||, !

Logic operations 2
Integer Operations
%/% Integer division

%% Module (remainder of a division)

Comparison Functions
all Are all elements TRUE?

any Is any element TRUE?

Math Functions
sqrt Square root
exp, log Exponential and logarithmic

sin, cos, tan Trigonometry

asin, acos, atan Inverse trigonometry

abs Absolute value

round, floor, ceiling Rounding

max, min Maximum and minimum

sum, prod Sums and products

 cumsum, cumprod, cummax, cummin Cumulative operations

factorial Factorial

choose Binomial coefficient

Vectors
c Create a vector (combine values)

seq: Creates a sequence

rep Repeat values

length Vector length

[ ] Item selection

sort Sorting

order Get indices ordered by magnitude

rev Reverse order

unique Get unique values

which Devuelve índices que cumplen condición

 union, intersect, setdiff, setequal, %in% Sets operations

Matrices
matrix Create a matrix

%*% Matrix multiplication

t Transpose a matrix
solve Inverse a matrix

colSums, rowSums Sum by rows or columns

colMeans, rowMeans Average by rows or columns

colnames, rownames Column or rows names

dim, nrow, ncol Dimensions

rbind, cbind Add columns or rows to a matrix matrix

 [, ] Items selection

Factors
factor Create a factor

gl Generate levels of a factor

expand.grid Generate factors combinations

Dates
as.Date Convert to date

format Returns a date in a given format

ISOweek Returns the week of a date in ISO format ( ISOweek package)

Character String
nchar Get number of characters

paste, paste0 Paste character strings

cat Print a character string in the console

Lists
list Create a list
Data Frames
data.frame Create a data frame

str Get data frame structure: column names, types, and sample data
 head, tail Get first or last rows of a data frame

subset Get a (filtered) subset of data

aggregate Get aggregate data applying a function over groups

Files
 download.file Download files

read.table Import data

read.csv2 Import data from csv file

write.csv2 Save csv data file

scan Read data from the console or text

save Save an R data file

load Load an R data file into the workspace

Data Simulation
set.seed Fix the seed 3

sample Draw a random sample from a set

rnorm, rbinom, rpois, … Draw random variates from probability distributions

(normal, binomial, Poisson, …)
Graphics
boxplot Box plot

hist Histogram

plot Scatter plot (for two numeric vectors)

barplot Bar plot (for counts)

par Graphical parameters (see ?par )

par Get or set graphical parameters

main Add a title to a plot (top)
sub Add a subtitle to a plot (bottom)
xlab, ylab Set horizontal and vertical axes labels
legend Add a legend
col Set color (see link at the end)
las Axes labels orientation
lty Line type
 lwd Line width
pch Symbol (for points)

graphics Graphical functions

points Add points to a plot

abline Draw a straight line (horizontal, vertical, or with a slope)
text Put text in the plot
mtext Add text in the margins
Descriptive Statistics
table Count the elements within each category

summary Five-num summary (plus the mean)

mean Average

median Median
 quantile Percentiles, quantiles

var Variance

sd Standard deviation

cor Correlation

max, min, range Maximum, minimum, range

Acceptance Sampling
Simple sampling plan
 Double sampling plan

Assess plan

Control Charts
qcc Library
qcc.groups Create object with grouped data

qcc Create control chart object. Some options:

data Vector, matrix or data frame with the data

type One of: “xbar”, “R”, “S”, “xbar.one”, “p”, “np”, “c”, “u”, “g”
sizes Vector with sample sizes for charts: “p”, “np”, o “u”
center Known center value
std.dev Known standard deviation
limits Phase I limits (vector with LCL, UCL)
plot If FALSE the chart is not shown
newdata Phase II data
newsizes Phase II sample sizes
nsigmas Number of standard deviations to compute control limits
confidence.level Confidence level to compute control limits (instead
of nsigmas)

Control charts for variables:

Control charts for attributes:
Process Capability
qcc Package
process.capability Needs a qcc object
SixSigma Package
 ss.study.ca Returns graphical and numerical capability analysis

qualityTools Package
cp Process capability indices
Pareto Analysis
qcc Package
cause.and.effect Cause-and-effect analysis
pareto.chart Pareto Chart
qualityTools Package
paretoChart Pareto chart
SixSigma Package
ss.ceDiag Cause-and-effect diagram
Probability
p* Distribution function for a given value

q* Quantile for a given cumulative probability probabilidad ¡¡¿¿= qnorm(p

= 0.95, mean = 10, sd = 1) @
d* Density for a given value (probability in discrete distributions)
Objects
str Get the structure of an object

class Get the class of an object

is.* Return a logic value TRUE if the object is of the specified class, for
example, numeric
as.* Coerce to the specified class

Vectorized Functions
tapply Apply a function to values for each level of a factor
lapply Apply a function to each element of a list returning a list

sapply Apply a function to each element of a list returning a list,vector, or

matrix

apply Apply a function to the rows or columns of a matrix

split Divide an object over factor levels returning a list

mappy Multivariate version of sapply

 rapply Recursive version of lapply

Programming
for Loop over the values of a vector or list

if …else Control flow

 function Create functions

Useful functions within a function:

warning

message

stop Stops the execution of code

Reports
xtable Package
xtable Create tables in different formats, e.g., LaTeX, HTML

caption Table caption

label Table label
align Alignment
 digits Number of significant digits
display Format (see ?xtable )

More options can be passed to the print generic function ?

print.xtable

col1 col2
1 1 3
2 2 4

Package knitr
knit Converts Rmd, Rhtml and Rnw files into HTML, MS Word o PDF
reports. See documentación at http://yihui.name/knitr/ .
Main options in a code chunk header:

echo Show code in the report

error Show error messages in the report
warning Show warning messages in the report
message Show messages in the report
eval Evaluate the chunk
fig.align Figure alignment
fig.width Figure width (in inches, 7 by default)
fig.height Figure height (in inches, 7 by default)
out.width Figure width within the report
out.height Figure height within the report
fig.keep Keep plots in the report
include Show text output in the report
results How to show the reports

Useful Links
 R-Project: http://www.r-project.org
RStudio: http://www.rstudio.com
Easy R practice: http://tryr.codeschool.com/
List of colours with names: http://www.stat.columbia.edu/~tzheng/files/
Rcolor.pdf
http://www.cyclismo.org/tutorial/R/
http://www.statmethods.net/index.html
Recipes : http://www.cookbook-r.com/
Search documentation: http://www.rdocumentation.org/
http://www.computerworld.com/s/article/9239625/Beginner_s_guide_to_
R_Introduction
http://www.inside-r.org/
http://www.r-bloggers.com/
Google R styleguide: http://google-styleguide.googlecode.com/svn/trunk/
Rguide.xml
Book Six Sigma with R : www.sixsigmawithr.com

R Packages and Functions Used in the Book

Symbols
:
;
¡-
¿
[
[[
$

A
AcceptanceSampling
assess
find.plan
OC2c
B
base
anyNA
array
as.Date
as.numeric
c
class
colMeans
colnames
cut
data.frame
detach
diff
dir
exp
expression
factor
format
getwd
gl
grep
is.na
is.numeric
length
library
list
list.dirs
list.files
load
log
ls
matrix
max
mean
names
ncol
nrow
 options
order
paste
range
rbind
rep
require
rev
rm
RNG
round
rownames
rowSums
sample
sapply
save
save.image
scan
search
seq
seq_along
set.seed
setwd
sort
source
sqrt
strsplit
subset
sum
summary
table
tapply
typeof
unlist
which

C
car
 bcPower
powerTransform
constants
LETTERS
letters
pi

D
Deducer

F
foreign

G
ggplot2
geom_boxplot
geom_histogram
geom_point
ggplot
labs
graphics
abline
barplot
boxplot
curve
hist
legend
lines
par
plot
points
polygon
rect
text
grDevices
grid
H
h5
Hmisc
binconf

I
IQCC
ISOweek

K
knitr

L
lattice
bwplot
dotplot
histogram
llines
panel.dotplot
panel.superpose
trellis.par.get
trellis.par.set
xyplot

M
MASS
boxcox
MSQC

N
nortest
ad.test

Q
qcc
 cause.and.effect
cusum
ewma
mqcc
oc.curves
pareto.chart
process.capability
qcc
qcc.groups
qcc.options
summary.qcc
qcr
qicharts
paretochart
qic
trc
qualityTools
cp
paretoChart

R
Rcmdr
RJDBC
RMongo
RMySQL
RODBC
ROracle
RPostgreSQL
RSQLite
rvest
html
html_nodes
html_text

S
SixSigma
climProfiles
outProfiles
 plotControlProfiles
plotProfiles
smoothProfiles
ss.ca.cp
ss.ca.cpk
ss.ca.study
ss.ca.z
ss.cc
ss.cc.getc4
ss.cc.getd2
ss.cc.getd3
ss.ceDiag
ss.data.bills
ss.data.wbx
ss.data.wby
special values
FALSE
i
Inf
LETTERS
letters
month.abb
month.name
NA
NaN
NULL
pi
TRUE
stats
aggregate
anova
arima
bartlett.test
binom.test
chisq.test
coef
complete.cases
cov
dbinom
 dhyper
dnorm
dpois
dyper
IQR
lm
mad
median
pbinom
phyper
pnorm
poisson.test
ppois
prop.test
qbinom
qchisq
qhyper
qnorm
qpois
qqline
qqnorm
quantile
rbinom
rhyper
rnorm
rpois
sd
shapiro.test
t.test
ts
stats ( cont. )
var
var.test
weighted.mean

T
tolerance
acc.samp
U
utils
apropos
available.packages
browseVignettes
demo
download.file
example
head
help
history
install.packages
installed.packages
loadhistory
read.csv
read.table
remove.packages
savehistory
str
Sweave
vignette
write.csv

X
XLConnect
XML
xmlSApply
xmlTreeParse
xpathApply
xtable
print.xtable
xtable

ISO Standards Referenced in the Book

I
ISO 10576-1:2003
ISO 10725:2000
ISO 11095:1996
ISO 11453:1996
ISO 11453:1996/Cor 1:1999
ISO 11462-1
ISO 11462-1:2001
ISO 11462-2
ISO 11462-2:2010
ISO 11648-1:2003
ISO 11648-2:2001
ISO 11843-1:1997
ISO 11843-2:2000
ISO 11843-3:2003
ISO 11843-4:2003
ISO 11843-5:2008
ISO 11843-6:2013
ISO 11843-7:2012
ISO 12207:2008
ISO 13053-1
ISO 13053-1:2011
ISO 13053-2
ISO 13053-2:2011
iso 13179-1:2012
ISO 13448-1:2005
ISO 13448-2:2004
ISO 14560:2004
ISO 15746-1:2015
ISO 16269-4:2010
ISO 16269-6:2014
ISO 16269-7:2001
ISO 16269-8:2004
ISO 16336:2014
ISO 17258:2015
ISO 18414:2006
ISO 19011:2011
ISO 21247:2005
ISO 21748:2010
ISO 22514-1:2014
ISO 22514-2:2013
ISO 22514-3:2008
ISO 22514-6:2013
ISO 22514-7:2012
ISO 22514-8:2014
ISO 24153:2009
ISO 2602:1980
ISO 2854:1976
ISO 2859-10:2006
ISO 2859-1:1999
ISO 2859-1:1999/Amd 1:2011
ISO 2859-3:2005
ISO 2859-4:2002
ISO 2859-5:2005
ISO 28640
ISO 28640:2010
ISO 28801:2011
ISO 3301:1975
ISO 3494:1976
ISO 3534-1
ISO 3534-1:2006
ISO 3534-2
ISO 3534-2:2006
ISO 3534-3:2013
ISO 3534-4
ISO 3534-4:2014
ISO 3951-1:2013
ISO 3951-2:2013
ISO 3951-3:2007
ISO 3951-4:2011
ISO 3951-5:2006
ISO 5479:1997
ISO 5725-1:1994
ISO 5725-2:1994
ISO 5725-3:1994
ISO 5725-4:1994
ISO 5725-5:1998
ISO 5725-6:1994
ISO 7870-1
ISO 7870-1:2014
ISO 7870-2
ISO 7870-2:2013
ISO 7870-3
ISO 7870-3:2012
ISO 7870-4
ISO 7870-4:2011
ISO 7870-5
ISO 7870-5:2014
ISO 8422:2006
ISO 8423:2008
ISO 8601:2004
ISO 9000:2005
ISO 9001:2008
ISO 9004:2009
ISO FDIS 7870-6
ISO.16336:2014
ISO.22514-3:2008
ISO/DIS 22514-4
ISO/IEC 14882:2014
ISO/IEC 9899
ISO/TR 12845
ISO/TR 12845:2010
ISO/TR 12888
ISO/TR 12888:2011
ISO/TR 13519:2012
ISO/TR 13587:2012
ISO/TR 14468:2010
ISO/TR 18532:2009
ISO/TR 22514-4:2007
ISO/TR 29901
ISO/TR 29901:2007
ISO/TR 29901:2007/Cor 1:2009
ISO/TS 21749:2005
ISO/TS 28037:2010

Subject Index
Symbols
C p
C pkL
C pkU
C pk
C pmk
F ( x )
H 0
H 1
P p
P pkL
P pkU
P pk
Q 1
Q 2
Q 3
α

β
X 2
δ
γ
λ
μ

σ
σ 2
σ ST
Ѳ
c 4
d 2
d 3
f ( x )
p -value
s 2
LATEX
FDIS
IDE
MDB
D 3
D 4
d 2
d 3
ANOVA
ANSI
AQL
ARL
BSI
CD
CLI
CL
CRAN
CSV
DBMS
DFSS
DIS
DMAIC
DPMO
DoE
EWMA
FAQ
FDA
FDIS
FOSS
GUI
HTML
ICS
IEC
IQR
ISO
JTC
LCL
LL
LSL
LTPD
LT
MAD
MR
NCD
NP
OBP
OC
ODBC
OS
PAS
PLC
PMBoK
PWI
QFD
RCA
RNG
RPD
RSS
RUG
R
SC
SDLC
SME
SPC
SR
SVM
TC
TMB
TR
TS
UCL
UL
URL
USL
VoC
VoP
WD
WG
XML
5Ms

A
acceptability constant
acceptance
acceptance quality level
acceptance sampling
for attributes
for variables
AENOR
aggregate values
alternative hypothesis
Anderson-Darling test
anonymous function
argument
assignable causes
asymmetric
attribute
attributes control charts
average

B
bar chart
baseline profile
Bernoulli trial
Big Data
binomial
binomial distribution
black belt
box plot
box-and-whisker plot
Box-Cox transformation
brainstorming
browser
C
c chart
capability analysis
capability index
capability indices
cause
cause-and-effect
cause-and-effect diagram
center line
central limit theorem
central tendency
character
characteristic
check sheet
chunk (code)
class
cluster
cluster sampling
column
comment
common causes
complementary event
confidence bands
confidence interval
confidence level
consumer’s risk
continuous distributions
continuous scale
continuous variable
control chart
constants
individuals
control chart power
control chart tests
control chart zones
control limits
cost of poor quality
crossings
csv file
customer
customer specification limit
customer’s risk
CUSUM chart
cusum coefficients
cycle

D
data
acquisition
database
export
raw data
treatment
data acquisition
data analysis
data cleaning
data collection plan
data frame
data import
data reuse
data structures
data transformation
data type
database
dataset
date
dates
debugging
decreasing
defect
defect-free
defective
defective fraction
defects
defects per million opportunities
defects per unit
degrees of freedom
density
design for six sigma
design of experiments
design specifications
destructive test
device
dimension
discrete distribution
distribution
normal distribution
distribution function
distribution parameters
distributions
binomial
exponential
geometric
hypergeometric
lognormal
negative binomial
normal
Poisson
uniform
Weibull
double
double sampling plan

E
effect
environment
error
error type I
error type II
escaping
especial variability
estimation
estimator
event
EWMA chart
exclude
expectation
expected value
experimental design
exploratory data analysis
export graphics
export data
extract
extraction
extreme values

F
factor
files
finite population
first-time yield
fishbone diagram
fitness of use
flow chart
formula
fraction defective
frequency
frequency table
function
functional tolerance

G
global standard deviation
goodness of fit
graphical system
graphics
grid
par
graphics options
green belt
groups
H
histogram
histogram bars
history
hypothesis test
hypothesis testing

I
I Chart
image
import csv
import data
improvement
in-control process
independent event
independent samples
independent trial
index
individuals chart
individuals control chart
inference
inspection
integer
interface
Internet
interquartile range
interval
interval estimation
interview
IQR
Ishikawa diagram
Ishikawa, Kaoru
ISO
ISO Council
ISO members
ISO website
K
knit

L
label
larger-the-better
Lean
length
liaison
liaisons
category A
category B
category D
limits
control limits
specification limits
linear relation
list
location
log-Likelihood
logical
logical expressions
long-term variation
longest run
Loss function analysis
lot
lower capability index
lower control limit
lower performance index
lower specification limit

M
manufacturing specification limit
markdown
matrix
maximum
mean
mean tests
measure
measurement
measurement methods
mechanical devices
median
median absolute deviation
Mersenne-Twister algorithm
minimum
minimum capability index
missing
missing values
mode
model
moving range
moving Range chart
MR Chart

N
names
national body
natural limits
natural variation
nomilan-is-best
nominal values
nomogram
nomograph
non normality
non-random variation
nonlinear profile
nonlinearity
normal distribution
normality
np chart
null hypothesis
numeric

O
O-member
object
one-sided test
open source
operating characteristic
operating characteristic curve
operator
optimization
order
order data frame
out of control
out-of-control process
outlier

P
p chart
P-member
p-value
package
package installation
package update
panel (lattice)
parameter
Pareto Analysis
Pareto chart
partial matching
path
pattern
percentile
Phase I control charts
Phase II control charts
plot
lines
text
plots
point estimation
point estimator
Poisson distribution
Poisson rate
poor quality
population
population mixture
power
predictive variable
print
probability
probability distribution
probability distribution model
problem-solving techniques
procedure
process
control
out of control
shift
process capability
process change
process control
process data
process map
process owner
process parameter
process performance indices
process shift
process tolerance
producer
producer’s risk
product
quality
profile
profiles control chart
programme of work
programming language
Ada
C
Fortran
Pascal
Prolog
 Python
R
Ruby
proportion of defects
proportions test
prototype profile
pseudo-random numbers

Q
quality
characteristic
quality (definition)
quality assurance
quality characteristic
larger-the-better
nominal-is-best
smaller-the-better
quality control
history
plan
quality definition
quality function deployment
quality management systems
quantile
quantile function
quantile-quantile plot
quartile

R
R
CRAN
FAQ
foundation
base
code
community
conferences
Console
 console
contributors
definition
documentation
environment
expression
Foundation
foundation
Help
history
installation
interface
journal
learning curve
libraries
mailing lists
manuals
markdown
object
output
packages
programming language
R-core
release
reports
script
scripts
session
source code
task views
website
R Commander
R Chart
R expression
random
random noise
random numbers
random sampling
random variable
random variate
randomness
range
Range chart
rare event
rational subgroups
raw data
recycling
reference interval
reference limits
regular expression
regularization
rejection
rejection region
relative frequency
reliability
replace
replacement
representative sample
reproducible report
reproducible research
requirement
response
response variable
reverse
risk management
robust parameter design
rolled throughput yield
row
RStudio
help
installation
layout
plots
plots pane
source
source editor
run chart
run tests
S
S Chart
sample
sample data
sample distribution
sample mean
sample size
sample standard deviation
sample variance
sampling
sampling distribution
sampling frequency
sampling plans
scatter plot
script
seasonality
secretariat (of a TC)
seed
sequence
Seven Quality Tools
Shapiro-Wilks test
Shewhart
constants
Shewhart charts
Shewhart control charts
shift
short-term variation
significance level
simple random sampling
simulation
single sampling plan
Six Sigma
skewness
smaller-the-better
smooth function
software
FOSS
Apache
 C
commercial software
eclipse
emacs
Fortran
freedoms
Internet
Java
JMP
LibreOffice
licence
Linux
mac
Microsoft Office
Minitab
MySQL
open source
pandoc
php
Python
R
RStudio
S
SAS
source code
SPSS
Stata
statistical software
support
Systat
Ubuntu
Windows
sort
source files
special causes
special values
specification
specification limit
specification limits
specification tolerance
specifications
specifications design
specifications limits
spreadsheet
standard
standard development
standard deviation
Standard deviation chart
standard errors
standard normal distribution
standard stages
standards
standards development
standards maintenance
statistic
statistical software
strata
stratification
stratified sampling
Student’s t
Sturges
subgroup
subprocess
subset data frame
summary
mean
summary statistic
support vector machines
system locale
systematic sampling

T
Taguchi capability index
Taguchi loss function
tally sheet
target
task view
TC/69 Secretariat
TC69
TC69/AHG1
TC69/SC1
TC69/SC1/WG2
TC69/SC1/WG5
TC69/SC1/WG6
TC69/SC4
TC69/SC4/WG10
TC69/SC4/WG11
TC69/SC4/WG12
TC69/SC5
TC69/SC5/WG10
TC69/SC5/WG2
TC69/SC5/WG3
TC69/SC5/WG8
TC69/SC6
TC69/SC6/WG1
TC69/SC6/WG5
TC69/SC6/WG7
TC69/SC6/WG9
TC69/SC7
TC69/SC7/AHG1
TC69/SC7/WG1
TC69/SC7/WG2
TC69/SC7/WG3
TC69/SC8
TC69/SC8/WG1
TC69/SC8/WG2
TC69/SC8/WG3
TC69/WG3
tendency
terminology
tier chart
time series
time-dependent process
tolerance limits
trend
trivial causes
two-sided test

U
u chart
unbiased estimator
unbiasedness
upper capability index
upper control limit
upper performance index
upper specification limit

V
variability
variable
categorical
continuous
discrete
qualitative
quatitative
variables control charts
variables relation
variance
variance tests
variation
assignable cause
assignable causes
assignable variation
causes
chance variation
common causes
special causes
vector
vectorized functions
vignette
vital causes
voice of stakeholders
voice of the customer
voice of the process
VoS

W
WD
webscrapping
weighted mean
weighted moving average
Wikipedia
working directory
workspace

X
x-bar chart

Y
yield

Footnotes
1 See ‘Edit’ menu for further options.

2 Double operators && and || are used to compare vectors globally. Single operators, element-wise.

3 This makes results reproducible.